Pub Date : 2025-02-06DOI: 10.1109/JSTARS.2025.3538911
Lang Wu;Fan Fan;Jun Huang;Guoqiang Xiao
With the increasing deployment of multiple-warhead missiles and UAV swarms, the challenge of achieving high-performance detection of clustered multismall targets has emerged as an imperative issue within infrared (IR) air defense systems. However, the existing methods struggle to accurately characterize the features of clustered targets, resulting in poor performance for clustered target detection. To address this challenge, we propose an IR small-target detection method based on improved density peak clustering (DPC) and gray-level contribution. Specifically, we first introduce attribute filtering to quickly extract candidate targets. Note that the attribute cannot only guide the parameter setting of the improved DPC (IDPC) but also derive the weights of feature fusion. Then, we construct an unsupervised clustering model based on IDPC, which is tailored for detecting clustered targets and can accurately represent the local features of these targets. In addition, a gray-level contribution model is proposed to extract the global features of small targets, leveraging the statistical properties of the gray level of small targets. By the weighted fusion of local and global features, the clustered targets are effectively enhanced, while the background clutter is further suppressed. Extensive experimental results demonstrate that our method exhibits a superior clustered target enhancement effect and a higher probability of multitarget detection compared with the state-of-the-art methods.
{"title":"Infrared Small-Target Detection via Improved Density Peak Clustering and Gray-Level Contribution","authors":"Lang Wu;Fan Fan;Jun Huang;Guoqiang Xiao","doi":"10.1109/JSTARS.2025.3538911","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538911","url":null,"abstract":"With the increasing deployment of multiple-warhead missiles and UAV swarms, the challenge of achieving high-performance detection of clustered multismall targets has emerged as an imperative issue within infrared (IR) air defense systems. However, the existing methods struggle to accurately characterize the features of clustered targets, resulting in poor performance for clustered target detection. To address this challenge, we propose an IR small-target detection method based on improved density peak clustering (DPC) and gray-level contribution. Specifically, we first introduce attribute filtering to quickly extract candidate targets. Note that the attribute cannot only guide the parameter setting of the improved DPC (IDPC) but also derive the weights of feature fusion. Then, we construct an unsupervised clustering model based on IDPC, which is tailored for detecting clustered targets and can accurately represent the local features of these targets. In addition, a gray-level contribution model is proposed to extract the global features of small targets, leveraging the statistical properties of the gray level of small targets. By the weighted fusion of local and global features, the clustered targets are effectively enhanced, while the background clutter is further suppressed. Extensive experimental results demonstrate that our method exhibits a superior clustered target enhancement effect and a higher probability of multitarget detection compared with the state-of-the-art methods.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6551-6566"},"PeriodicalIF":4.7,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10876578","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553134","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-06DOI: 10.1109/JSTARS.2025.3539395
Kai Huang;Chenkai Teng;Jialong Zhang;Rui Bao;Yi Liao;Yunrun He;Bo Qiu;Mingrui Xu
Landsat time-series (LTS) archived the multitemporal hyperspectral images, providing freely accessible and long-term optical data for estimating forest aboveground carbon stock (ACS). Due to LTS carrying noise, there were such issues as bias, outliers, and missing values in ACS estimation. Hence, a new filtering method named terrain-perceive spatiotemporal filtering (TP-STF) was developed to improve the estimation accuracy. In TP-STF, landforms were classified based on the terrain data. A computer-recognizable identifier was generated by perceiving each terrain unit. Combining the discriminative criteria with the spatiotemporal information, the TP-STF adaptively selected performant filtering to reconstruct LTS. Then, the random forests regression (RFR) was employed to estimate ACS of Pinus densata in Shangri-La, Yunnan, China. Compared with the other filtering, the TP-STF method's reconstructed LTS had the best modeling accuracy and the highest prediction accuracy, with R2 = 0.903, RMSE = 17.049 t/hm2, P = 81.080%, and rRMSE = 19.691%. The ACS results using TP-STF and RFR were: 6.56 million tons in 1987, 6.44 million tons in 1992, 6.33 million tons in 1997, 6.35 million tons in 2002, 6.72 million tons in 2007, 6.70 million tons in 2012, and 7.04 million tons in 2017. The TP-STF could effectively denoise the LTS images in high-altitude regions, providing a new approach to improve the accuracy of remote sensing-based forest ACS estimation.
