Wide-Angle Threat Detection: Fisheye Imaging and Machine Learning for Enhanced Drone and Bird Surveillance

The research investigates expanding security threats which unauthorized drones and birds present within vulnerable zones of airports and military installations. Solar energy systems present challenges because their narrow surveillance angles lead to areas where monitoring is absent. The proposed work introduces a new system which utilizes fisheye lenses with deep learning algorithms to establish wideangle real-time threat detection capabilities. The research used a total of 4,101 images where 2,091 depicted drones alongside 2,010 bird images which underwent fisheye transformation and data augmentation procedures to model actual site simulations. The evaluation tested seven deep learning models with CNN and both ResNet and DenseNet among them. Our performance evaluation consisted of measuring model accuracy as well as precision and recall rate and F1-score. DenseNet delivered the greatest accuracy which equated to 99.66% and F1-score which reached 99.66% for detecting distorted fisheye images with high precision. GoogleNet performed just behind the primary models with 99.55% accuracy while the custom CNN model achieved 99.22% accuracy. The established system shows its capability to detect and discriminate drones and birds reliably in panoramic image evaluations. The study demonstrates how deep learning with fisheye imaging creates a distinguished surveillance approach which operates efficiently along with scalability. Future research will minimize the model size for edge processing and expand the data collection for enhancing model generalization. The security enhancement method shows potential to improve airspace protection in critical structures and metropolitan areas.