Abstract: High resolution remote sensing images often has the characteristics of huge data amount, complex background and low ground objects proportion. It is inefficient to use RCNN model for object recognition in high resolution remote sensing images directly. The cascade AdaBoost has the advantages of high recall rate and fast calculating speed, while it is likely to detect more false targets. In this article, a coarse to fine aircraft recognition method is proposed by combining RCNN model with cascade AdaBoost. Firstly, a HOG based cascade AdaBoost is applied to extract the candidate aircraft regions quickly. Then we classify the candidate regions with the support vector machine (SVM) classifier using features computed from convolutional neural networks. The results show that this method can ensure the accuracy and improve the computational efficiency.

Key words: AdaBoost, RCNN, aircraft recognition, high resolution remote sensing images