【摘要】 针对现有的车辆检测网络模型大、不易部署的问题,提出一种基于改进YOLOv3-tiny的轻量级车辆检测网络。改进YOLOv3-tiny的特征提取网络,提高车辆检测的速度和准确性,将空间金字塔池化(Spatial Pyramid Pooling,SPP)融合到网络中,进行特征的拼接,提高网络的学习能力,利用距离交并比(Distance Intersection over Union,DIoU)损失函数来提高网络的性能。实验结果表明,所提出的轻量级网络与YOLOv3-tiny网络相比,模型缩小了0.1 Mb,检测精度提高了5.64%,检测速度满足车辆实时检测的需求。更多还原

【Abstract】 To address the problem that existing vehicle detection networks have large models and are not easy to deploy, this paper proposes a lightweight vehicle detection network based on the improved YOLOv3-tiny. The feature extraction network of YOLOv3-tiny is improved to enhance the speed and accuracy of vehicle detection, the Spatial Pyramid Pooling(SPP) is fused into the network for feature stitching to improve the learning capability of the network, and the Distance Intersection over Union(DIoU) loss function is used to improve the performance of the network. The experimental results show that the proposed lightweight network in this paper, compared with the YOLOv3-tiny network, has a reduced model of 0.1 Mb, an improved detection accuracy of 5.64% and a detection speed that meets the demand for real-time vehicle detection.更多还原