Pub Date : 2018-08-01DOI: 10.1109/PRRS.2018.8486368
Chen-Gui Lu, Xiaomei Yang, Zhihua Wang, Yueming Liu
Automatic extraction of rural residential areas from high-resolution remote sensing images in large regions is a challenging task, because all kinds of background features, such as roads, green houses, and urban areas, must be excluded effectively by an extraction method. For the unsupervised methods of rural residential areas extraction, it is difficult to manually design features which are only sensitive to residential areas. At the same time, the supervised methods utilize training samples to obtain the discrimination between rural residential areas and the background features. However, manual labeling in large regions is tedious and time-consuming. The drawbacks of the existing methods for extracting rural residential areas limit their application in large regions. Therefore, we proposed a novel methodology for extracting rural residential areas in large regions based on fast acquisition of training samples and the fully convoluted network (FCN). A block-based method was proposed to extract rural residential areas rapidly and acquire training samples. Then, the large amount of training samples were used to train the FCN for rural residential area extraction. Finally, all ZY-3 satellite images in in the coastal area of Shandong, China were feed into the FCN, and the extraction result were obtained.
{"title":"Extracting Rural Residential Areas from High-Resolution Remote Sensing Images in the Coastal Area of Shandong, China Based on Fast Acquisition of Training Samples and Fully Convoluted Network","authors":"Chen-Gui Lu, Xiaomei Yang, Zhihua Wang, Yueming Liu","doi":"10.1109/PRRS.2018.8486368","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486368","url":null,"abstract":"Automatic extraction of rural residential areas from high-resolution remote sensing images in large regions is a challenging task, because all kinds of background features, such as roads, green houses, and urban areas, must be excluded effectively by an extraction method. For the unsupervised methods of rural residential areas extraction, it is difficult to manually design features which are only sensitive to residential areas. At the same time, the supervised methods utilize training samples to obtain the discrimination between rural residential areas and the background features. However, manual labeling in large regions is tedious and time-consuming. The drawbacks of the existing methods for extracting rural residential areas limit their application in large regions. Therefore, we proposed a novel methodology for extracting rural residential areas in large regions based on fast acquisition of training samples and the fully convoluted network (FCN). A block-based method was proposed to extract rural residential areas rapidly and acquire training samples. Then, the large amount of training samples were used to train the FCN for rural residential area extraction. Finally, all ZY-3 satellite images in in the coastal area of Shandong, China were feed into the FCN, and the extraction result were obtained.","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":"129951871","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.2018.8486248
Han Hai-hui, Wang Yilin, Zhang Zhuan, Ren Guang-li, Yang Min
Researchers have found that the mixed pixels exist in complex geological conditions often lead to distortion of altered minerals' spectral curves, and the accuracy of altered mineral extract from remote sensing data was reduced. Fortunately, the nonlinear analysis is a feasible solution. In this paper, by analyzing the nonlinear characteristics of the geological anomalies, the Fractal Dimension Change Point Method (FDCPM) will be used to extract the altered minerals' threshold from multispectral image. The realization theory and access mechanism of the model are elaborated by an experiment with ASTER data in Xinjinchang and Laojinchang gold deposits. The results show that the findings produced by FDCPM are agreed with well with a mounting body of evidence from different perspectives. The extracting accuracy over 86% show that FDCPM is an effective extrating method for remote sensing alteration anomalies, and it could be used as an useful tool for mineral exploration in similar areas in Beishan mineralization belt in northwest China.
