Aircraft detection in remote sensing image based on corner clustering and deep learning

Owing to the variations of aircraft type, pose, size and complex background, it remains difficult to detect aircraft effectively in remote sensing images, which plays a great significance in civilian and military. Classical aircraft detection algorithms still produce thousands of candidate regions and extract the features of candidate regions manually, which affects the detection performance. To address these difficulties encountered, an aircraft detection scheme based on corner clustering and Convolutional Neural Network (CNN) is proposed in this paper. The scheme is divided into two main steps: region proposal and classification. First, candidate regions are generated by utilizing mean-shift clustering algorithm to the corners detected on binary images. Then, the CNN is used for the feature extraction and classification of candidate regions that possibly contain the aircraft, and the location of the aircraft is finally determined after further screening. Compared with other classical methods, such as selective search (SS) + CNN, Edgeboxes + CNN and histogram of oriented gradient (HOG) + support vector machine (SVM), the proposed approach has a high accuracy and efficiency since it can automatically learn the essential features of the object from a large amount of data and produce fewer high quality candidate regions.