{"title":"A New Spatiotemporal Filtering Method to Reconstruct Landsat Time-Series for Improving Estimation Accuracy of Forest Aboveground Carbon Stock","authors":"Kai Huang;Chenkai Teng;Jialong Zhang;Rui Bao;Yi Liao;Yunrun He;Bo Qiu;Mingrui Xu","doi":"10.1109/JSTARS.2025.3539395","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3539395","url":null,"abstract":"Landsat time-series (LTS) archived the multitemporal hyperspectral images, providing freely accessible and long-term optical data for estimating forest aboveground carbon stock (ACS). Due to LTS carrying noise, there were such issues as bias, outliers, and missing values in ACS estimation. Hence, a new filtering method named terrain-perceive spatiotemporal filtering (TP-STF) was developed to improve the estimation accuracy. In TP-STF, landforms were classified based on the terrain data. A computer-recognizable identifier was generated by perceiving each terrain unit. Combining the discriminative criteria with the spatiotemporal information, the TP-STF adaptively selected performant filtering to reconstruct LTS. Then, the random forests regression (RFR) was employed to estimate ACS of <italic>Pinus densata</i> in Shangri-La, Yunnan, China. Compared with the other filtering, the TP-STF method's reconstructed LTS had the best modeling accuracy and the highest prediction accuracy, with <italic>R</i><sup>2</sup> = 0.903, RMSE = 17.049 t/hm<sup>2</sup>, <italic>P</i> = 81.080%, and rRMSE = 19.691%. The ACS results using TP-STF and RFR were: 6.56 million tons in 1987, 6.44 million tons in 1992, 6.33 million tons in 1997, 6.35 million tons in 2002, 6.72 million tons in 2007, 6.70 million tons in 2012, and 7.04 million tons in 2017. The TP-STF could effectively denoise the LTS images in high-altitude regions, providing a new approach to improve the accuracy of remote sensing-based forest ACS estimation.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6503-6519"},"PeriodicalIF":4.7,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10876592","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-05DOI: 10.1109/JSTARS.2025.3539217
Suraj A. Yadav;Xin Zhang;Nuwan K. Wijewardane;Max Feldman;Ruijun Qin;Yanbo Huang;Sathishkumar Samiappan;Wyatt Young;Francisco G. Tapia
The study demonstrated the efficacy of integrating time-series uncrewed aerial system (UAS) multispectral imaging with data-driven deep learning methodologies to systematically and precisely predict field-scale crop yield throughout the growing seasons. A UAS equipped with a micasense rededge MX+ sensor was used for data acquisition at the Hermiston Agricultural Research and Extension Center, Oregon State University. The data were collected throughout the potato (Solanum tuberosum L.) growing seasons under varied nitrogen (N)-rates ranging from 0 to 639 kg/ha. The raw data were preprocessed using Pix4Dmapper and the quantum geographic information system. A linear unmixing model followed by Otsu-based adaptive autosegmentation was implemented to generate soil-masked spatio-spectral fusion maps for accurate vegetation feature extraction. The proposed feature engineering and prediction model followed a two-fold approach: first, adoption of partial least squares regression (PLSR) algorithm to extract features relevant to yield, and second, a novel context-aware attention and residual connection convolution-bidirectional gated recurrent unit bidirectional long short-term memory-network (CAR Conv1D-BiGRU-BiLSTM-Net) to exploit time-series multifeatures information to predict final yield. On integrating the PLSR-derived robust features, the proposed model demonstrated an increase in predictive capability from emergence (T1) to bulking (T4) growth stage by effectively capturing the temporal dynamics of physiological and biological traits. Overall, using multifeatures such as simple ratio, Chlorophyll Green, modified anthocyanin reflectance index, vegetation fraction ($V_{f}$), and N-rate from T1–T4 growth stage resulted in predictive accuracy with high $text{R}^{2}$ = 0.775 and low root mean square error of 16.4%, outperforming other deep learning models.