{"title":"Extraction of Altered Mineral from Remote Sensing Data in Gold Exploration Based on the Nonlinear Analysis Technology","authors":"Han Hai-hui, Wang Yilin, Zhang Zhuan, Ren Guang-li, Yang Min","doi":"10.1109/PRRS.2018.8486248","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486248","url":null,"abstract":"Researchers have found that the mixed pixels exist in complex geological conditions often lead to distortion of altered minerals' spectral curves, and the accuracy of altered mineral extract from remote sensing data was reduced. Fortunately, the nonlinear analysis is a feasible solution. In this paper, by analyzing the nonlinear characteristics of the geological anomalies, the Fractal Dimension Change Point Method (FDCPM) will be used to extract the altered minerals' threshold from multispectral image. The realization theory and access mechanism of the model are elaborated by an experiment with ASTER data in Xinjinchang and Laojinchang gold deposits. The results show that the findings produced by FDCPM are agreed with well with a mounting body of evidence from different perspectives. The extracting accuracy over 86% show that FDCPM is an effective extrating method for remote sensing alteration anomalies, and it could be used as an useful tool for mineral exploration in similar areas in Beishan mineralization belt in northwest China.","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":"134021214","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.2018.8486163
S. Lou, Q. Fan, Feng Chen, Cheng Wang, Jonathan Li
Because of impacts resulted from sensor malfunction and clouds, there is usually a great deal of missing regions (pixels) in remotely sensed imagery. To make full use of the remotely sensed imagery affected, different algorithms for remote sensing images inpainting have been proposed. In this paper, an unsupervised convolutional neural network (CNN) context generate model was modified to recover the affected (or un-recorded) pixels in a single image without auxiliary information. Unlike existing nonparametric algorithms in which pixels located in surrounding region are used to estimate the unrecorded pixel, the proposed method directly generates content based on a neural network. To ensure recovered results with high quality, a modified reconstruction loss was used in training the model, which included structural similarity index (SSIM) loss and Ll loss. Comparison of the proposed model with bilinear interpolation was indicated through relative error. The performances of two methods in scenes with different complexity were discussed further. Results show that the proposed model performed better in simple scenes (i.e., with relative homogeneity), compared to the traditional method. Meanwhile, the corrupted images of channel blue were recovered more accurately, compared to the corrupted images of other channels (i.e., channel green and channel red). The relationship between scene complexities and channels shows that same scene has different complexities in different channels. The scene complexity presents significant correlation with recovered results, high complexity images are always accompanied by poor recovered results. It suggests that the recovering accuracy depends on scene complexity.
{"title":"Preliminary Investigation on Single Remote Sensing Image Inpainting Through a Modified GAN","authors":"S. Lou, Q. Fan, Feng Chen, Cheng Wang, Jonathan Li","doi":"10.1109/PRRS.2018.8486163","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486163","url":null,"abstract":"Because of impacts resulted from sensor malfunction and clouds, there is usually a great deal of missing regions (pixels) in remotely sensed imagery. To make full use of the remotely sensed imagery affected, different algorithms for remote sensing images inpainting have been proposed. In this paper, an unsupervised convolutional neural network (CNN) context generate model was modified to recover the affected (or un-recorded) pixels in a single image without auxiliary information. Unlike existing nonparametric algorithms in which pixels located in surrounding region are used to estimate the unrecorded pixel, the proposed method directly generates content based on a neural network. To ensure recovered results with high quality, a modified reconstruction loss was used in training the model, which included structural similarity index (SSIM) loss and Ll loss. Comparison of the proposed model with bilinear interpolation was indicated through relative error. The performances of two methods in scenes with different complexity were discussed further. Results show that the proposed model performed better in simple scenes (i.e., with relative homogeneity), compared to the traditional method. Meanwhile, the corrupted images of channel blue were recovered more accurately, compared to the corrupted images of other channels (i.e., channel green and channel red). The relationship between scene complexities and channels shows that same scene has different complexities in different channels. The scene complexity presents significant correlation with recovered results, high complexity images are always accompanied by poor recovered results. It suggests that the recovering accuracy depends on scene complexity.","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":"131138816","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.2018.8486413
Qiaoqiao Shi, Wei Li, R. Tao
Ship classification in optical remote sensing images is a fundamental but challenging problem with wide range of applications. Deep convolutional neural network (CNN) has shown excellent performance in object classification; however, limited available training samples prevent CNN for ship classification. In this paper, a novel ship-classification framework consisting of two-branch CNN and two dimensional discrete fractional Fourier transform (2D-DFrFT) is proposed. Firstly, the amplitude and phase information of ship image in 2D-DFrFT is extracted. Due to the fact that different orders of 2D-DFrFT have different contribution on the process of feature extraction of ship image. Thus the amplitude (M) and phase (P) value obtained in different orders are regarded as the input of two-branch CNN that can learn the high-level features automatically. After multiple features learning, decision-level fusion is adopted for final classification. The remote sensing image data, named as BCCT200-resize, is utilized for validation. Compared to the existing state-of-art algorithms, the proposed method has superior performance.
