【摘要】 为提高车辆检测算法精度，提出一种基于YOLOv5框架上增添新型轻量化注意力机制(novel lightweight attention module, NLAM)和多尺度特征检测层的算法。NLAM模块将深度可分离卷积的空间注意力模块和一维卷积的通道注意力模块进行并联融合，使NLAM模块参数量仅为8;增添多尺度特征检测层，提升小目标的检测精度。该算法在KITTI数据集训练和测试，实验结果表明，改进后算法平均精度为89.9%,相较于原始算法平均精度上涨2%,检测帧率为90 frame/s。该算法对车辆检测具有更高的小目标检测精度和更好的鲁棒性。更多还原

【Abstract】 To improve the accuracy of the vehicle detection algorithm, an algorithm based on the YOLOv5 framework with the addition of a novel lightweight attention module(NLAM) and a multi-scale feature detection layer was proposed. The NLAM module was a parallel fusion of the spatial attention module with depth-separable convolution and the channel attention module with one-dimensional convolution, which made the number of parameters of the NLAM module only 8. The multi-scale feature detection layer was added to improve the detection accuracy of small targets. The algorithm was trained and tested on the KITTI dataset. Experimental results show that the average accuracy of the improved algorithm is 89.9%, which is 2% higher than that of the original algorithm, and the detection frame rate is 90 frames/s. The algorithm has higher small target detection accuracy and better robustness for vehicle detection.更多还原