Applied Computing and Geosciences最新文献_第9页

Irrigated rice-field mapping in Brazil using phenological stage information and optical and microwave remote sensing 利用物候阶段信息和光学及微波遥感技术在巴西进行灌溉稻田制图

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100223

Andre Dalla Bernardina Garcia , MD Samiul Islam , Victor Hugo Rohden Prudente , Ieda Del’Arco Sanches , Irene Cheng

Irrigated rice-field mapping methodologies have been rapidly evolving as a result of advanced remote sensing (RS) technology. However, current methods rely on extensive time-series data and a wide range of multi-spectral bands. These methods often struggle with classification accuracy with contaminated satellite data due to environmental factors or acquisition device constraints, e.g., cloud cover, shadows, noise, and the temporal and spectral resolution trade-off. Our goal is map irrigated rice-field by using a suitable satellite image band composition instead of time-series data. We divide the growth cycle into different rice phenological stages: beginning, middle and end of season, as well as the season transition periods. Near-infrared (NIR), short-wave infrared (SWIR) and red bands of MultiSpectral Instrument - MSI/Sentinel-2 (optical RS), along with polarizations of VV (vertical–vertical) and VH (vertical–horizontal) of Sentinel-1 C-band Synthetic Aperture Radar (SAR) (microwave RS), were used to create ten different false-color image composites. Ground truth maps from two consecutive growth seasons (2017/2018 and 2018/2019) served as references. We applied a modified version of the Fusion Adaptive Patch Network (FAPNET), named as Patch Layer Adaptive Network (PLANET) convolutional neural network (CNN) to obtain binary rice mapping, which was evaluated using the traditional Mean Intersection over Union (MIoU) and Dice coefficient. Analytic results show that the end of season is the most suitable for obtaining a reliable classification based on optical and SAR sensors. Although complex rice-field pose challenges, our predictions consistently scored a MIoU above 0.9. We conclude that choosing the right phenological stage for rice mapping combined with deep learning model can greatly improve the classification results. These results indicate that the choice of composition significantly impacts classification accuracy, especially in more complex environments.

由于先进的遥感技术，灌溉稻田测绘方法得到了迅速发展。然而，目前的方法依赖于大量的时间序列数据和大范围的多光谱波段。由于环境因素或采集设备的限制，例如，云层、阴影、噪声以及时间和光谱分辨率的权衡，这些方法通常在受污染卫星数据的分类精度方面存在困难。我们的目标是利用合适的卫星图像波段组成代替时间序列数据来绘制灌溉稻田。我们将水稻的生长周期划分为不同的物候阶段：季初、季中、季末，以及季节过渡期。利用MSI/Sentinel-2多光谱仪（光学RS）的近红外（NIR）、短波红外（SWIR）和红色波段，以及Sentinel-1 c波段合成孔径雷达（SAR）（微波RS）的VV（垂直-垂直）和VH（垂直-水平）极化，合成了10幅不同的伪彩色图像。连续两个增长季节（2017/2018和2018/2019）的地面真值图作为参考。我们采用了一种改进的融合自适应补丁网络（FAPNET），即补丁层自适应网络（PLANET）卷积神经网络（CNN）来获得二元水稻映射，并使用传统的平均交联（MIoU）和Dice系数对其进行评估。分析结果表明，基于光学和SAR传感器的分类最适合在季末进行可靠分类。尽管复杂的稻田构成了挑战，但我们的预测MIoU始终在0.9以上。研究表明，选择合适的物候阶段进行水稻的分类，结合深度学习模型可以大大提高分类结果。这些结果表明，成分的选择显著影响分类精度，特别是在更复杂的环境中。

{"title":"Irrigated rice-field mapping in Brazil using phenological stage information and optical and microwave remote sensing","authors":"Andre Dalla Bernardina Garcia , MD Samiul Islam , Victor Hugo Rohden Prudente , Ieda Del’Arco Sanches , Irene Cheng","doi":"10.1016/j.acags.2025.100223","DOIUrl":"10.1016/j.acags.2025.100223","url":null,"abstract":"<div><div>Irrigated rice-field mapping methodologies have been rapidly evolving as a result of advanced remote sensing (RS) technology. However, current methods rely on extensive time-series data and a wide range of multi-spectral bands. These methods often struggle with classification accuracy with contaminated satellite data due to environmental factors or acquisition device constraints, e.g., cloud cover, shadows, noise, and the temporal and spectral resolution trade-off. Our goal is map irrigated rice-field by using a suitable satellite image band composition instead of time-series data. We divide the growth cycle into different rice phenological stages: beginning, middle and end of season, as well as the season transition periods. Near-infrared (NIR), short-wave infrared (SWIR) and red bands of MultiSpectral Instrument - MSI/Sentinel-2 (optical RS), along with polarizations of VV (vertical–vertical) and VH (vertical–horizontal) of Sentinel-1 C-band Synthetic Aperture Radar (SAR) (microwave RS), were used to create ten different false-color image composites. Ground truth maps from two consecutive growth seasons (2017/2018 and 2018/2019) served as references. We applied a modified version of the Fusion Adaptive Patch Network (FAPNET), named as Patch Layer Adaptive Network (PLANET) convolutional neural network (CNN) to obtain binary rice mapping, which was evaluated using the traditional Mean Intersection over Union (MIoU) and Dice coefficient. Analytic results show that the end of season is the most suitable for obtaining a reliable classification based on optical and SAR sensors. Although complex rice-field pose challenges, our predictions consistently scored a MIoU above 0.9. We conclude that choosing the right phenological stage for rice mapping combined with deep learning model can greatly improve the classification results. These results indicate that the choice of composition significantly impacts classification accuracy, especially in more complex environments.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100223"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143165475","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deformation analysis by an improved similarity transformation 一种改进的相似变换变形分析方法