{"title":"2D-DFrFT Based Deep Network for Ship Classification in Remote Sensing Imagery","authors":"Qiaoqiao Shi, Wei Li, R. Tao","doi":"10.1109/PRRS.2018.8486413","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486413","url":null,"abstract":"Ship classification in optical remote sensing images is a fundamental but challenging problem with wide range of applications. Deep convolutional neural network (CNN) has shown excellent performance in object classification; however, limited available training samples prevent CNN for ship classification. In this paper, a novel ship-classification framework consisting of two-branch CNN and two dimensional discrete fractional Fourier transform (2D-DFrFT) is proposed. Firstly, the amplitude and phase information of ship image in 2D-DFrFT is extracted. Due to the fact that different orders of 2D-DFrFT have different contribution on the process of feature extraction of ship image. Thus the amplitude (M) and phase (P) value obtained in different orders are regarded as the input of two-branch CNN that can learn the high-level features automatically. After multiple features learning, decision-level fusion is adopted for final classification. The remote sensing image data, named as BCCT200-resize, is utilized for validation. Compared to the existing state-of-art algorithms, the proposed method has superior performance.","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":"124553983","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.2018.8486227
Xiangguang Leng, K. Ji, Shilin Zhou
Traditional ship segmentation methods in synthetic aperture radar (SAR) imagery are mainly based on the intensity/amplitude information. They cannot take fully advantage of the complex information in SAR imagery. This paper proposes a novel ship segmentation method based on kurtosis test in the complex-valued SAR imagery. It can take benefit of the complex information of the SAR imagery. The segmentation rationale is that sea clutter usually obey a Gaussian distribution while ship targets usually obey a sup-Gaussian distribution. Thus, their kurtosis can be different. Kurtosis is invariant with respect to location shift and positive scale changes. It follows that kurtosis of sea clutter remains approximately constant while the amplitude decreases with the incidence angle increasing. Preliminary experimental results based on Gaofen-3 and Sentinel-1 data show that the proposed method can achieve good performance.
{"title":"A Novel Ship Segmentation Method Based on Kurtosis Test in Complex-Valued SAR Imagery","authors":"Xiangguang Leng, K. Ji, Shilin Zhou","doi":"10.1109/PRRS.2018.8486227","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486227","url":null,"abstract":"Traditional ship segmentation methods in synthetic aperture radar (SAR) imagery are mainly based on the intensity/amplitude information. They cannot take fully advantage of the complex information in SAR imagery. This paper proposes a novel ship segmentation method based on kurtosis test in the complex-valued SAR imagery. It can take benefit of the complex information of the SAR imagery. The segmentation rationale is that sea clutter usually obey a Gaussian distribution while ship targets usually obey a sup-Gaussian distribution. Thus, their kurtosis can be different. Kurtosis is invariant with respect to location shift and positive scale changes. It follows that kurtosis of sea clutter remains approximately constant while the amplitude decreases with the incidence angle increasing. Preliminary experimental results based on Gaofen-3 and Sentinel-1 data show that the proposed method can achieve good performance.","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":"129017931","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.2018.8486259
Wei Li, Zhongjian Wang, Lu Li, Q. Du
Spectral-Spatial features are helpful for hyperspectral image classification. One of the most successful approaches based morphology is Extinction Profiles (EPs), which is constructed based on the component trees (Max-tree/Mintree) and can accurately extract spatial and contextual information from remote sensing images. However, the dimension of feature extracted by EPs with component trees is large, which potentially causes high redundancy. In order to reduce redundancy information and achieve better feature extraction, we propose a modified EP with the Topological trees (Inclusion tree). The proposed method is carried out on two commonlyused hyperspectral datasets captured over North-western Indiana and Salinas, California. The results show that the proposed method has significantly improved in terms of both accuracy and complexity on the basis of a reduction of half of the feature dimensions compared to the original EPs.
