講演情報
[U06-03]Web Services and Convolutional Neural Networks for Satellite Image Classification
*Joel Bandibas1、Shinji Takarada1 (1.Geological Survey of Japan)
キーワード：
Web Map Service、Web Processing Service、Convolutional Neural Networks、Satellite Image Classification、WebGIS
Web Map Service (WMS) and Web Processing Service (WPS) are the main services currently used for geospatial data rendition and processing within the online environment, respectively. While many satellite image data providers are currently sharing remotely sensed data as WMS, very few WPS are available for satellite image processing. Formulating WPS to classify satellite images shared as WMS provides a very accessible and useful computing resource for many applications. Convolutional Neural Networks (CNN) have been used for pattern recognition in images, including satellite image classification. Unlike the conventional parametric statistic classifiers, CNN can accurately classify images that have high spatial frequency where the spectral patterns of land cover types are not normally distributed. Furthermore, CNN can be trained to include pixel contextual information, the spatial signature of land cover types, increasing the classification accuracy. This study focuses on the formulation of WPS using CNN to classify satellite images served as WMS. The study successfully formulated the CNN WPS and used to classify satellite images served as color composite WMS. The classification results also show that the CNN classifier is more accurate compared to the conventional Maximum Likelihood Classifier (MLC). A WebGIS portal is also successfully developed as WMS and WPS client for easy web service implementation. The study highlights the feasibility of using color composite WMS, from data exclusively used for visual interpretation, to inputs for important image processing computations.