Land Cover Mapping for LISS IV Multispectral Satellite Images Using Self Attention Convolutional Neural Network

Abstract

Deep learning techniques is the one of the emerging fields in machine learning and it has a various application such as object classification in remote sensing images, medical image analysis, recognition of various patterns, object identification, handwriting generation etc. Deep learning techniques are using supervised and unsupervised approaches to learn multiple levels illustrations of spatial, spectral and textural features of hierarchical architecture for the roles of land cover monitoring including land use and land cover classification and change detection. In this research work, we analyzed the performance of self-attention convolutional neural network (SACNN) based on the application of land cover mapping. Experiment result proved that, classification accuracy of SACNN is higher than traditional CNN and FCM algorithm.