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100221

Vahid Mahboub

In this contribution, deformation analysis is rigorously performed by a non-linear 3-D similarity transformation. In contrast to traditional methods based on linear least-squares (LLS), here we solve a non-linear problem without any linearization. To achieve this goal, a new weighted total least-squares (WTLS) approach with general dispersion matrix is implemented to deformation analysis problem. Although some researchers have been trying to solve deformation analysis using TLS approaches, these attempts require modification since they used to apply unstructured TLS techniques such as Generalized TLS (GTLS) to similarity transformation which requires structured TLS (STLS) techniques while the WTLS approach preserves the structure of the functional model when based on the perfect description of the variance-covariance matrix. As a secondary scope, here it is analytically proved that LLS is not identical to nonlinear estimations such as the WTLS methods and rigorous nonlinear least-square (RNLS) as opposed to what in some contributions has been claimed. The third attainment of this contribution is proposing another algorithm for rigorous similarity transformation with arbitrary rotational angles. It is based on the RNLS method which can obtain the correct update of misclosure. Moreover, compared to transformation methods that deal with arbitrary rotational angles, we do not need to impose any orthogonality constraints here. Two case studies numerically confirm that the WTLS and RNLS methods provide the most accurate results among the LLS, GTLS, RNLS and WTLS approaches in two landslide areas.

在这个贡献中，变形分析是通过非线性三维相似变换严格执行的。与传统的基于线性最小二乘（LLS）的方法相比，我们在没有任何线性化的情况下解决了一个非线性问题。为实现这一目标，提出了一种基于广义色散矩阵的加权总最小二乘（WTLS）方法。虽然一些研究人员已经尝试使用TLS方法解决变形分析，但这些尝试需要修改，因为他们使用非结构化TLS技术，如广义TLS （GTLS）来进行相似性转换，这需要结构化TLS （STLS）技术，而WTLS方法在基于方差-协方差矩阵的完美描述时保留了功能模型的结构。作为次要范围，本文分析证明了LLS与非线性估计（如WTLS方法和严格非线性最小二乘（RNLS））不相同，这与某些贡献中所声称的相反。这一贡献的第三个成就是提出了另一种具有任意旋转角度的严格相似变换算法。该方法基于RNLS方法，可以获得误闭的正确更新。此外，与处理任意旋转角的变换方法相比，我们在这里不需要施加任何正交性约束。两个算例表明，在两个滑坡区，WTLS和RNLS方法在LLS、GTLS、RNLS和WTLS方法中提供了最准确的结果。

{"title":"Deformation analysis by an improved similarity transformation","authors":"Vahid Mahboub","doi":"10.1016/j.acags.2025.100221","DOIUrl":"10.1016/j.acags.2025.100221","url":null,"abstract":"<div><div>In this contribution, deformation analysis is rigorously performed by a non-linear 3-D similarity transformation. In contrast to traditional methods based on linear least-squares (LLS), here we solve a non-linear problem without any linearization. To achieve this goal, a new weighted total least-squares (WTLS) approach with general dispersion matrix is implemented to deformation analysis problem. Although some researchers have been trying to solve deformation analysis using TLS approaches, these attempts require modification since they used to apply unstructured TLS techniques such as Generalized TLS (GTLS) to similarity transformation which requires structured TLS (STLS) techniques while the WTLS approach preserves the structure of the functional model when based on the perfect description of the variance-covariance matrix. As a secondary scope, here it is analytically proved that LLS is not identical to nonlinear estimations such as the WTLS methods and rigorous nonlinear least-square (RNLS) as opposed to what in some contributions has been claimed. The third attainment of this contribution is proposing another algorithm for rigorous similarity transformation with arbitrary rotational angles. It is based on the RNLS method which can obtain the correct update of misclosure. Moreover, compared to transformation methods that deal with arbitrary rotational angles, we do not need to impose any orthogonality constraints here. Two case studies numerically confirm that the WTLS and RNLS methods provide the most accurate results among the LLS, GTLS, RNLS and WTLS approaches in two landslide areas.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100221"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143165022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Rapid mapping of landslides using satellite SAR imagery: A progressive learning approach 利用卫星SAR图像快速绘制滑坡地图：渐进式学习方法

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100224

Nikhil Prakash , Andrea Manconi , Alessandro Cesare Mondini

Rapid detection of landslides after an exceptional event is critical for planning effective disaster management. Previous works have typically used machine learning-based methods, including the recently popular deep-learning approaches, to identify characteristics surface features from satellite remote sensing data, especially from optical images. However, data acquisition from optical images is not possible in cloudy conditions, leading to unpredictable delays in any mapping task from future events. These methods also rely on large manually labelled inventories for training, which is often not available before the event. In this work, we propose an active training strategy to generate a landslide map after an event using the first available synthetic-aperture radar (SAR) image and improve it once subsequent cloud-free optical images are acquired. The proposed active learning workflow can start with a small (

\sim 100 m^{2}

) and incomplete inventory,- and can grow the extent and completeness in iterative steps with manual updates after each step. This significantly reduces the slow manual mapping typically required for generating a large training inventory. We designed our experiments to map the landslides triggered by the

M_{w}

6.6 Hokkaido Eastern Iburi earthquake of 2018 in Japan using sequentially ALOS-2 (SAR) and PlanetScope (Optical) scenes in the order they are acquired. The choice of active learning prioritizes speed over accuracy. However, we note only a modest reduction in performance (

\sim 10 %

drop in F1 and MCC scores), with our method allowing a preliminary landslide inventory to be completed within a single day. This is of major importance in disaster response, improving performance and reducing the potential subjectivity associated with manual mapping.

