基于改进YOLO v5算法的道路小目标检测

Road small target detection based on improved YOLO v5 algorithm

为解决交通道路小目标检测难度大、精度低,容易出现错检漏检的问题,提出一种基于YOLO v5(you only look once v5)算法的多尺度特征融合目标检测改进算法。首先,增加小目标检测头用于适应小目标尺寸,缓解漏检情况。然后,引入可变形卷积网络v2(deformable convolutional networks V2,DCN V2)提高模型对运动中小目标的学习能力;同时,增加上下文增强模块,提升对远距离小目标的识别能力。最后,在替换损失函数、提高边界框定位精度的同时,使用空间金字塔池化和上下文空间金字塔卷积分组模块,提高网络的感受野和特征表达能力。实验结果表明,所提算法在KITTI数据集小目标类别上平均识别精度达到了95.2%,相较于原始YOLO v5,算法总体平均识别精度提升了2.7%,对小目标的检测效果更佳,平均识别精度提升了3.1%,证明所提算法在道路小目标检测方面的有效性。

In order to solve the problems that small targets on traffic roads faces including detection difficulty,low precision,detection failures,a multi-scale feature fusion target detection improvement algorithm based on the YOLO v5(you only look once v5)algorithm is proposed.Firstly,the small target detection head is added for adapting to the small target size and alleviating the missed detection.Then,deformable convolutional networks V2(DCN V2)is introduced to improve the model's learning ability for small targets in motion.The context augmentation module(CAM)is introduced to improve the recognition ability of small targets at a long distance.The replacement loss function is used to improve the bounding box's localization accuracy,and the spatial pyramid pooling and context spatial pyramid convolution_group(SPPCSPC_group)module is also used to improve the sensory field and feature expression ability of the network.The experiment results show that the proposed algorithm achieves an average accuracy of 95.2%in the category of small targets in the KITTI dataset,compared with the original YOLO v5 algorithm,the overall average accuracy is improved by 2.7%.For the detection of small targets,the average accuracy is improved by 3.1%with a better detection effect,which proves the effectiveness of the proposed algorithm for the detection of small targets on roads.