{"title":"Context-Aware Deep Learning Model for Yield Prediction in Potato Using Time-Series UAS Multispectral Data","authors":"Suraj A. Yadav;Xin Zhang;Nuwan K. Wijewardane;Max Feldman;Ruijun Qin;Yanbo Huang;Sathishkumar Samiappan;Wyatt Young;Francisco G. Tapia","doi":"10.1109/JSTARS.2025.3539217","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3539217","url":null,"abstract":"The study demonstrated the efficacy of integrating time-series uncrewed aerial system (UAS) multispectral imaging with data-driven deep learning methodologies to systematically and precisely predict field-scale crop yield throughout the growing seasons. A UAS equipped with a micasense rededge MX+ sensor was used for data acquisition at the Hermiston Agricultural Research and Extension Center, Oregon State University. The data were collected throughout the potato (<italic>Solanum tuberosum L.</i>) growing seasons under varied nitrogen (N)-rates ranging from 0 to 639 kg/ha. The raw data were preprocessed using Pix4Dmapper and the quantum geographic information system. A linear unmixing model followed by Otsu-based adaptive autosegmentation was implemented to generate soil-masked spatio-spectral fusion maps for accurate vegetation feature extraction. The proposed feature engineering and prediction model followed a two-fold approach: first, adoption of partial least squares regression (PLSR) algorithm to extract features relevant to yield, and second, a novel context-aware attention and residual connection convolution-bidirectional gated recurrent unit bidirectional long short-term memory-network (CAR Conv1D-BiGRU-BiLSTM-Net) to exploit time-series multifeatures information to predict final yield. On integrating the PLSR-derived robust features, the proposed model demonstrated an increase in predictive capability from emergence (T1) to bulking (T4) growth stage by effectively capturing the temporal dynamics of physiological and biological traits. Overall, using multifeatures such as simple ratio, Chlorophyll Green, modified anthocyanin reflectance index, vegetation fraction (<inline-formula><tex-math>$V_{f}$</tex-math></inline-formula>), and N-rate from T1–T4 growth stage resulted in predictive accuracy with high <inline-formula><tex-math>$text{R}^{2}$</tex-math></inline-formula> = 0.775 and low root mean square error of 16.4%, outperforming other deep learning models.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6096-6115"},"PeriodicalIF":4.7,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10873293","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143512905","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-05DOI: 10.1109/JSTARS.2025.3538920
Xi Cheng;Jieyu Yang;Bin Li;Bin Zhao;Deng Pan;Zhanfeng Shen;Qian Zhu;Miaomiao Liu
Urban vacant land (UVL) presents significant environmental and urban planning challenges as cities expand, necessitating effective identification and management strategies. This study proposes an enhanced framework for UVL extraction, based on an improved SegFormer model, which incorporates the densely connected atrous spatial pyramid pooling module and the progressive feature pyramid network for expanded receptive field and achieve multiscale feature integration. The framework first applies a region-based stratification approach, dividing the study area into the central and expanded areas to handle varying land characteristics in different urban regions. Both pretrained and non-pretrained models were utilized to assess their effectiveness in segmentation accuracy, using high-resolution remote sensing images of Chengdu. The experimental results demonstrate the effectiveness of the framework, with the pretrained model, trained on urbanized area data from Chinese cities, achieving F1-scores of 91.34 and 90.05 and IoU values of 84.21 and 81.91 for the central and expanded areas, respectively. In contrast, the non-pretrained model yielded F1-scores of 93.08 and 92.32, with corresponding IoU values of 87.16 and 85.74. Ablation studies and robustness tests further confirm the model's stability and precision in complex application scenarios. This framework provides the accurate and efficient tool for UVL identification, contributing to improved urban land utilization and offering valuable insights for future research and urban planning.
