Detecting Suspicious Military Tents from Optical Images using Deep Learning Model

Locating suspicious military tents or encampments is crucial for national security. Recently, deep learning methods have become increasingly popular for detecting military objects at country borders. Traditional methods, which rely on manual checks and morphological algorithms, are often inaccurate and slow at detecting such unauthorized activities. To overcome this issue, a solution that automatically detects suspicious tents from satellite images using deep learning techniques is crucial. An enhanced YOLOv8x-obb model is proposed and trained with a satellite image dataset that is well-prepared with various preprocessing and augmentation techniques. This model was validated and produced an accuracy of 91.9% and false alarm rate of 6.08%, specifying that the model works well in different terrains and situations, and it has the advantage of better border surveillance.