Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486179
Xinle Liu, Hui-bin Yan, H. Huo, T. Fang
The goal of mining middle-level visual elements is to discover a set of image patches that are representative of and discriminative for a target category. The commonly used mid-level feature representations such as bag-of-visual-words (BOW) models or part-based models in high-resolution remote sensing (HRS) images, seldom consider the discriminability of visual words or parts in object detection. To address this problem, we propose a novel and effective HRS image object detection method based on mid-level visual element representations. First, we employ an iterative procedure that alternates between retraining discriminative classifiers and mining for additional patch instances to discover the discriminative patches, i.e., discriminative mid-level visual elements. Then, a novel mid-level feature representation for an image is constructed based on these visual elements to achieve object detection in HRS images. The experiments on the two HRS image datasets demonstrated the effectiveness of the proposed method compared with several state-of-the-art BOW-based and part-based models.
{"title":"Mining Mid-Level Visual Elements for Object Detection in High-Resolution Remote Sensing Images","authors":"Xinle Liu, Hui-bin Yan, H. Huo, T. Fang","doi":"10.1109/PRRS.2018.8486179","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486179","url":null,"abstract":"The goal of mining middle-level visual elements is to discover a set of image patches that are representative of and discriminative for a target category. The commonly used mid-level feature representations such as bag-of-visual-words (BOW) models or part-based models in high-resolution remote sensing (HRS) images, seldom consider the discriminability of visual words or parts in object detection. To address this problem, we propose a novel and effective HRS image object detection method based on mid-level visual element representations. First, we employ an iterative procedure that alternates between retraining discriminative classifiers and mining for additional patch instances to discover the discriminative patches, i.e., discriminative mid-level visual elements. Then, a novel mid-level feature representation for an image is constructed based on these visual elements to achieve object detection in HRS images. The experiments on the two HRS image datasets demonstrated the effectiveness of the proposed method compared with several state-of-the-art BOW-based and part-based models.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126972396","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486272
Song Xiaorui, Wu Lingda
In view of the special properties of hyperspectral images(HSI) in the transform domain, in this paper, a new denoising method of HSI based on principal component analysis(PCA) and adaptive sparse coding is proposed. Firstly, the principal component image of each channel is obtained by performing PCA transform on the noisy HSI. Then, the first PCA output channels which contain a majority of the total energy of an HSI data cube are retained, and the rest PCA output channels which contain a small amount of energy, termed noise component images, are subjected to noise reduction through an adaptive sparse coding method. The encoding dictionaries are learned from each channel of noise component images by an approach of online dictionary learning. Finally, the denoised HSI is obtained by the inverse PCA transform. The proposed method takes the advantages of PCA and adaptive sparse representation that has better adaptability to the HSI. It not only performs better in denoising, but also preserves the details and alleviates the blocking artifacts well. The effectiveness of the proposed approach to hyperspectral denoising, termed PCASpC, is illustrated in a series of experiments with synthetic and realworld data where it outperforms the state-of-the-art.
{"title":"Denoising of Hyperspectral Images Based on Principal Component Analysis and Adaptive Sparse Coding","authors":"Song Xiaorui, Wu Lingda","doi":"10.1109/PRRS.2018.8486272","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486272","url":null,"abstract":"In view of the special properties of hyperspectral images(HSI) in the transform domain, in this paper, a new denoising method of HSI based on principal component analysis(PCA) and adaptive sparse coding is proposed. Firstly, the principal component image of each channel is obtained by performing PCA transform on the noisy HSI. Then, the first PCA output channels which contain a majority of the total energy of an HSI data cube are retained, and the rest PCA output channels which contain a small amount of energy, termed noise component images, are subjected to noise reduction through an adaptive sparse coding method. The encoding dictionaries are learned from each channel of noise component images by an approach of online dictionary learning. Finally, the denoised HSI is obtained by the inverse PCA transform. The proposed method takes the advantages of PCA and adaptive sparse representation that has better adaptability to the HSI. It not only performs better in denoising, but also preserves the details and alleviates the blocking artifacts well. The effectiveness of the proposed approach to hyperspectral denoising, termed PCASpC, is illustrated in a series of experiments with synthetic and realworld data where it outperforms the state-of-the-art.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125801428","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486359
Song Wenping, Niu Changling
Matching is the knotty point in photogrammetry and computer vision. Aiming at inaccurate corresponding points after preliminary matching, this paper proposed an image matching correction method of integrating least squares and phase correlation using window series. The method firstly uses least squares and phase correlation matching to correct corresponding points in utilizing of window series, and simultaneously calculates correlation coefficients using windows of different size. And then the correlation coefficients are used as the index of evaluating whether the corresponding image points are accurate or not. So the matching results with the largest correlation coefficients are chosen as the final results. Based on experimental data-set 1 and data-set 2, the experimental results revealed that the use of window series can significantly improve the correction accuracy of preliminary matching results. And the proposed method can correct the corresponding points of preliminary matching effectively and greatly improve the overall matching accuracy, which is better than least squares matching or phase correlation matching using window series and fixed windows.