{"title":"Enhancing Urban Land Utilization Through SegFormer: A Vacant Land Analysis in Chengdu","authors":"Xi Cheng;Jieyu Yang;Bin Li;Bin Zhao;Deng Pan;Zhanfeng Shen;Qian Zhu;Miaomiao Liu","doi":"10.1109/JSTARS.2025.3538920","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538920","url":null,"abstract":"Urban vacant land (UVL) presents significant environmental and urban planning challenges as cities expand, necessitating effective identification and management strategies. This study proposes an enhanced framework for UVL extraction, based on an improved SegFormer model, which incorporates the densely connected atrous spatial pyramid pooling module and the progressive feature pyramid network for expanded receptive field and achieve multiscale feature integration. The framework first applies a region-based stratification approach, dividing the study area into the central and expanded areas to handle varying land characteristics in different urban regions. Both pretrained and non-pretrained models were utilized to assess their effectiveness in segmentation accuracy, using high-resolution remote sensing images of Chengdu. The experimental results demonstrate the effectiveness of the framework, with the pretrained model, trained on urbanized area data from Chinese cities, achieving <italic>F</i>1-scores of 91.34 and 90.05 and IoU values of 84.21 and 81.91 for the central and expanded areas, respectively. In contrast, the non-pretrained model yielded <italic>F</i>1-scores of 93.08 and 92.32, with corresponding IoU values of 87.16 and 85.74. Ablation studies and robustness tests further confirm the model's stability and precision in complex application scenarios. This framework provides the accurate and efficient tool for UVL identification, contributing to improved urban land utilization and offering valuable insights for future research and urban planning.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6070-6085"},"PeriodicalIF":4.7,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10873816","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489227","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-05DOI: 10.1109/JSTARS.2025.3538572
Sihan Li;Siqi Qi;Xiaojun Ma
In this work, a set of retrieval methods for evaluating the aerosol mass concentration suitable for high-pollution areas is used to study the vertical distribution of the $text{PM}_{2.5}$ mass concentration in different regions of East Asia from 2007 to 2016, high-resolution satellite active remote sensing data. Research has revealed that high-pollution areas in East Asia are located mainly in the Taklamakan Desert, northern India, and northern China, followed by Southeast Asia and southern China. The aerosol thickness in the dust source areas of eastern China is generally 4 km, with an average $text{PM}_{2.5}$ mass concentration of 60.3 μg · m−1 in the aerosol layer. In densely populated areas, the aerosol thickness is 2 km, with an average $text{PM}_{2.5}$ mass concentration of 40.9 μg · m−1 in the aerosol layer. Results from this work can be used for further analysis of the aerosol effects on weather and climate in East Asia.
{"title":"Vertical Distribution of the Aerosol Mass Concentration in East Asia Using CALIPSO Data","authors":"Sihan Li;Siqi Qi;Xiaojun Ma","doi":"10.1109/JSTARS.2025.3538572","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538572","url":null,"abstract":"In this work, a set of retrieval methods for evaluating the aerosol mass concentration suitable for high-pollution areas is used to study the vertical distribution of the <inline-formula><tex-math>$text{PM}_{2.5}$</tex-math></inline-formula> mass concentration in different regions of East Asia from 2007 to 2016, high-resolution satellite active remote sensing data. Research has revealed that high-pollution areas in East Asia are located mainly in the Taklamakan Desert, northern India, and northern China, followed by Southeast Asia and southern China. The aerosol thickness in the dust source areas of eastern China is generally 4 km, with an average <inline-formula><tex-math>$text{PM}_{2.5}$</tex-math></inline-formula> mass concentration of 60.3 <italic>μ</i>g · m<sup>−1</sup> in the aerosol layer. In densely populated areas, the aerosol thickness is 2 km, with an average <inline-formula><tex-math>$text{PM}_{2.5}$</tex-math></inline-formula> mass concentration of 40.9 <italic>μ</i>g · m<sup>−1</sup> in the aerosol layer. Results from this work can be used for further analysis of the aerosol effects on weather and climate in East Asia.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6041-6046"},"PeriodicalIF":4.7,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10873820","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489072","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-05DOI: 10.1109/JSTARS.2025.3538998
Songyi Xiao;Liangliang Zhu;Xinwen Zhu;Xiurui Geng
When spectral bands are considered as scatter points in the image space, hyperspectral band selection can be transformed into the column subset selection problem for matrices. We consider the problem of finding the submatrix of maximum volume of a matrix $mathbf {X}in mathbb {R}^{Ntimes L}$$(Ngg L)$ within the column space perspective. To solve this problem, we present a novel determinant formula that utilizes the scalar product of arbitrary-order antisymmetric tensors in wedge product form, providing a fresh perspective for interpreting the volume gradient. Based on the variant of this formula, we can mitigate the local extremum problem inherent in existing subset selection algorithms to a certain extent. Subsequently, we introduce exterior algebra theory into the remote sensing community and develop a hyperspectral band selection algorithm via higher order joint volume gradient. To address bias in the volume criterion and reduce algorithmic complexity, we integrate a general subspace partitioning framework that accounts for spectral information. In addition, we propose a method for automatically determining the optimal number of bands. Experiments conducted on both publicly available and Gaofen-5 satellite hyperspectral datasets demonstrate the superiority and stability of our proposed algorithms in terms of classification accuracy and computational efficiency.
