Leveraging mayfly optimization with deep learning for secure remote sensing scene image classification

Abstract

Current advancement in satellite sensor and remote sensing image (RSI) technologies allows high-resolution RSI, which provide spatial information about the world. In these circumstances, RSI scene classification is attracted considerable interest owing to its wide-ranging applications like environmental monitoring, urban planning, national defence security, and natural hazard detection. New deep learning (DL) achievements have resulted in the newest technology of RSISC applications; however, they lack trustworthiness and explainability. But RSIs of sensitive area is vulnerable to theft, interception, and loss due to the transmission through public channel or the storage in the cloud environment. Therefore, this study develops a mayfly optimization with deep learning (DL) based robust remote sensing scene image classification (MFODL-RRSSIC) model. The presented MFODL-RRSSIC model majorly aims to accomplish two objectives namely scene classification and security. For scene classification, the presented MFODL-RRSSIC technique employs NasNet feature extraction, MFO based hyperparameter tuning, and stacked autoencoder (SAE) classifier. Besides, the presented MFODL-RRSSIC model exploits modified deep belief network (MDBN) for intrusion detection and attaining security. A wide-ranging simulation analysis is made on popular datasets to illustrate the enhanced performance of the MFODL-RRSSIC technique.