Military Aircraft Detection from Optical Satellite Images Using Convolutional Neural Networks

Aircraft detection using optical images plays a vital role in various military operations contributing to security, safety, efficiency, and environmental sustainability in the aviation sector. Recent works such as CGC-net, YOLOv5, DenseNet have showed remarkable outcomes in military aircraft detection. Although the models demonstrate significant results, they often exhibit low speed and unreliable accuracy in detecting aircrafts from foggy optical images. To address these issues, we plan on developing a deep learning model that detects military aircrafts with high speed and accuracy particularly in challenging weather conditions. This paper proposes an enhanced CNN model that has been well trained with MAR20 dataset, with various pre-processing techniques applied. The trained model has consistently shown reliable results when tested with unseen data. This model was validated and it yielded an accuracy of 92.35 % and over all false positive rate is 0.41 %, indicating that it performs well in a variety of landscapes and circumstances, with an additional advantage of improved border surveillance.