{"title":"Optimized Subset Selection for Matrices and its Application in Hyperspectral Band Selection","authors":"Songyi Xiao;Liangliang Zhu;Xinwen Zhu;Xiurui Geng","doi":"10.1109/JSTARS.2025.3538998","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538998","url":null,"abstract":"When spectral bands are considered as scatter points in the image space, hyperspectral band selection can be transformed into the column subset selection problem for matrices. We consider the problem of finding the submatrix of maximum volume of a matrix <inline-formula><tex-math>$mathbf {X}in mathbb {R}^{Ntimes L}$</tex-math></inline-formula> <inline-formula><tex-math>$(Ngg L)$</tex-math></inline-formula> within the column space perspective. To solve this problem, we present a novel determinant formula that utilizes the scalar product of arbitrary-order antisymmetric tensors in wedge product form, providing a fresh perspective for interpreting the volume gradient. Based on the variant of this formula, we can mitigate the local extremum problem inherent in existing subset selection algorithms to a certain extent. Subsequently, we introduce exterior algebra theory into the remote sensing community and develop a hyperspectral band selection algorithm via higher order joint volume gradient. To address bias in the volume criterion and reduce algorithmic complexity, we integrate a general subspace partitioning framework that accounts for spectral information. In addition, we propose a method for automatically determining the optimal number of bands. Experiments conducted on both publicly available and Gaofen-5 satellite hyperspectral datasets demonstrate the superiority and stability of our proposed algorithms in terms of classification accuracy and computational efficiency.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6464-6479"},"PeriodicalIF":4.7,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10873828","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553135","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-04DOI: 10.1109/JSTARS.2025.3538488
Jie Guo;Shuchang Jiao;Hao Sun;Bin Song;Yuhao Chi
Multilabel remote sensing image classification (MLRSIC) can provide comprehensive object-level semantic descriptions of remote sensing images. However, most existing methods struggle to effectively integrate visual features of images and high-level semantic information of labels, which limits their ability to obtain fine-grained image features with rich semantic information for classification. To address these issues, we propose a novel cross-modal compositional learning (CMCL) model for MLRSIC, which fully utilizes label information to improve classification performance. CMCL introduces rich label semantic information into the image feature extraction process, which enhances the visual features of each class with the label semantic. Meanwhile, CMCL constructs a feature space shared by image features and label correlation features for classification. The multilabel image classification task is modeled as the feature distance measurement task, and the visual features of images and the semantic information of labels are mutually promoted by the method. Experimental results on UCM, AID, and DFC15 multilabel datasets show that the proposed CMCL outperforms existing state-of-the-art methods.
{"title":"Cross-Modal Compositional Learning for Multilabel Remote Sensing Image Classification","authors":"Jie Guo;Shuchang Jiao;Hao Sun;Bin Song;Yuhao Chi","doi":"10.1109/JSTARS.2025.3538488","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538488","url":null,"abstract":"Multilabel remote sensing image classification (MLRSIC) can provide comprehensive object-level semantic descriptions of remote sensing images. However, most existing methods struggle to effectively integrate visual features of images and high-level semantic information of labels, which limits their ability to obtain fine-grained image features with rich semantic information for classification. To address these issues, we propose a novel cross-modal compositional learning (CMCL) model for MLRSIC, which fully utilizes label information to improve classification performance. CMCL introduces rich label semantic information into the image feature extraction process, which enhances the visual features of each class with the label semantic. Meanwhile, CMCL constructs a feature space shared by image features and label correlation features for classification. The multilabel image classification task is modeled as the feature distance measurement task, and the visual features of images and the semantic information of labels are mutually promoted by the method. Experimental results on UCM, AID, and DFC15 multilabel datasets show that the proposed CMCL outperforms existing state-of-the-art methods.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"5810-5823"},"PeriodicalIF":4.7,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10872829","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Landscape character identification (LCI) has traditionally relied on manual methods to integrate and visually interpret multiple layers of natural and cultural data across regions. While effective, these methods are constrained by subjectivity in manual categorization and face challenges in scalability and consistency when applied to larger regions. In this article, we propose a novel deep learning-based LCI method that leverages natural language processing (NLP) to enable greater flexibility in identifying landscape types through natural language guidance. Focusing on the Western Sichuan Plains, our approach integrates nine geographic information system-derived landscape elements with natural language descriptions that specify desired styles of landscape characteristics. The model features three key components: a transformer-based module for extracting natural language features, a vision transformer (ViT) for spatial feature extraction, and a feature pyramid network for decision-making. Through multimodal information fusion across the feature extraction and decision-making stages, natural language inputs effectively guide the prioritization of landscape attributes and control the granularity of the generated landscape character maps. Visualization experiments demonstrate the model's capability to produce accurate and detailed landscape character maps, with a particular emphasis on identifying agricultural landscapes in the Western Sichuan Plains. This study validates the potential of integrating NLP into LCI, offering a significant advancement in precision and adaptability for landscape characterization.