在异常事件发生后迅速发现山体滑坡对于规划有效的灾害管理至关重要。以前的工作通常使用基于机器学习的方法，包括最近流行的深度学习方法，从卫星遥感数据（特别是光学图像）中识别特征表面特征。然而，在多云条件下，从光学图像中获取数据是不可能的，这会导致未来事件的任何映射任务出现不可预测的延迟。这些方法还依赖于大量的人工标记的培训清单，而这些清单在事件发生之前通常是不可用的。在这项工作中，我们提出了一种主动训练策略，在事件发生后使用第一张可用的合成孔径雷达（SAR）图像生成滑坡地图，并在获得后续无云光学图像后对其进行改进。提出的主动学习工作流可以从一个小的（~ 100m2）和不完整的库存开始，并且可以在迭代步骤中增加范围和完整性，在每个步骤之后进行手动更新。这大大减少了生成大型训练清单所需的缓慢的手动映射。我们设计了实验，按照获得的顺序使用ALOS-2 （SAR）和PlanetScope（光学）场景绘制2018年日本北海道东伊武里6.6级地震引发的山体滑坡。主动学习的选择优先考虑速度而不是准确性。然而，我们注意到只有适度的性能下降（F1和MCC分数下降约10%），我们的方法允许在一天内完成初步的滑坡清单。这在灾难响应、提高性能和减少与手工绘图相关的潜在主观性方面非常重要。

{"title":"Rapid mapping of landslides using satellite SAR imagery: A progressive learning approach","authors":"Nikhil Prakash , Andrea Manconi , Alessandro Cesare Mondini","doi":"10.1016/j.acags.2025.100224","DOIUrl":"10.1016/j.acags.2025.100224","url":null,"abstract":"<div><div>Rapid detection of landslides after an exceptional event is critical for planning effective disaster management. Previous works have typically used machine learning-based methods, including the recently popular deep-learning approaches, to identify characteristics surface features from satellite remote sensing data, especially from optical images. However, data acquisition from optical images is not possible in cloudy conditions, leading to unpredictable delays in any mapping task from future events. These methods also rely on large manually labelled inventories for training, which is often not available before the event. In this work, we propose an active training strategy to generate a landslide map after an event using the first available synthetic-aperture radar (SAR) image and improve it once subsequent cloud-free optical images are acquired. The proposed active learning workflow can start with a small (<span><math><mrow><mo>∼</mo><mn>100</mn><msup><mrow><mi>m</mi></mrow><mrow><mn>2</mn></mrow></msup></mrow></math></span>) and incomplete inventory,- and can grow the extent and completeness in iterative steps with manual updates after each step. This significantly reduces the slow manual mapping typically required for generating a large training inventory. We designed our experiments to map the landslides triggered by the <span><math><msub><mrow><mi>M</mi></mrow><mrow><mi>w</mi></mrow></msub></math></span> 6.6 Hokkaido Eastern Iburi earthquake of 2018 in Japan using sequentially ALOS-2 (SAR) and PlanetScope (Optical) scenes in the order they are acquired. The choice of active learning prioritizes speed over accuracy. However, we note only a modest reduction in performance (<span><math><mrow><mo>∼</mo><mn>10</mn><mtext>%</mtext></mrow></math></span> drop in F1 and MCC scores), with our method allowing a preliminary landslide inventory to be completed within a single day. This is of major importance in disaster response, improving performance and reducing the potential subjectivity associated with manual mapping.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100224"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143395700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Chemical map classification in XMapTools XMapTools中的化学图分类

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100230

Pierre Lanari , Mahyra Tedeschi

Chemical mapping using electron beam or laser instruments is an important analytical technique that allows the study of the compositional variability of materials in two dimensions. While quantitative compositional mapping of minerals has received considerable attention over the last two decades, pixel misclassification in commonly used software solutions remains a fundamental limitation affecting several applications. Calibration of intensity maps to fully quantitative compositional maps requires accurate classification, for example when a calibration curve is applied to a group of pixels that are assumed to have the same matrix behavior under the electron beam or the laser. This paper compares seven automated supervised machine learning classification algorithms implemented in the open source XMapTools software along with various tools for manual classification, for selecting training data and assessing the quality of a classification result. This new implementation aims to provide the research and industry communities with a free software tool for fast and robust classification of chemical maps. A standardized color scheme with reference colors for minerals and mineral groups is proposed to improve the readability of the classified maps in petrological studies. The performance of each algorithm varies depending on the data set, especially when minerals exhibit strong compositional zoning or when different minerals have similar compositions for a given element. The random forest algorithm based on bootstrap aggregation provides satisfactory results in most situations and is recommended for general use in XMapTools.

使用电子束或激光仪器的化学作图是一种重要的分析技术，它允许在二维空间研究材料的成分变化。虽然矿物的定量成分制图在过去二十年中受到了相当大的关注，但常用软件解决方案中的像素错误分类仍然是影响若干应用的基本限制。将强度图校准为完全定量的成分图需要精确的分类，例如，当将校准曲线应用于假定在电子束或激光下具有相同矩阵行为的一组像素时。本文比较了在开源的XMapTools软件中实现的7种自动监督机器学习分类算法，以及用于选择训练数据和评估分类结果质量的各种人工分类工具。这个新的实现旨在为研究和工业界提供一个免费的软件工具，用于快速和强大的化学图分类。为了提高岩石学研究中分类图的可读性，提出了一种具有矿物和矿物群参考色的标准化配色方案。每种算法的性能因数据集而异，特别是当矿物表现出强烈的成分分带或当不同矿物具有给定元素的相似成分时。基于自举聚合的随机森林算法在大多数情况下提供了令人满意的结果，建议在XMapTools中一般使用。

