基于改进YOLOX-s的车辆检测方法研究

Research on Vehicle Detection Method Based on Improved YOLOX-s

为缓解车辆小目标漏检及误检问题,提出一种基于YOLOX网络的多尺度特征融合的改进车辆检测模型。设计基于深度可分离卷积的Ghost-CSP(cross stage partial),替换网络的部分跨阶段局部结构,加快检测速度;将模型的最大池化方式改进为Softpool方式,并引入坐标注意力机制,增强待检测目标的特征表达,优化目标漏检问题;选用Focal Loss作为模型置信度损失函数以增加分类不准确样本的权重,提高模型对小目标的预测能力。实验结果表明:改进算法平均准确率提高到74.96%,速度达到73帧/s,在满足实时性要求下可以更好地完成车辆目标检测要求。

A improved vehicle detection model based on multi-scale feature fusion of YOLOX network is proposed to solve the problem of missing and false detection of small vehicle targets.Ghost-cross stage partial(CSP)based on the depth separable convolution is designed to replace part of cross stage partial in network to speed up the speed of detection.The max pooling mode of model is improved to Softpool mode,and coordinate attention mechanism is introduced to enhance the feature expression of target to be detected and to optimize the problem of target missing detection.Focal Loss is selected as the confidence loss function of model to increase the weight of inaccurate classification samples and improve the prediction ability of the model for small targets.The experimental results show that the average accuracy of the improved algorithm is improved to 74.96%,and the speed is up to 73 frames per second,which can better meet the requirements of real-time vehicle target detection.