{"title":"Integrating Natural Language Processing With Vision Transformer for Landscape Character Identification","authors":"Tingting Huang;Haiyue Zhao;Bo Huang;Sha Li;Jianning Zhu","doi":"10.1109/JSTARS.2025.3538174","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538174","url":null,"abstract":"Landscape character identification (LCI) has traditionally relied on manual methods to integrate and visually interpret multiple layers of natural and cultural data across regions. While effective, these methods are constrained by subjectivity in manual categorization and face challenges in scalability and consistency when applied to larger regions. In this article, we propose a novel deep learning-based LCI method that leverages natural language processing (NLP) to enable greater flexibility in identifying landscape types through natural language guidance. Focusing on the Western Sichuan Plains, our approach integrates nine geographic information system-derived landscape elements with natural language descriptions that specify desired styles of landscape characteristics. The model features three key components: a transformer-based module for extracting natural language features, a vision transformer (ViT) for spatial feature extraction, and a feature pyramid network for decision-making. Through multimodal information fusion across the feature extraction and decision-making stages, natural language inputs effectively guide the prioritization of landscape attributes and control the granularity of the generated landscape character maps. Visualization experiments demonstrate the model's capability to produce accurate and detailed landscape character maps, with a particular emphasis on identifying agricultural landscapes in the Western Sichuan Plains. This study validates the potential of integrating NLP into LCI, offering a significant advancement in precision and adaptability for landscape characterization.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"5838-5852"},"PeriodicalIF":4.7,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10870433","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143480742","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-04DOI: 10.1109/JSTARS.2025.3538701
Na Liu;Guodong Wu;Yonggui Huang;Xi Chen;Qingdu Li;Lihong Wan
In large-scale multimodal remote sensing data archives, the application of cross-modal technology to achieve fast retrieval between different modalities has attracted great attention. In this article, we focus on cross-modal retrieval technology between remote sensing images and text. At present, there is still a large heterogeneity problem in the semantic information extracted from different modal data in the remote sensing field, which leads to the inability to effectively utilize intraclass similarities and interclass differences in the hash learning process, ultimately resulting in low cross-modal retrieval accuracy. In addition, supervised learning-based methods require a large number of labeled training samples, which limits the large-scale application of hash-based cross-modal retrieval technology in the remote sensing field. To address this problem, this article proposes a new unsupervised cross-autoencoder contrast hashing method for RS retrieval. This method constructs an end-to-end deep hashing model, which mainly includes a feature extraction module and a hash representation module. The feature extraction module is mainly responsible for extracting deep semantic information from different modal data and sends the different modal semantic information to the hash representation module through the intermediate layer to learn and generate binary hash codes. In the hashing module, we introduce a new multiobjective loss function to increase the expression of intramodal and intermodal semantic consistency through multiscale semantic similarity constraints and contrastive learning and add a cross-autoencoding module to reconstruct and compare hash features to reduce the loss of semantic information during the learning process. This article conducts a large number of experiments on the UC Merced Land dataset and the RSICD dataset. The experimental results of these two popular benchmark datasets show that the proposed CACH method outperforms the most advanced unsupervised cross-modal hashing methods in RS.