{"title":"Chemical map classification in XMapTools","authors":"Pierre Lanari , Mahyra Tedeschi","doi":"10.1016/j.acags.2025.100230","DOIUrl":"10.1016/j.acags.2025.100230","url":null,"abstract":"<div><div>Chemical mapping using electron beam or laser instruments is an important analytical technique that allows the study of the compositional variability of materials in two dimensions. While quantitative compositional mapping of minerals has received considerable attention over the last two decades, pixel misclassification in commonly used software solutions remains a fundamental limitation affecting several applications. Calibration of intensity maps to fully quantitative compositional maps requires accurate classification, for example when a calibration curve is applied to a group of pixels that are assumed to have the same matrix behavior under the electron beam or the laser. This paper compares seven automated supervised machine learning classification algorithms implemented in the open source XMapTools software along with various tools for manual classification, for selecting training data and assessing the quality of a classification result. This new implementation aims to provide the research and industry communities with a free software tool for fast and robust classification of chemical maps. A standardized color scheme with reference colors for minerals and mineral groups is proposed to improve the readability of the classified maps in petrological studies. The performance of each algorithm varies depending on the data set, especially when minerals exhibit strong compositional zoning or when different minerals have similar compositions for a given element. The random forest algorithm based on bootstrap aggregation provides satisfactory results in most situations and is recommended for general use in XMapTools.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100230"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143509291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Classification of geological borehole descriptions using a domain adapted large language model 基于域适应大语言模型的地质钻孔描述分类

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100229

Hossein Ghorbanfekr, Pieter Jan Kerstens, Katrijn Dirix

Geological borehole descriptions contain detailed textual information about the composition of the subsurface. However, their unstructured format presents significant challenges for extracting relevant features into a structured format. This paper introduces GEOBERTje: a domain adapted large language model trained on geological borehole descriptions from Flanders (Belgium) in the Dutch language. This model effectively extracts relevant information from the borehole descriptions and represents it into a numeric vector space. Showcasing just one potential application of GEOBERTje, we finetune a classifier model on a limited number of manually labeled observations. This classifier categorizes borehole descriptions into a main, second and third lithology class. We show that our classifier outperforms a rule-based approach (by 30% on average), non-contextual Word2Vec embeddings combined with a random forest classifier (by 38% on average), and a prompt engineering method with large language models (i.e., GPT-4 (by 11% on average) and Gemma 2 (by 28% on average)). This study exemplifies how domain adapted large language models enhance the efficiency and accuracy of extracting information from complex, unstructured geological descriptions. This offers new opportunities for geological analysis and modeling using vast amounts of data.

地质钻孔描述包含有关地下成分的详细文本信息。然而，它们的非结构化格式对将相关特征提取为结构化格式提出了重大挑战。本文介绍了GEOBERTje：一种基于荷兰语法兰德斯（比利时）地质钻孔描述训练的领域适应大语言模型。该模型有效地从井眼描述中提取相关信息，并将其表示为数值向量空间。仅展示GEOBERTje的一个潜在应用，我们在有限数量的手动标记观测值上微调分类器模型。该分类器将井眼描述分为主要、第二和第三岩性类。我们表明，我们的分类器优于基于规则的方法（平均30%），非上下文Word2Vec嵌入结合随机森林分类器（平均38%），以及具有大型语言模型的提示工程方法（即GPT-4（平均11%）和Gemma 2（平均28%））。本研究举例说明了领域适应大语言模型如何提高从复杂、非结构化的地质描述中提取信息的效率和准确性。这为使用大量数据进行地质分析和建模提供了新的机会。

{"title":"Classification of geological borehole descriptions using a domain adapted large language model","authors":"Hossein Ghorbanfekr, Pieter Jan Kerstens, Katrijn Dirix","doi":"10.1016/j.acags.2025.100229","DOIUrl":"10.1016/j.acags.2025.100229","url":null,"abstract":"<div><div>Geological borehole descriptions contain detailed textual information about the composition of the subsurface. However, their unstructured format presents significant challenges for extracting relevant features into a structured format. This paper introduces GEOBERTje: a domain adapted large language model trained on geological borehole descriptions from Flanders (Belgium) in the Dutch language. This model effectively extracts relevant information from the borehole descriptions and represents it into a numeric vector space. Showcasing just one potential application of GEOBERTje, we finetune a classifier model on a limited number of manually labeled observations. This classifier categorizes borehole descriptions into a main, second and third lithology class. We show that our classifier outperforms a rule-based approach (by 30% on average), non-contextual Word2Vec embeddings combined with a random forest classifier (by 38% on average), and a prompt engineering method with large language models (i.e., GPT-4 (by 11% on average) and Gemma 2 (by 28% on average)). This study exemplifies how domain adapted large language models enhance the efficiency and accuracy of extracting information from complex, unstructured geological descriptions. This offers new opportunities for geological analysis and modeling using vast amounts of data.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100229"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143520867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

lasertram: A Python library for time resolved analysis of laser ablation inductively coupled plasma mass spectrometry data lasertram：用于激光烧蚀电感耦合等离子体质谱数据时间分辨分析的Python库

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100225

Jordan Lubbers , Adam J.R. Kent , Chris Russo

Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) data has a wide variety of uses in the geosciences for in-situ chemical analysis of complex natural materials. Improvements to instrument capabilities and operating software have drastically reduced the time required to generate large volumes of data relative to previous methodologies. Raw data from LA-ICP-MS, however, is in counts per unit time (typically counts per second), not elemental concentrations and converting these count ratesto concentrations requires additional processing. For complex materials where the ablated volume may contain a range of material compositions, a moderate amount of user input is also required if appropriate concentrations are to be accurately calculated. In geologic materials such as glasses and minerals that potentially have numerous heterogeneities (e.g., microlites or other inclusions) within them, this is typically determiningwhether the total ablation signal should be filtered to remove these heterogeneities. This necessitates that the LA-ICP-MS data processing pipeline is one that is not automated, but is also designed to enable rapid and efficient processing of large volumes of data.