{"title":"Modified Extinction Profiles for Hyperspectral Image Classification","authors":"Wei Li, Zhongjian Wang, Lu Li, Q. Du","doi":"10.1109/PRRS.2018.8486259","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486259","url":null,"abstract":"Spectral-Spatial features are helpful for hyperspectral image classification. One of the most successful approaches based morphology is Extinction Profiles (EPs), which is constructed based on the component trees (Max-tree/Mintree) and can accurately extract spatial and contextual information from remote sensing images. However, the dimension of feature extracted by EPs with component trees is large, which potentially causes high redundancy. In order to reduce redundancy information and achieve better feature extraction, we propose a modified EP with the Topological trees (Inclusion tree). The proposed method is carried out on two commonlyused hyperspectral datasets captured over North-western Indiana and Salinas, California. The results show that the proposed method has significantly improved in terms of both accuracy and complexity on the basis of a reduction of half of the feature dimensions compared to the original EPs.","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":"130841538","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.2018.8486155
C. Qiu, M. Schmitt, Lichao Mou, Xiaoxiang Zhu
This study presents a classification framework for the urban Local Climate Zones (LCZs) based on a Residual Convolutional Neural Network (ResNet) architecture. In order to make full use of the temporal and spectral information contained in modern Earth observation data, multi-seasonal Sentinel-2 images are exploited. After training the ResNet, independent predictions are made from the multi-seasonal images. Subsequently, the seasonal predictions are fused in a decision fusion step based on majority voting. A systematical experiment is carried out in a large-scale study area located in the center of Europe. A significant accuracy improvement can be achieved by applying majority voting on multi-seasonal predictions. Based on the results, the main challenges and possible solutions of urban LCZ classification are further discussed, providing guidance for large-scale urban LCZ mapping.
{"title":"Urban Local Climate Zone Classification with a Residual Convolutional Neural Network and Multi-Seasonal Sentinel-2 Images","authors":"C. Qiu, M. Schmitt, Lichao Mou, Xiaoxiang Zhu","doi":"10.1109/PRRS.2018.8486155","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486155","url":null,"abstract":"This study presents a classification framework for the urban Local Climate Zones (LCZs) based on a Residual Convolutional Neural Network (ResNet) architecture. In order to make full use of the temporal and spectral information contained in modern Earth observation data, multi-seasonal Sentinel-2 images are exploited. After training the ResNet, independent predictions are made from the multi-seasonal images. Subsequently, the seasonal predictions are fused in a decision fusion step based on majority voting. A systematical experiment is carried out in a large-scale study area located in the center of Europe. A significant accuracy improvement can be achieved by applying majority voting on multi-seasonal predictions. Based on the results, the main challenges and possible solutions of urban LCZ classification are further discussed, providing guidance for large-scale urban LCZ mapping.","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":"121014650","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.2018.8486292
Jiao Jiao, Wu Lingda
In order to effectively combine the spectral information of the multispectral (MS) image with the spatial details of the panchromatic (PAN) image and improve the fusion quality, a fusion method based on morphological operator and improved pulse coupled neural network (PCNN) in mixed multi-scale (MM) domain is proposed. Firstly, the MS and PAN images are decomposed by nonsubsampled shearlet transform (NSST) to low- and high-frequency coefficients, respectively; secondly, morphological filter-based intensity modulation (MFIM) technology and stationary wavelet transform (SWT) are applied to the fusion of the low-frequency coefficients; an improved PCNN model is employed to the fusion of the high-frequency coefficients; thirdly, the final coefficients are reconstructed with inverse NSST. The experimental results on QuickBird satellite demonstrate that the proposed method is superior to five other kinds of traditional and popular methods: HIS, PCA, SWT, NSCT-PCNN and NSST-PCNN. The proposed method can improve the spatial resolution effectively while maintaining the spectral information well. The experimental results show that the proposed method outperforms the other methods in visual effect and objective evaluations.
