关键词: 目标检测, 航拍图像, 深度学习, YOLOv3算法

Abstract:

Recently, object detection of aerial image based on deep learning has been widely used in the field of unmanned driving, military reconnaissance, and disaster detection. More accurate and efficient algorithms are the hotspots and difficulties in this field. An aerial target detection method based on improved YOLOv3 algorithm is proposed. The effectiveness of network detection improvement is achieved by clustering target priori box dimension and optimizing the anchor parameters on aerial dataset. In addition, this paper reduces partial convolution operation and adds short cut to reduce feature redundancy, thereby improving the detection accuracy, reducing the false and missed detection rate of small target. The experimental shows that the improved YOLOv3 algorithm has a significant improvement over the original effect. For high-resolution aerial image, the convergence speed of the network is accelerated, and the mean Average Precision of detection is improved by 12.7% under the premise of ensuring real-time performance.

Key words: object detection, aerial image, deep learning, YOLOv3 algorithm