Here we introduce

, a Python library for the time resolved analysis of LA-ICP-MS data. We outline its mathematical theory, code structure, and provide an example of how it can be used to provide the time resolved analysis necessitated by LA-ICP-MS data of complex geologic materials. Throughout the

pipeline we show how metadata and data are incrementally added to the objects created such that virtually any aspect of an experiment may be interrogated and its quality assessed. We also show, that when combined with other Python libraries for building graphical user interfaces, it can be utilized outside of a pure scripting environment.

can be found at https://doi.org/10.5066/P1DZUR3Z.

激光烧蚀电感耦合等离子体质谱（LA-ICP-MS）数据在复杂天然材料的原位化学分析地球科学中具有广泛的用途。与以前的方法相比，仪器功能和操作软件的改进大大减少了生成大量数据所需的时间。然而，来自LA-ICP-MS的原始数据是单位时间（通常是每秒）的计数，而不是元素浓度，并且将这些计数率转换为浓度需要额外的处理。对于复杂材料，其中烧蚀体积可能包含一系列材料成分，如果要准确计算适当的浓度，也需要适量的用户输入。在诸如玻璃和矿物等地质材料中，可能存在大量的非均质（例如，微晶岩或其他包裹体），这通常是决定是否应该过滤总烧蚀信号以去除这些非均质。这就要求LA-ICP-MS数据处理管道不是自动化的，但也被设计为能够快速有效地处理大量数据。本文介绍了一个用于LA-ICP-MS数据时间分辨分析的Python库。我们概述了它的数学理论、代码结构，并提供了一个例子，说明如何使用它来提供复杂地质材料的LA-ICP-MS数据所需的时间分辨分析。在整个流程中，我们展示了如何将元数据和数据增量地添加到创建的对象中，以便几乎可以询问实验的任何方面并评估其质量。我们还表明，当与其他Python库结合用于构建图形用户界面时，它可以在纯脚本环境之外使用。可在https://doi.org/10.5066/P1DZUR3Z找到。

{"title":"lasertram: A Python library for time resolved analysis of laser ablation inductively coupled plasma mass spectrometry data","authors":"Jordan Lubbers , Adam J.R. Kent , Chris Russo","doi":"10.1016/j.acags.2025.100225","DOIUrl":"10.1016/j.acags.2025.100225","url":null,"abstract":"<div><div>Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) data has a wide variety of uses in the geosciences for in-situ chemical analysis of complex natural materials. Improvements to instrument capabilities and operating software have drastically reduced the time required to generate large volumes of data relative to previous methodologies. Raw data from LA-ICP-MS, however, is in counts per unit time (typically counts per second), not elemental concentrations and converting these count ratesto concentrations requires additional processing. For complex materials where the ablated volume may contain a range of material compositions, a moderate amount of user input is also required if appropriate concentrations are to be accurately calculated. In geologic materials such as glasses and minerals that potentially have numerous heterogeneities (e.g., microlites or other inclusions) within them, this is typically determiningwhether the total ablation signal should be filtered to remove these heterogeneities. This necessitates that the LA-ICP-MS data processing pipeline is one that is not automated, but is also designed to enable rapid and efficient processing of large volumes of data.</div><div>Here we introduce <figure><img></figure> , a Python library for the time resolved analysis of LA-ICP-MS data. We outline its mathematical theory, code structure, and provide an example of how it can be used to provide the time resolved analysis necessitated by LA-ICP-MS data of complex geologic materials. Throughout the <figure><img></figure> pipeline we show how metadata and data are incrementally added to the objects created such that virtually any aspect of an experiment may be interrogated and its quality assessed. We also show, that when combined with other Python libraries for building graphical user interfaces, it can be utilized outside of a pure scripting environment. <figure><img></figure> can be found at <span><span>https://doi.org/10.5066/P1DZUR3Z</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100225"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Do more with less: Exploring semi-supervised learning for geological image classification 少花钱多办事：探索地质图像分类的半监督学习

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2024.100216

Hisham I. Mamode, Gary J. Hampson, Cédric M. John

Labelled datasets within geoscience can often be small, with data acquisition both costly and challenging, and their interpretation and downstream use in machine learning difficult due to data scarcity. Deep learning algorithms require large datasets to learn a robust relationship between the data and its label and avoid overfitting. To overcome the paucity of data, transfer learning has been employed in classification tasks. But an alternative exists: there often is a large corpus of unlabeled data which may enhance the learning process. To evaluate this potential for subsurface data, we compare a high-performance semi-supervised learning (SSL) algorithm (SimCLRv2) with supervised transfer learning on a Convolutional Neural Network (CNN) in geological image classification.

We tested the two approaches on a classification task of sediment disturbance from cores of International Ocean Drilling Program (IODP) Expeditions 383 and 385. Our results show that semi-supervised transfer learning can be an effective strategy to adopt, with SimCLRv2 capable of producing representations comparable to those of supervised transfer learning. However attempts to enhance the performance of semi-supervised transfer learning with task-specific unlabeled images during self-supervision degraded representations. Significantly, we demonstrate that SimCLRv2 trained on a dataset of core disturbance images can out-perform supervised transfer learning of a CNN once a critical number of task-specific unlabeled images are available for self-supervision. The gain in performance compared to supervised transfer learning is 1% and 3% for binary and multi-class classification, respectively.

Supervised transfer learning can be deployed with comparative ease, whereas the current SSL algorithms such as SimCLRv2 require more effort. We recommend that SSL be explored in cases when large amounts of unlabeled task-specific images exist and improvement of a few percent in metrics matter. When examining small, highly specialized datasets, without large amounts of unlabeled images, supervised transfer learning might be the best strategy to adopt. Overall, SSL is a promising approach and future work should explore this approach utilizing different dataset types, quantity, and quality.

