Deep Learning Based Approach for Aerial Surveillance System

Military operations, urban planning, environmental monitoring, and disaster management all benefit from modern aerial observation. In order to identify critical infrastructure, including airports, highways, ports, railroad stations, and defense zones, our work focuses on deep learning-based classification of high-resolution aerial photos. We train and assess three CNN models: MobileNetV2, VGG16, and DenseNet121. VGG16 enhances feature extraction, DenseNet121 increases accuracy by effective feature reuse, and MobileNetV2 offers a lightweight solution for real-time applications. To increase robustness and generalization, data augmentation is used. Accuracy, precision, recall, and F1-score are used to assess model performance. According to experimental results, all models function well, with MobileNetV2 being appropriate for real-time aerial surveillance applications and DenseNet121 offering the highest accuracy.