Block-Scrambling-Based Encryption with Deep-Learning-Driven Remote Sensing Image Classification

• Published in: Remote sensing (Basel, Switzerland) Vol. 15; no. 4; p. 1022
• Main Authors: Alsubaei, Faisal S, Alneil, Amani A, Mohamed, Abdullah, Mustafa Hilal, Anwer
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.02.2023
• Subjects: Access control; Airports; Automation; Biometrics; Classification; Cloud computing; Data analysis; Data encryption; Data integrity; Datasets; Deep learning; Distance measurement; Domains; Environmental protection; Equipment and supplies; Feature extraction; hyperparameter tuning; Image classification; image encryption; Image processing; Image retrieval; Internet; Machine learning; Medical imaging equipment; Multimedia; National security; Privacy; privacy-preserving; Remote sensing; remote-sensing images; Resource surveys; Safety and security measures; security; Servers; Surveys; Weather forecasting
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs15041022

Remote sensing is a long-distance measuring technology that obtains data about a phenomenon or an object. Remote sensing technology plays a crucial role in several domains, such as weather forecasts, resource surveys, disaster evaluation and environment protection. The application of remote-sensing images (RSIs) is extensive in some specific domains, such as national security and business secrets. Simple multimedia distribution techniques and the development of the Internet make the content security of RSIs a significant problem for both engineers and scientists. In this background, RSI classification using deep learning (DL) models becomes essential. Therefore, the current research article develops a block-scrambling-based encryption with privacy preserving optimal deep-learning-driven classification (BSBE-PPODLC) technique for the classification of RSIs. The presented BSBE-PPODLC technique follows a two-stage process, i.e., image encryption and classification. Initially, the RSI encryption process takes place based on a BSBE approach. In the second stage, the image classification process is performed, and it encompasses multiple phases, such as densely connected network (DenseNet) feature extraction, extreme gradient boosting (XGBoost) classifier and artificial gorilla troops optimizer (AGTO)-based hyperparameter tuning. The proposed BSBE-PPODLC technique was simulated using the RSI dataset, and the outcomes were assessed under different aspects. The outcomes confirmed that the presented BSBE-PPODLC approach accomplished improved performance compared to the existing models.