基于GhostNet与注意力机制的YOLOv5交通目标检测

针对交通目标检测模型参数量大、检测精度低、检测速度慢、泛化性差等问题,提出一种基于GhostNet与注意力机制的YOLOv5交通目标实时检测模型.采用基于遗传算法的K-means聚类方法获取适用于车辆检测的最佳预选框;采用轻量的Ghost卷积提取目标特征,并构建基于CSP结构的C3Ghost模块,大幅度压缩模型参数量,降低计算成本,提高计算速度;在特征融合层添加Transformer block和CBAM注意力模块,来探索模型特征提取潜力以及为模型在密集对象的场景中寻找注意力区域; UA-DETRAC数据集上的消融实验和综合性能评价结果表明所提模型平均精度达到98.68%,参数量为47 M,检测速度为65 FPS,与YOLOv5相比,参数量压缩了34%,速度提升43%,平均精度提高了1.05%.

YOLOv5 Traffic Object Detection Based on GhostNet and Attention Mechanism

Traffic object detection models have massive parameters, low detection accuracy and speed, and poor generalization. In view of these problems, YOLOv5 real-time traffic object detection model based on GhostNet and attention mechanism is proposed. The K-means clustering method based on genetic algorithms is used to obtain the best prior bounding box suitable for vehicle detection. The lightweight GhostConv is used to extract target features, and the C3Ghost module based on the CSP structure is constructed, which can greatly reduce the number of model parameters, reduce the calculation cost, and improve the calculation speed. Transformer block and CBAM attention module are added in the feature fusion layer to explore the potential of feature extraction of the model and find attention regions for the model in scenarios with dense objects. The results of ablation experiments and comprehensive performance evaluation on the UA-DETRAC data set show that the average accuracy of the proposed model reaches 98.68%, the number of parameters is 47 M, and the detection speed is 65 FPS. Compared with YOLOv5, the number of parameters is reduced by 34%, the speed is increased by 43%, and the average accuracy is increased by 1.05%.