Advances in Deep Learning for Remote Sensing: Automatic Target Detection and High-Resolution Image Classification

The research has been conducted by introducing deep learning models for automatic target detection and high resolution image classification on remote sensing, especially based upon a selected region. The research further evaluate its performance using four of the state-of-the-art deep learning architectures - Convolutional Neural Network (CNN), Residual Network (ResNet), Densely Connected Convolutional Networks (DenseNet) and Visual Geometry Group Network. For each model, evaluation is using several performance metrics that quantitatively assess our approach's ability to detect reports of human and vehicle presence in satellite and camera images. The results show that the ResNet obtained a better performance metric where one can observe it has accuracy of 0.94, 0.92 precision of 0.92, recall of 0.93 and F1 Score of 0.92. The second best algorithm was DenseNet with the highest accuracy of 0.93, the precision of about 0.91 and recall also around at 0.92, F1-score similar to that of ResNet. CNN highly performed with accuracy of 0.92, precision of 0.89, and recall of 0.91 and F1-score of 0.90. VGG had competitive performance as it achieved an accuracy of 0.91, precision of 0.88, recall of 0.90 and F1-score of 0.89. These findings summaries our understanding of the potential for technological intrusion in ecology, yielding exciting new techniques and methods that can be used to monitor environmental change, perform security surveillance or coordinate disaster response over extremely large spatial scales across remote regions.