{"title":"An Image Matching Correction Method of Integrating Least Squares and Phase Correlation Using Window Series","authors":"Song Wenping, Niu Changling","doi":"10.1109/PRRS.2018.8486359","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486359","url":null,"abstract":"Matching is the knotty point in photogrammetry and computer vision. Aiming at inaccurate corresponding points after preliminary matching, this paper proposed an image matching correction method of integrating least squares and phase correlation using window series. The method firstly uses least squares and phase correlation matching to correct corresponding points in utilizing of window series, and simultaneously calculates correlation coefficients using windows of different size. And then the correlation coefficients are used as the index of evaluating whether the corresponding image points are accurate or not. So the matching results with the largest correlation coefficients are chosen as the final results. Based on experimental data-set 1 and data-set 2, the experimental results revealed that the use of window series can significantly improve the correction accuracy of preliminary matching results. And the proposed method can correct the corresponding points of preliminary matching effectively and greatly improve the overall matching accuracy, which is better than least squares matching or phase correlation matching using window series and fixed windows.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124424153","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A model to estimate heavy metal content based on spectral analysis can provide the theory and method to rapidly obtain the heavy metal content in leaves. This study established a multiple stepwise regression model for selecting sensitive spectral bands, then used an extreme learning machine model optimized by particle swarm algorithm (PSOELM) to invert the contents of six metals in leaves in the Duobaoshan copper mine area in Heilongjiang Province, China. The results show that the Cu content of some leaves decreased with the distance from the copper mine therefore, the heavy metal content of leaves is related to mineral information. The PSOELM model is superior to both the back propagation model and extreme learning machine models in inversion accuracy and trend analysis.
{"title":"Inversion of Heavy Metal Content in a Copper Mining Area Based on Extreme Learning Machine Optimized by Particle Swarm Algorithm","authors":"Xinyue Zhang, X. Niu, Fengyan Wang, Xu Zeshuang, Xuqing Zhang, Shengbo Chen, Mingchang Wang","doi":"10.1109/PRRS.2018.8486172","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486172","url":null,"abstract":"A model to estimate heavy metal content based on spectral analysis can provide the theory and method to rapidly obtain the heavy metal content in leaves. This study established a multiple stepwise regression model for selecting sensitive spectral bands, then used an extreme learning machine model optimized by particle swarm algorithm (PSOELM) to invert the contents of six metals in leaves in the Duobaoshan copper mine area in Heilongjiang Province, China. The results show that the Cu content of some leaves decreased with the distance from the copper mine therefore, the heavy metal content of leaves is related to mineral information. The PSOELM model is superior to both the back propagation model and extreme learning machine models in inversion accuracy and trend analysis.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124141124","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486194
Qi Bi, K. Qin, Han Zhang, Wenjun Han, Zhili Li, Kai Xu
Building change detection is of great significance in high resolution remote sensing applications. Multi-index learning, one of the state-of-the-art building change detection methods, still has drawbacks like incapability to find change types directly and heavy computation consumption of MBI. In this paper, a two-stage building change detection method is proposed to address these problems. In the first stage, a multi-scale filtering building index (MFBI) is calculated to detect building areas in each temporal with fast speed and moderate accuracy. In the second stage, images and the corresponding building maps are partitioned into grids. In each grid, the ratio of building areas in time T2 and time T1 is calculated. Each grid is classified into one of the three change patterns, i.e., significantly increase, significantly decrease and approximately unchanged. Exhaustive experiments indicate that the proposed method can detect building change types directly and outperform the current multi-index learning method.