{"title":"Fusion of Panchromatic and Multispectral Images via Morphological Operator and Improved PCNN in Mixed Multiscale Domain","authors":"Jiao Jiao, Wu Lingda","doi":"10.1109/PRRS.2018.8486292","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486292","url":null,"abstract":"In order to effectively combine the spectral information of the multispectral (MS) image with the spatial details of the panchromatic (PAN) image and improve the fusion quality, a fusion method based on morphological operator and improved pulse coupled neural network (PCNN) in mixed multi-scale (MM) domain is proposed. Firstly, the MS and PAN images are decomposed by nonsubsampled shearlet transform (NSST) to low- and high-frequency coefficients, respectively; secondly, morphological filter-based intensity modulation (MFIM) technology and stationary wavelet transform (SWT) are applied to the fusion of the low-frequency coefficients; an improved PCNN model is employed to the fusion of the high-frequency coefficients; thirdly, the final coefficients are reconstructed with inverse NSST. The experimental results on QuickBird satellite demonstrate that the proposed method is superior to five other kinds of traditional and popular methods: HIS, PCA, SWT, NSCT-PCNN and NSST-PCNN. The proposed method can improve the spatial resolution effectively while maintaining the spectral information well. The experimental results show that the proposed method outperforms the other methods in visual effect and objective evaluations.","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":"129196916","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 : 2018-07-24DOI: 10.1109/PRRS.2018.8486170
Shihao Sun, Lei Yang, Wenjie Liu, Ruirui Li
In recent years, Fully Convolutional Networks (FCN) has been widely used in various semantic segmentation tasks, including multi-modal remote sensing imagery. How to fuse multi-modal data to improve the segmentation performance has always been a research hotspot. In this paper, a novel end-to-end fully convolutional neural network is proposed for semantic segmentation of natural color, infrared imagery and Digital Surface Models (DSM). It is based on a modified DeepUNet and perform the segmentation in a multi-task way. The channels are clustered into groups and processed on different task pipelines. After a series of segmentation and fusion, their shared features and private features are successfully merged together. Experiment results show that the feature fusion network is efficient. And our approach achieves good performance in ISPRS Semantic Labeling Contest (2D).
{"title":"Feature Fusion Through Multitask CNN for Large-scale Remote Sensing Image Segmentation","authors":"Shihao Sun, Lei Yang, Wenjie Liu, Ruirui Li","doi":"10.1109/PRRS.2018.8486170","DOIUrl":"https://doi.org/10.1109/PRRS.2018.8486170","url":null,"abstract":"In recent years, Fully Convolutional Networks (FCN) has been widely used in various semantic segmentation tasks, including multi-modal remote sensing imagery. How to fuse multi-modal data to improve the segmentation performance has always been a research hotspot. In this paper, a novel end-to-end fully convolutional neural network is proposed for semantic segmentation of natural color, infrared imagery and Digital Surface Models (DSM). It is based on a modified DeepUNet and perform the segmentation in a multi-task way. The channels are clustered into groups and processed on different task pipelines. After a series of segmentation and fusion, their shared features and private features are successfully merged together. Experiment results show that the feature fusion network is efficient. And our approach achieves good performance in ISPRS Semantic Labeling Contest (2D).","PeriodicalId":197319,"journal":{"name":"2018 10th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2018-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133416171","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：