{"title":"Unsupervised Contrastive Hashing With Autoencoder Semantic Similarity for Cross-Modal Retrieval in Remote Sensing","authors":"Na Liu;Guodong Wu;Yonggui Huang;Xi Chen;Qingdu Li;Lihong Wan","doi":"10.1109/JSTARS.2025.3538701","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538701","url":null,"abstract":"In large-scale multimodal remote sensing data archives, the application of cross-modal technology to achieve fast retrieval between different modalities has attracted great attention. In this article, we focus on cross-modal retrieval technology between remote sensing images and text. At present, there is still a large heterogeneity problem in the semantic information extracted from different modal data in the remote sensing field, which leads to the inability to effectively utilize intraclass similarities and interclass differences in the hash learning process, ultimately resulting in low cross-modal retrieval accuracy. In addition, supervised learning-based methods require a large number of labeled training samples, which limits the large-scale application of hash-based cross-modal retrieval technology in the remote sensing field. To address this problem, this article proposes a new unsupervised cross-autoencoder contrast hashing method for RS retrieval. This method constructs an end-to-end deep hashing model, which mainly includes a feature extraction module and a hash representation module. The feature extraction module is mainly responsible for extracting deep semantic information from different modal data and sends the different modal semantic information to the hash representation module through the intermediate layer to learn and generate binary hash codes. In the hashing module, we introduce a new multiobjective loss function to increase the expression of intramodal and intermodal semantic consistency through multiscale semantic similarity constraints and contrastive learning and add a cross-autoencoding module to reconstruct and compare hash features to reduce the loss of semantic information during the learning process. This article conducts a large number of experiments on the UC Merced Land dataset and the RSICD dataset. The experimental results of these two popular benchmark datasets show that the proposed CACH method outperforms the most advanced unsupervised cross-modal hashing methods in RS.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6047-6059"},"PeriodicalIF":4.7,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10870405","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489228","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Airborne platforms and satellites provide rich sensor data in the form of hyperspectral images (HSI), which are crucial for numerous vision-related tasks, such as feature extraction, image enhancement, and data synthesis. This article reviews the contextual importance and applications of generative artificial intelligence (GAI) in the advancement of HSI processing. GAI methods address the inherent challenges of HSI data, such as high dimensionality, noise, and the need to preserve spectral-spatial correlations, rendering them indispensable for modern HSI analysis. Generative neural networks, including generative adversarial networks and denoising diffusion probabilistic models, are highlighted for their superior performance in classification, segmentation, and object identification tasks, often surpassing traditional approaches, such as U-Nets, autoencoders, and deep convolutional neural networks. Diffusion models showed competitive performance in tasks, such as feature extraction and image resolution enhancement, particularly in terms of inference time and computational cost. Transformer architectures combined with attention mechanisms further improved the accuracy of generative methods, particularly for preserving spectral and spatial information in tasks, such as image translation, data augmentation, and data synthesis. Despite these advancements, challenges remain, particularly in developing computationally efficient models for super-resolution and data synthesis. In addition, novel evaluation metrics tailored to the complex nature of HSI data are needed. This review underscores the potential of GAI in addressing these challenges while presenting its current strengths, limitations, and future research directions.
{"title":"Generative Artificial Intelligence for Hyperspectral Sensor Data: A Review","authors":"Diaa Addeen Abuhani;Imran Zualkernan;Raghad Aldamani;Mohamed Alshafai","doi":"10.1109/JSTARS.2025.3538759","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3538759","url":null,"abstract":"Airborne platforms and satellites provide rich sensor data in the form of hyperspectral images (HSI), which are crucial for numerous vision-related tasks, such as feature extraction, image enhancement, and data synthesis. This article reviews the contextual importance and applications of generative artificial intelligence (GAI) in the advancement of HSI processing. GAI methods address the inherent challenges of HSI data, such as high dimensionality, noise, and the need to preserve spectral-spatial correlations, rendering them indispensable for modern HSI analysis. Generative neural networks, including generative adversarial networks and denoising diffusion probabilistic models, are highlighted for their superior performance in classification, segmentation, and object identification tasks, often surpassing traditional approaches, such as U-Nets, autoencoders, and deep convolutional neural networks. Diffusion models showed competitive performance in tasks, such as feature extraction and image resolution enhancement, particularly in terms of inference time and computational cost. Transformer architectures combined with attention mechanisms further improved the accuracy of generative methods, particularly for preserving spectral and spatial information in tasks, such as image translation, data augmentation, and data synthesis. Despite these advancements, challenges remain, particularly in developing computationally efficient models for super-resolution and data synthesis. In addition, novel evaluation metrics tailored to the complex nature of HSI data are needed. This review underscores the potential of GAI in addressing these challenges while presenting its current strengths, limitations, and future research directions.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"6422-6439"},"PeriodicalIF":4.7,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10870282","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：