{"title":"Building Change Detection Based on Multi-Scale Filtering and Grid Partition","authors":"Qi Bi, K. Qin, Han Zhang, Wenjun Han, Zhili Li, Kai Xu","doi":"10.1109/PRRS.2018.8486194","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486194","url":null,"abstract":"Building change detection is of great significance in high resolution remote sensing applications. Multi-index learning, one of the state-of-the-art building change detection methods, still has drawbacks like incapability to find change types directly and heavy computation consumption of MBI. In this paper, a two-stage building change detection method is proposed to address these problems. In the first stage, a multi-scale filtering building index (MFBI) is calculated to detect building areas in each temporal with fast speed and moderate accuracy. In the second stage, images and the corresponding building maps are partitioned into grids. In each grid, the ratio of building areas in time T2 and time T1 is calculated. Each grid is classified into one of the three change patterns, i.e., significantly increase, significantly decrease and approximately unchanged. Exhaustive experiments indicate that the proposed method can detect building change types directly and outperform the current multi-index learning method.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131008987","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486330
Xia Wei, Sun Shikai, L. Jian
This paper studies to detect the change of road network from remote sensing images. Our purpose is to apply the method for practical usages, such as navigation map updating, road construction supervision, disaster survey, and so on. The proposed approach assumes that there is an outdated road map and the updating job is performed by detecting new road network and comparing the changes. The deep convolution network is utilized for precisely segmenting road areas. An image registration and correction procedure is performed to unify the spatial coordinate reference between the old map and the new road detection results. Then, we modify and standardize the extracted road segments, and apply it to determine the road variation of different periods. Experiments show that, the proposed method successfully identifies road changes, which is useful for fast map update in remote areas.
{"title":"Road Map Update from Satellite Images by Object Segmentation and Change Analysis","authors":"Xia Wei, Sun Shikai, L. Jian","doi":"10.1109/PRRS.2018.8486330","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486330","url":null,"abstract":"This paper studies to detect the change of road network from remote sensing images. Our purpose is to apply the method for practical usages, such as navigation map updating, road construction supervision, disaster survey, and so on. The proposed approach assumes that there is an outdated road map and the updating job is performed by detecting new road network and comparing the changes. The deep convolution network is utilized for precisely segmenting road areas. An image registration and correction procedure is performed to unify the spatial coordinate reference between the old map and the new road detection results. Then, we modify and standardize the extracted road segments, and apply it to determine the road variation of different periods. Experiments show that, the proposed method successfully identifies road changes, which is useful for fast map update in remote areas.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126227924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486201
Qian Wang, Pengfei Liu, Lifu Zhang
Simplex maximum distance (SMD) is an algorithm based on that the pixel with the biggest distance from simplex formed by known endmembers is most likely to be the next endmember. However, SMD involves calculation of some intermediate variables, such as simplex's normal vector, and intersection point of simplex and line, leading to computation complexity. In addition, high brightness points, outliers and isolated noise points in hyperspectral image are often extracted as endmembers in SMD. To overcome these two shortages, an improved simplex maximum distance (ISMD) algorithm is presented in the paper. To simplify computation procedure, ISMD defines the distance from pixel to simplex as ratio of volumes of parallel polyhedrons with adjacent dimensions. Once distances of all pixels from existing simplex are received, a set of similar pixels was selected from multiple pixels with a larger distance according to the spectral angle. The set of pixels is averaged to be the new endmember. The ISMD algorithm was assessed using simulated and real AVIRIS images. Compared with SMD, ISMD better extracted real endmembers in simulated image. And spectral angle between endmember obtained by ISMD and corresponding mineral from USGS spectral library is less for AVIRIS image.