地球科学中的标记数据集通常很小，数据采集既昂贵又具有挑战性，而且由于数据稀缺，它们的解释和在机器学习中的下游使用也很困难。深度学习算法需要大型数据集来学习数据与其标签之间的稳健关系，并避免过拟合。为了克服数据的缺乏，迁移学习被用于分类任务。但另一种选择是存在的：通常存在大量未标记数据的语料库，这可能会增强学习过程。为了评估地下数据的潜力，我们比较了高性能半监督学习（SSL）算法（SimCLRv2）与卷积神经网络（CNN）上的监督迁移学习在地质图像分类中的应用。我们在国际海洋钻探计划（IODP）远征383和385岩心沉积物扰动的分类任务中测试了这两种方法。我们的研究结果表明，半监督迁移学习可以是一种有效的策略，SimCLRv2能够产生与监督迁移学习相当的表示。然而，试图在自我监督过程中使用特定任务的未标记图像来提高半监督迁移学习的性能会降低表征。值得注意的是，我们证明了在核心干扰图像数据集上训练的SimCLRv2可以胜过CNN的监督迁移学习，一旦有临界数量的特定任务的未标记图像可用于自我监督。与监督迁移学习相比，在二元分类和多类分类中，性能的提高分别为1%和3%。有监督的迁移学习可以相对容易地部署，而当前的SSL算法（如SimCLRv2）则需要更多的努力。我们建议在存在大量未标记的特定于任务的图像并且度量提高几个百分点很重要的情况下探索SSL。当检查小型的、高度专业化的数据集，没有大量未标记的图像时，监督迁移学习可能是最好的策略。总的来说，SSL是一种很有前途的方法，未来的工作应该利用不同的数据集类型、数量和质量来探索这种方法。

{"title":"Do more with less: Exploring semi-supervised learning for geological image classification","authors":"Hisham I. Mamode, Gary J. Hampson, Cédric M. John","doi":"10.1016/j.acags.2024.100216","DOIUrl":"10.1016/j.acags.2024.100216","url":null,"abstract":"<div><div>Labelled datasets within geoscience can often be small, with data acquisition both costly and challenging, and their interpretation and downstream use in machine learning difficult due to data scarcity. Deep learning algorithms require large datasets to learn a robust relationship between the data and its label and avoid overfitting. To overcome the paucity of data, transfer learning has been employed in classification tasks. But an alternative exists: there often is a large corpus of unlabeled data which may enhance the learning process. To evaluate this potential for subsurface data, we compare a high-performance semi-supervised learning (SSL) algorithm (SimCLRv2) with supervised transfer learning on a Convolutional Neural Network (CNN) in geological image classification.</div><div>We tested the two approaches on a classification task of sediment disturbance from cores of International Ocean Drilling Program (IODP) Expeditions 383 and 385. Our results show that semi-supervised transfer learning can be an effective strategy to adopt, with SimCLRv2 capable of producing representations comparable to those of supervised transfer learning. However attempts to enhance the performance of semi-supervised transfer learning with task-specific unlabeled images during self-supervision degraded representations. Significantly, we demonstrate that SimCLRv2 trained on a dataset of core disturbance images can out-perform supervised transfer learning of a CNN once a critical number of task-specific unlabeled images are available for self-supervision. The gain in performance compared to supervised transfer learning is 1% and 3% for binary and multi-class classification, respectively.</div><div>Supervised transfer learning can be deployed with comparative ease, whereas the current SSL algorithms such as SimCLRv2 require more effort. We recommend that SSL be explored in cases when large amounts of unlabeled task-specific images exist and improvement of a few percent in metrics matter. When examining small, highly specialized datasets, without large amounts of unlabeled images, supervised transfer learning might be the best strategy to adopt. Overall, SSL is a promising approach and future work should explore this approach utilizing different dataset types, quantity, and quality.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100216"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143166137","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A new inversion algorithm (PyMDS) based on the Pyro library to use chlorine 36 data as a paleoseismological tool on normal fault scarps 基于Pyro库的氯36古地震反演算法（PyMDS）

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100234

Maureen Llinares, Ghislain Gassier, Sophie Viseur, Lucilla Benedetti

Paleoseismology (study of earthquakes that occurred before records were kept and before instruments can record them) provides useful information such as recurrence periods and slip rate to assess seismic hazard and better understand fault mechanisms. Chlorine 36 is one of the paleoseismological tools that can be used to date scarp exhumation associated with earthquakes events.

We propose an algorithm, PyMDS, that uses chlorine 36 data sampled on a fault scarp to retrieve seismic sequences (age and slip associated to each earthquake) and long term slip rate on a normal fault.

We show that the algorithm, based on Hamiltonian kernels, can successfully retrieve earthquakes and long term slip rate on a synthetic dataset. The precision on the ages can vary between few thousand years for old earthquakes (>5000 yr BP) and down to few hundreds of years for the most recent ones (<2000 yr BP). The resolution on the slip is ∼30–50 cm and on the slip rate is ∼ 1 mm/yr. Diagnostic tools (R_hat and divergences on chains) are used to check the convergence of the results.

Our new code is applied to a site in Central Italy, the results yielded are in agreement with the ones obtained previously with another inversion procedure. We found 4 events 7800±400 yr, 4700±400 yr, 3000±200 and 400 ±20 yr BP on the MA3 site. The associated slips were of 130±10 cm, 140±20 cm, 580 ± 20 cm and 205±20 cm. The results are comparable with a previous study made by (Schlagenhauf et al., 2010). The yielded slip rate of 2.7 mm/yr ± 0.4 mm/yr is also coherent with the one determined by Tesson et al. (2020).

