Ahmed R. El-gabri, Hussein A. Aly, Mohamed A. Elshafey, Tarek S. Ghoniemy
{"title":"EAS $$^2$$ KAM: enhanced adaptive source-selection kernel with attention mechanism for hyperspectral image classification","authors":"Ahmed R. El-gabri, Hussein A. Aly, Mohamed A. Elshafey, Tarek S. Ghoniemy","doi":"10.1007/s12145-024-01466-5","DOIUrl":null,"url":null,"abstract":"<p>Hyperspectral Images (HSIs) possess extensive applications in remote sensing, especially material discrimination and earth observation monitoring. However, constraints in spatial resolution increase sensitivity to spectral noise, limiting the ability to adjust Receptive Fields (RFs). Convolutional Neural Networks (CNNs) with fixed RFs are a common choice for HSI classification tasks. However, their potential in leveraging the appropriate RF remains under-exploited, thus affecting feature discriminative capabilities. This study introduces an Enhanced Adaptive Source-Selection Kernel with Attention Mechanism (EAS<span>\\(^2\\)</span>KAM) for HSI Classification. The model incorporates a Three Dimensional Enhanced Function Mixture (3D-EFM) with a distinct RF for local low-rank contextual exploitation. Furthermore, it incorporates diverse global RF branches enriched with spectral attention and an additional spectral-spatial mixing branch to adjust RFs, enhancing multiscale feature discrimination. The 3D-EFM is integrated with a 3D Residual Network (3D ResNet) that includes a Channel-Pixel Attention Module (CPAM) in each segment, improving spectral-spatial feature utilization. Comprehensive experiments on four benchmark datasets show marked advancements, including a maximum rise of 0.67% in Overall Accuracy (OA), 0.87% in Average Accuracy (AA), and 1.33% in the Kappa Coefficient (<span>\\(\\kappa \\)</span>), outperforming the top two HSI classifiers from a list of eleven state-of-the-art deep learning models. A detailed ablation study evaluates model complexity and runtime, confirming the superior performance of the proposed model.</p>","PeriodicalId":49318,"journal":{"name":"Earth Science Informatics","volume":null,"pages":null},"PeriodicalIF":2.7000,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Earth Science Informatics","FirstCategoryId":"89","ListUrlMain":"https://doi.org/10.1007/s12145-024-01466-5","RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}
引用次数: 0
Abstract
Hyperspectral Images (HSIs) possess extensive applications in remote sensing, especially material discrimination and earth observation monitoring. However, constraints in spatial resolution increase sensitivity to spectral noise, limiting the ability to adjust Receptive Fields (RFs). Convolutional Neural Networks (CNNs) with fixed RFs are a common choice for HSI classification tasks. However, their potential in leveraging the appropriate RF remains under-exploited, thus affecting feature discriminative capabilities. This study introduces an Enhanced Adaptive Source-Selection Kernel with Attention Mechanism (EAS\(^2\)KAM) for HSI Classification. The model incorporates a Three Dimensional Enhanced Function Mixture (3D-EFM) with a distinct RF for local low-rank contextual exploitation. Furthermore, it incorporates diverse global RF branches enriched with spectral attention and an additional spectral-spatial mixing branch to adjust RFs, enhancing multiscale feature discrimination. The 3D-EFM is integrated with a 3D Residual Network (3D ResNet) that includes a Channel-Pixel Attention Module (CPAM) in each segment, improving spectral-spatial feature utilization. Comprehensive experiments on four benchmark datasets show marked advancements, including a maximum rise of 0.67% in Overall Accuracy (OA), 0.87% in Average Accuracy (AA), and 1.33% in the Kappa Coefficient (\(\kappa \)), outperforming the top two HSI classifiers from a list of eleven state-of-the-art deep learning models. A detailed ablation study evaluates model complexity and runtime, confirming the superior performance of the proposed model.
期刊介绍:
The Earth Science Informatics [ESIN] journal aims at rapid publication of high-quality, current, cutting-edge, and provocative scientific work in the area of Earth Science Informatics as it relates to Earth systems science and space science. This includes articles on the application of formal and computational methods, computational Earth science, spatial and temporal analyses, and all aspects of computer applications to the acquisition, storage, processing, interchange, and visualization of data and information about the materials, properties, processes, features, and phenomena that occur at all scales and locations in the Earth system’s five components (atmosphere, hydrosphere, geosphere, biosphere, cryosphere) and in space (see "About this journal" for more detail). The quarterly journal publishes research, methodology, and software articles, as well as editorials, comments, and book and software reviews. Review articles of relevant findings, topics, and methodologies are also considered.