单纯形最大距离(Simplex maximum distance, SMD)是一种基于与已知端元构成的单纯形距离最大的像素最有可能成为下一个端元的算法。然而，SMD涉及到单纯形法向量、单纯形与直线交点等中间变量的计算，计算量较大。此外，在SMD中，高光谱图像中的高亮度点、离群点和孤立噪声点经常被提取作为端元。为了克服这两个缺点，本文提出了一种改进的单纯形最大距离(ISMD)算法。为了简化计算过程，ISMD将像素到单纯形的距离定义为维度相邻的平行多面体的体积之比。一旦接收到现有单纯形中所有像元的距离，根据光谱角度从距离较大的多个像元中选择一组相似的像元。该像素集被平均为新的端元。采用模拟和真实的AVIRIS图像对ISMD算法进行了评估。与SMD相比，ISMD能更好地提取模拟图像中的真实端元。ISMD获取的端元与USGS光谱库中对应矿物的光谱角较小。
{"title":"An Improved Simplex Maximum Distance Algorithm for Endmember Extraction in Hyperspectral Image","authors":"Qian Wang, Pengfei Liu, Lifu Zhang","doi":"10.1109/PRRS.2018.8486201","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486201","url":null,"abstract":"Simplex maximum distance (SMD) is an algorithm based on that the pixel with the biggest distance from simplex formed by known endmembers is most likely to be the next endmember. However, SMD involves calculation of some intermediate variables, such as simplex's normal vector, and intersection point of simplex and line, leading to computation complexity. In addition, high brightness points, outliers and isolated noise points in hyperspectral image are often extracted as endmembers in SMD. To overcome these two shortages, an improved simplex maximum distance (ISMD) algorithm is presented in the paper. To simplify computation procedure, ISMD defines the distance from pixel to simplex as ratio of volumes of parallel polyhedrons with adjacent dimensions. Once distances of all pixels from existing simplex are received, a set of similar pixels was selected from multiple pixels with a larger distance according to the spectral angle. The set of pixels is averaged to be the new endmember. The ISMD algorithm was assessed using simulated and real AVIRIS images. Compared with SMD, ISMD better extracted real endmembers in simulated image. And spectral angle between endmember obtained by ISMD and corresponding mineral from USGS spectral library is less for AVIRIS image.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114884422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486338
Gou Mao, Yuan Yuan, Lu Xiaoqiang
Remote sensing image retrieval has many important applications in civilian and military fields, such as disaster monitoring and target detecting. However, the existing research on image retrieval, mainly including to two directions, text based and content based, cannot meet the rapid and convenient needs of some special applications and emergency scenes. Based on text, the retrieval is limited by keyboard inputting because of its lower efficiency for some urgent situations and based on content, it needs an example image as reference, which usually does not exist. Yet speech, as a direct, natural and efficient human-machine interactive way, can make up these shortcomings. Hence, a novel cross-modal retrieval method for remote sensing image and spoken audio is proposed in this paper. We first build a large-scale remote sensing image dataset with plenty of manual annotated spoken audio captions for the cross-modal retrieval task. Then a Deep Visual-Audio Network is designed to directly learn the correspondence of image and audio. And this model integrates feature extracting and multi-modal learning into the same network. Experiments on the proposed dataset verify the effectiveness of our approach and prove that it is feasible for speech-to-image retrieval.
{"title":"Deep Cross-Modal Retrieval for Remote Sensing Image and Audio","authors":"Gou Mao, Yuan Yuan, Lu Xiaoqiang","doi":"10.1109/PRRS.2018.8486338","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486338","url":null,"abstract":"Remote sensing image retrieval has many important applications in civilian and military fields, such as disaster monitoring and target detecting. However, the existing research on image retrieval, mainly including to two directions, text based and content based, cannot meet the rapid and convenient needs of some special applications and emergency scenes. Based on text, the retrieval is limited by keyboard inputting because of its lower efficiency for some urgent situations and based on content, it needs an example image as reference, which usually does not exist. Yet speech, as a direct, natural and efficient human-machine interactive way, can make up these shortcomings. Hence, a novel cross-modal retrieval method for remote sensing image and spoken audio is proposed in this paper. We first build a large-scale remote sensing image dataset with plenty of manual annotated spoken audio captions for the cross-modal retrieval task. Then a Deep Visual-Audio Network is designed to directly learn the correspondence of image and audio. And this model integrates feature extracting and multi-modal learning into the same network. Experiments on the proposed dataset verify the effectiveness of our approach and prove that it is feasible for speech-to-image retrieval.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121948353","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486379
K. Wohlfarth, C. Schröer, Maximilian Klass, Simon Hakenes, Maike Venhaus, S. Kauffmann, T. Wilhelm, C. Wohler
In this paper we explore the capabilities of two state-of-the-art machine learning techniques, transfer learning with convolutional neural networks (CNN) and support vector machines (SVM) to distinguish between 10 cloud genera. We will evaluate these methods using images acquired by the satellite Landsat 8. The classification of cloud genera is of high general relevance for remote sensing applications such as the surveillance of atmospheric or meteorological processes. Transfer learning is of advantage because it exploits neural networks, which are known to perform well, and enables the adaption to a specific problem with relatively small training data size. We will utilize Landsat 8 images for evaluating the examined machine learning approaches because these image data are freely available in large amounts. Downscaling the Landsat 8 images utilized for training to a resolution of about 300 meters per pixel will allow for keeping the CNN and SVM size reasonably low, such that our training data set can be restricted to a moderate size.