古地震学（研究在有记录和仪器记录之前发生的地震）提供了有用的信息，如复发周期和滑动率，以评估地震危害并更好地了解断层机制。氯36是一种古地震学工具，可用于确定与地震事件有关的陡崖发掘的年代。我们提出了一种算法PyMDS，该算法使用断层崖上采样的氯36数据来检索正常断层的地震序列（与每次地震相关的年龄和滑动）和长期滑动率。我们表明，该算法基于哈密顿核，可以成功地检索合成数据集上的地震和长期滑动率。年龄的精确度可以从几千年的老地震（距今5000年）到几百年的最近地震（距今2000年）不等。滑移的分辨率为~ 30-50 cm，滑移速率为~ 1 mm/yr。诊断工具（Rhat和链上的散度）用于检查结果的收敛性。我们的新代码应用于意大利中部的一个站点，所得结果与以前使用另一种反演程序获得的结果一致。我们在MA3位点发现了4个事件（7800±400 yr, 4700±400 yr， 3000±200和400±20 yr）。相关滑移分别为130±10 cm、140±20 cm、580±20 cm和205±20 cm。该结果与（Schlagenhauf et al., 2010）先前的研究结果相当。产生的滑移率为2.7 mm/yr±0.4 mm/yr，也与Tesson等人（2020）确定的滑移率一致。

{"title":"A new inversion algorithm (PyMDS) based on the Pyro library to use chlorine 36 data as a paleoseismological tool on normal fault scarps","authors":"Maureen Llinares, Ghislain Gassier, Sophie Viseur, Lucilla Benedetti","doi":"10.1016/j.acags.2025.100234","DOIUrl":"10.1016/j.acags.2025.100234","url":null,"abstract":"<div><div>Paleoseismology (study of earthquakes that occurred before records were kept and before instruments can record them) provides useful information such as recurrence periods and slip rate to assess seismic hazard and better understand fault mechanisms. Chlorine 36 is one of the paleoseismological tools that can be used to date scarp exhumation associated with earthquakes events.</div><div>We propose an algorithm, PyMDS, that uses chlorine 36 data sampled on a fault scarp to retrieve seismic sequences (age and slip associated to each earthquake) and long term slip rate on a normal fault.</div><div>We show that the algorithm, based on Hamiltonian kernels, can successfully retrieve earthquakes and long term slip rate on a synthetic dataset. The precision on the ages can vary between few thousand years for old earthquakes (>5000 yr BP) and down to few hundreds of years for the most recent ones (<2000 yr BP). The resolution on the slip is ∼30–50 cm and on the slip rate is ∼ 1 mm/yr. Diagnostic tools (R<sub>hat</sub> and divergences on chains) are used to check the convergence of the results.</div><div>Our new code is applied to a site in Central Italy, the results yielded are in agreement with the ones obtained previously with another inversion procedure. We found 4 events 7800±400 yr, 4700±400 yr, 3000±200 and 400 ±20 yr BP on the MA3 site. The associated slips were of 130±10 cm, 140±20 cm, 580 ± 20 cm and 205±20 cm. The results are comparable with a previous study made by (Schlagenhauf et al., 2010). The yielded slip rate of 2.7 mm/yr ± 0.4 mm/yr is also coherent with the one determined by Tesson et al. (2020).</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100234"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143631906","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SeisAug: A data augmentation python toolkit 一个数据增强python工具包

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2025.100232

D. Pragnath , G. Srijayanthi , Santosh Kumar , Sumer Chopra

A common limitation in applying any deep learning and machine learning techniques is the limited labelled dataset which can be addressed through Data augmentation (DA). SeisAug is a DA python toolkit to address this challenge in seismological studies. DA. DA helps to balance the imbalanced classes of a dataset by creating more examples of under-represented classes. It significantly mitigates overfitting by increasing the volume of training data and introducing variability, thereby improving the model's performance on unseen data. Given the rapid advancements in deep learning for seismology, ‘SeisAug’ assists in extensibility by generating a substantial amount of data (2–6 times more data) which can aid in developing an indigenous robust model. Further, this study demonstrates the role of DA in developing a robust model. For this we utilized a basic two class identification models between earthquake/signal and noise/(non-earthquake). The model is trained with original, 1 and 5 times augmented datasets and their performance metrics are evaluated. The model trained with 5X times augmented dataset significantly outperforms with accuracy of 0.991, AUC 0.999 and AUC-PR 0.999 compared to the model trained with original dataset with accuracy of 0.50, AUC 0.75 and AUC-PR 0.80. Furthermore, by making all codes available on GitHub, the toolkit facilitates the easy application of DA techniques, empowering end-users to enhance their seismological waveform datasets effectively and overcome the initial drawbacks posed by the scarcity of labelled data.

应用任何深度学习和机器学习技术的一个常见限制是有限的标记数据集，可以通过数据增强（DA）来解决。SeisAug是一个数据处理python工具包，用于解决地震学研究中的这一挑战。哒。数据分析通过创建更多代表性不足的类的示例来帮助平衡数据集的不平衡类。它通过增加训练数据量和引入可变性来显著减轻过拟合，从而提高模型在未见数据上的性能。鉴于地震学深度学习的快速发展，“SeisAug”通过生成大量数据（2-6倍的数据）来帮助扩展，这可以帮助开发本地鲁棒模型。此外，本研究证明了数据分析在建立稳健模型中的作用。为此，我们利用了地震/信号和噪声/（非地震）之间的基本两类识别模型。使用原始、1倍和5倍增强数据集训练模型，并评估其性能指标。5倍增强数据集训练的模型准确率为0.991，AUC为0.999，AUC- pr为0.999，明显优于原始数据集训练的模型，准确率为0.50，AUC为0.75，AUC- pr为0.80。此外，通过在GitHub上提供所有代码，该工具包促进了数据分析技术的简单应用，使最终用户能够有效地增强他们的地震波形数据集，并克服了标记数据稀缺所带来的最初缺点。

