【摘要】 为提升施工现场管理水平，实现复杂施工场景下施工机械的实时检测，本文提出一种基于改进YOLOv4的施工机械检测方法。基于YOLOv4目标检测模型，将网络中普通3×3卷积替换为深度可分离卷积，使用轻量化特征提取网络替换模型主干网络，降低模型大小和参数量，提升模型检测速度。除此之外，在网络中添加注意力机制，在不影响检测速度的前提下提升模型检测精度。实验证明，本文提出的M1-DSC-YOLOv4+ECA算法在自制施工机械数据集上的平均准确率均值(mAP)达到了86.46%,检测速度为31.39 FPS,模型大小仅为原来的1/5,拥有高检测精度和实时检测速率，表明该算法能够满足施工场景下施工机械检测的准确性、实时性需求。更多还原

【Abstract】 In order to improve the management level of construction site and realize real-time detection of construction machinery in complex construction scenes, this paper proposes a construction machinery detection method based on improved YOLOv4. Based on YOLOv4 object detection network, the original 3×3 convolution is replaced with depthwise separable convolution, and lightweight feature extraction network is used to replace the backbone network for reducing the model size and the number of parameters and improving the detection speed. In addition, adding an attention mechanism to the network improves the model detection accuracy without affecting the detection speed. Experiments show that the mean Average Precision(mAP) of the M1-DSC-YOLOv4+ECA algorithm reaches 86.46% on the self-made construction machinery data set, the detection speed is 31.39 FPS, and the model size is only 1/5 of the original. It has high detection accuracy and real-time detection rate, indicating that the algorithm can meet the accuracy and real-time requirements of construction machinery detection in the construction scene.更多还原