Aircraft Detection in Satellite Images Using CNN and Sliding Window Method

Object detection from remote sensing imagery is an important avenue of research, covering applications in urban planning, agriculture, transport, civil safety, and military activities. Detecting stationary aircraft in and around airport locations has always remained very challenging due to the complex images and overlapping objects. This paper aims at developing a CNN-based model to classify if there is an aircraft or not and employs a sliding window detection algorithm that minimizes overlap for detection of aircraft. The proposed detection algorithm includes the CNN model and takes a satellite image as input that preliminary scans and identifies only high-confidence regions again rescan low-confidence regions for improved detection with smaller stride. Experimental results show that detection is highly accurate that have demonstrated efficiency in working with varied inputs from satellite images. These findings indicate the strength of Convolutional Neural Network in achieving exactness and computational efficiency in detecting aircraft from complex remote sensing scenarios.