【摘要】 针对YOLOv5s基准模型参数量大、特征尺度差异化等问题，提出基于高效卷积注意力特征融合的轻量级目标检测模型.首先，构建基于幻影操作的轻量化特征提取模块，在保证检测精度接近原模型的前提下，提高模型的实时性.其次，优化通道注意力和空间注意力模块，提出基于高效卷积的注意力特征融合模块，并设计兼具检测精度与实时性的轻量级目标检测模型.在具有不同复杂道路场景的数据集BDD100K上开展实验.结果表明，相较于基准模型，所提模型的检测精度和推理速度均得到提升，其中全类平均检测精度提升了1.4%,帧率提升了28.2%.相较于当前行业应用中主流的深度学习模型，所提模型在精度与速度的均衡性上表现出显著优势.更多还原

【Abstract】 A lightweight object detection model based on efficient convolutional attention feature fusion was proposed to address the issues of large number of parameters and feature scale differences in the YOLOv5s benchmark model. Firstly, a lightweight feature extraction module based on phantom operation was constructed to improve the real-time performance of the model while ensuring detection accuracy close to the original model. Secondly, the channel attention and spatial attention modules were optimized, and an attention feature fusion module based on efficient convolution was proposed. Meanwhile, a lightweight object detection model with high detection accuracy and real-time performance was designed. Experiments were conducted on the dataset BDD100K with different complex road scenes. The results show that the designed model is improved in detection accuracy and inference speed compared with the benchmark model. The average detection accuracy of the entire class is improved by 1.4%, and the frame rate is improved by 28.2%. Compared with mainstream deep learning models in current industry applications, the proposed model shows significant advantages in the balance between accuracy and speed.更多还原