{"title":"Dense Cloud Classification on Multispectral Satellite Imagery","authors":"K. Wohlfarth, C. Schröer, Maximilian Klass, Simon Hakenes, Maike Venhaus, S. Kauffmann, T. Wilhelm, C. Wohler","doi":"10.1109/PRRS.2018.8486379","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486379","url":null,"abstract":"In this paper we explore the capabilities of two state-of-the-art machine learning techniques, transfer learning with convolutional neural networks (CNN) and support vector machines (SVM) to distinguish between 10 cloud genera. We will evaluate these methods using images acquired by the satellite Landsat 8. The classification of cloud genera is of high general relevance for remote sensing applications such as the surveillance of atmospheric or meteorological processes. Transfer learning is of advantage because it exploits neural networks, which are known to perform well, and enables the adaption to a specific problem with relatively small training data size. We will utilize Landsat 8 images for evaluating the examined machine learning approaches because these image data are freely available in large amounts. Downscaling the Landsat 8 images utilized for training to a resolution of about 300 meters per pixel will allow for keeping the CNN and SVM size reasonably low, such that our training data set can be restricted to a moderate size.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123637811","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486378
Peidi Yang, Wei Li, Xuebin Li, Lianru Gao
Currently, remote sensing image processing raises much higher requirements on the computing platform and processing speed. The high speed, low power, reconfigurable and radiation resistance features of Field Programmable Gate Arrays (FPGA) makes it become a better choice for real-time processing in hyperspectral imagery. In this paper, we have optimized the newly proposed hyperspectral target detection algorithm based on FPGA. The collaborative representation is a high-efficiency target detection (CRD) algorithm in hyperspectral imagery, which is directly based on the concept that the target pixels can be approximately represented by its spectral signatures, while the other cannot. Using the Sherman-Morrison formula to calculate the matrix inversion and the difficulty of implementing the overall CRD algorithm on the FPGA is reduced. The running speed of parallel programming is greatly promoted on the FPGA under the premise of reasonable resources. The experimental results demonstrate that the proposed system has significantly improved the processing time when compared to the pre-optimized system and the 3.40 GHzCPU.
{"title":"FPGA Optimization for Hyperspectral Target Detection with Collaborative Representation","authors":"Peidi Yang, Wei Li, Xuebin Li, Lianru Gao","doi":"10.1109/PRRS.2018.8486378","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486378","url":null,"abstract":"Currently, remote sensing image processing raises much higher requirements on the computing platform and processing speed. The high speed, low power, reconfigurable and radiation resistance features of Field Programmable Gate Arrays (FPGA) makes it become a better choice for real-time processing in hyperspectral imagery. In this paper, we have optimized the newly proposed hyperspectral target detection algorithm based on FPGA. The collaborative representation is a high-efficiency target detection (CRD) algorithm in hyperspectral imagery, which is directly based on the concept that the target pixels can be approximately represented by its spectral signatures, while the other cannot. Using the Sherman-Morrison formula to calculate the matrix inversion and the difficulty of implementing the overall CRD algorithm on the FPGA is reduced. The running speed of parallel programming is greatly promoted on the FPGA under the premise of reasonable resources. The experimental results demonstrate that the proposed system has significantly improved the processing time when compared to the pre-optimized system and the 3.40 GHzCPU.","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126605185","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","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：