Secure DenseNet121-Powered Deep Learning for Satellite-Based Urban Flood Classification

Urban flood is a serious threat to facilities, life and economic development hence requires fast and efficient classification system for rescue missions. In this research, it is suggested to develop a Secure DenseNet 121 driven deep learning model for distinguishing urban flood from satellite images which will adopt manifested adversarial defense and blockchain techniques for better security and efficacy. The approach is based on multispectral and SAR imagery, and DenseNet121 of the images to improve the accuracy of flood mapping. The three key techniques involved in security improvement are as follows: Adversarial training, Homomorphic encryption, and Federated learning, which results in making data secure as well as accurate. By comparing the result obtained on this problem, it is found out that the proposed model has better performance than the existing deep learning architectures with classification accuracy of 96.8 %, precision of 97.5 %, recall of 96.3 %, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F 1$</tex>-score of 96.9 % for the classification of urban flood. Moreover, the developed system is implemented in a GIS web environment for monitoring and early warning purposes. The outcomes reveal that the developed model improves the result of flood classification and makes disaster management and mitigation safer, efficient, and secure from the privacy issues at the same time.