改进YOLOv3在航拍目标检测中的应用

Application of Improved YOLOv3 in Aerial Target Detection

近年来,基于深度学习的航拍目标检测在无人驾驶、军事侦察、灾害检测等领域有着广泛的应用,更精确、高效的算法是目前航拍目标检测研究的热点与难点。提出一种基于改进YOLOv3算法的航拍目标检测方法,对航拍数据集进行目标先验框维度聚类、优化锚点参数,提高了网络对航拍目标的检测有效性。同时对原网络进行改进,减少部分卷积操作并引入跳跃连接机制降低特征冗余,提高了检测准确率,并降低了小目标的误检率与漏检率。实验结果表明,改进YOLOv3算法相较于原始YOLOv3算法的效果有明显提升,对于较高分辨率的航拍图像,加快了网络的收敛速度,并在保证实时性的前提下,将检测平均准确率(mean Average Precision,mAP)提高了12.7%。

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.