{"title":"SeisAug: A data augmentation python toolkit","authors":"D. Pragnath , G. Srijayanthi , Santosh Kumar , Sumer Chopra","doi":"10.1016/j.acags.2025.100232","DOIUrl":"10.1016/j.acags.2025.100232","url":null,"abstract":"<div><div>A common limitation in applying any deep learning and machine learning techniques is the limited labelled dataset which can be addressed through Data augmentation (DA). SeisAug is a DA python toolkit to address this challenge in seismological studies. DA. DA helps to balance the imbalanced classes of a dataset by creating more examples of under-represented classes. It significantly mitigates overfitting by increasing the volume of training data and introducing variability, thereby improving the model's performance on unseen data. Given the rapid advancements in deep learning for seismology, ‘SeisAug’ assists in extensibility by generating a substantial amount of data (2–6 times more data) which can aid in developing an indigenous robust model. Further, this study demonstrates the role of DA in developing a robust model. For this we utilized a basic two class identification models between earthquake/signal and noise/(non-earthquake). The model is trained with original, 1 and 5 times augmented datasets and their performance metrics are evaluated. The model trained with 5X times augmented dataset significantly outperforms with accuracy of 0.991, AUC 0.999 and AUC-PR 0.999 compared to the model trained with original dataset with accuracy of 0.50, AUC 0.75 and AUC-PR 0.80. Furthermore, by making all codes available on GitHub, the toolkit facilitates the easy application of DA techniques, empowering end-users to enhance their seismological waveform datasets effectively and overcome the initial drawbacks posed by the scarcity of labelled data.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100232"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143591747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Semantic segmentation framework for atoll satellite imagery: An in-depth exploration using UNet variants and Segmentation Gym 环礁卫星图像的语义分割框架：使用UNet变体和分割体育馆的深入探索

IF 2.6 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Applied Computing and Geosciences

Pub Date : 2025-02-01 DOI: 10.1016/j.acags.2024.100217

Ray Wang , Tahiya Chowdhury , Alejandra C. Ortiz

This paper presents a framework for semantic segmentation of satellite imagery aimed at studying atoll morphometrics. Recent advances in deep neural networks for automated segmentation have been valuable across a variety of satellite and aerial imagery applications, such as land cover classification, mineral characterization, and disaster impact assessment. However, identifying an appropriate segmentation approach for geoscience research remains challenging, often relying on trial-and-error experimentation for data preparation, model selection, and validation. Building on prior efforts to create reproducible research pipelines for aerial image segmentation, we propose a systematic framework for custom segmentation model development using Segmentation Gym, a software tool designed for efficient model experimentation. Additionally, we evaluate state-of-the-art U-Net model variants to identify the most accurate and precise model for specific segmentation tasks. Using a dataset of 288 Landsat images of atolls as a case study, we conduct a detailed analysis of various annotation techniques, image types, and training methods, offering a structured framework for practitioners to design and explore segmentation models. Furthermore, we address dataset imbalance, a common challenge in geographical data, and discuss strategies to mitigate its impact on segmentation outcomes. Based on our findings, we provide recommendations for applying this framework to other geoscience research areas to address similar challenges.

针对环礁形态计量学的研究，提出了一种卫星图像语义分割框架。深度神经网络在自动分割方面的最新进展在各种卫星和航空图像应用中都很有价值，例如土地覆盖分类、矿物表征和灾害影响评估。然而，为地球科学研究确定合适的分割方法仍然具有挑战性，通常依赖于反复试验来进行数据准备、模型选择和验证。在先前为航空图像分割创建可重复研究管道的努力的基础上，我们提出了一个系统的框架，用于使用segmentation Gym开发自定义分割模型，这是一个专为高效模型实验而设计的软件工具。此外，我们评估了最先进的U-Net模型变体，以确定最准确和精确的模型，用于特定的分割任务。以288张环礁陆地卫星图像数据集为例，详细分析了各种标注技术、图像类型和训练方法，为从业者设计和探索分割模型提供了结构化框架。此外，我们解决了数据集不平衡，这是地理数据中的一个常见挑战，并讨论了减轻其对分割结果影响的策略。基于我们的发现，我们提供了将该框架应用于其他地球科学研究领域以解决类似挑战的建议。

{"title":"Semantic segmentation framework for atoll satellite imagery: An in-depth exploration using UNet variants and Segmentation Gym","authors":"Ray Wang , Tahiya Chowdhury , Alejandra C. Ortiz","doi":"10.1016/j.acags.2024.100217","DOIUrl":"10.1016/j.acags.2024.100217","url":null,"abstract":"<div><div>This paper presents a framework for semantic segmentation of satellite imagery aimed at studying atoll morphometrics. Recent advances in deep neural networks for automated segmentation have been valuable across a variety of satellite and aerial imagery applications, such as land cover classification, mineral characterization, and disaster impact assessment. However, identifying an appropriate segmentation approach for geoscience research remains challenging, often relying on trial-and-error experimentation for data preparation, model selection, and validation. Building on prior efforts to create reproducible research pipelines for aerial image segmentation, we propose a systematic framework for custom segmentation model development using Segmentation Gym, a software tool designed for efficient model experimentation. Additionally, we evaluate state-of-the-art U-Net model variants to identify the most accurate and precise model for specific segmentation tasks. Using a dataset of 288 Landsat images of atolls as a case study, we conduct a detailed analysis of various annotation techniques, image types, and training methods, offering a structured framework for practitioners to design and explore segmentation models. Furthermore, we address dataset imbalance, a common challenge in geographical data, and discuss strategies to mitigate its impact on segmentation outcomes. Based on our findings, we provide recommendations for applying this framework to other geoscience research areas to address similar challenges.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"25 ","pages":"Article 100217"},"PeriodicalIF":2.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143166135","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0