基于YOLOv5-DSC的漂浮物目标检测

YOLOv5-DSC based floating object target detection

针对河道漂浮物检测识别困难的问题,提出一种改进型的检测识别网络YOLOv5-DSC。首先,在YOLOv5骨干网络中加入无参数注意力机制SimAM,在不增加模型参数的情况下,提高模型的特征提取能力;其次,在特征融合网络中使用基于深度可分离卷积(DSC)的DSCSP结构,减少模型的计算量;最后,采用SIoU损失函数代替原YOLOv5网络模型中的CIoU损失函数。SIoU损失函数重新定义了回归距离损失,加快了网络的收敛速度。在漂浮物数据集上进行了实验验证,结果表明,YOLOv5-DSC平均精度均值达到了98.5%,检测速度为145 f/s。

In allusion to the problem of river floating object detection and recognition,an improved detection and recognition network YOLOv5-DSC is proposed.SimAM,a parameter-free attention mechanism,is added to the YOLOv5 backbone network to improve the feature extraction capability of the model without increasing the model parameters.The DSCSP structure based on depth-separable convolution is used in the feature fusion network to reduce the computation of the model.The SIoU loss function is used to replace the CIoU loss function in the original YOLOv5 network model.The SIoU loss function can redefine the regression distance loss and speed up the convergence of the network.The experimental validation was performed on the floater dataset,and the results show that the mean accuracy of YOLOv5-DSC can reach 98.5%on average,with a detection speed of 145 f/s.