Weighted multi-deep feature extraction for hybrid deep convolutional LSTM-based remote sensing image scene classification model

Abstract

The classification with remote sensing image scene model is the most critical issue in the perception of better visualization of remote sensing images. However, deep learning techniques have not been effective in seizing the hierarchical configuration of the elements in remote-sensing images. Thus, this work aims to expand a novel classification with remote sensing image scene representation for performing capable feature extraction and categorization. The multi-deep features are transformed into a normal weighted function and tuned by a hybrid optimization algorithm using DA and EFO named Adaptive Sense Area-based Dragon Fly Electric Fish Optimization (ASA-DEFO). From the simulation results, the precision of introduced ASA-DEFO-LSTM + DCN provides 11, 17, 14, 18, 20 and 18% better than SVM, pre-trained CNN, CNN, VGG16, Resnet-150 and LSTM-DCN for dataset 1. Finally, the experimental results show an improved performance of the suggested model using varied performance measures.