【摘要】 随着我国机动车保有量激增，路内停车的管理问题愈发凸显。为解决现有检测算法模型精度低、相对误差大、难以准确判定违停行为的问题，作者提出了基于改进A-YOLOM的路内泊车的模型。针对模型在特征提取过程中未能减少通道相似性的问题，提出了利用具有通道混洗功能的RCS-OSA模块改进骨干网络；针对模型检测颈部网络未能区分特征通道重要性，提出了利用EMA多尺度注意力机制改进颈部网络。改进后的模型命名为REA-YOLOM模型，实验结果表明，改进策略单一作用时有效，且共同作用时模型在目标检测精度AP50与mAP50-95上分别提高了2.4%和2.2%,分割精度mIoU提升0.3%,有效提高了文中检测对象的检测精度。更多还原

【Abstract】 With the rapid growth in the number of motor vehicles in China, the management of on-street parking has become an increasingly critical issue.To address the limitations of existing detection algorithms, such as low precision, large relative errors, and the inability to accurately identify illegal parking, this paper proposes a model for on-street parking detection based on an improved A-YOLOM.To mitigate the issue of channel similarity during the feature extraction process, we introduce the RCS-OSA module, which incorporates channel shuffling, to enhance the backbone network.Additionally, to address the failure of the neck network to differentiate the importance of feature channels, we propose the integration of an EMA-based multi-scale attention mechanism to refine the neck network.The resulting model is named REA-YOLOM.Experimental results demonstrate that each of the proposed improvements is effective when applied individually, and when combined, the model achieves an increase of 2.4% in AP50 and 2.2% in mAP50-95.Moreover, the segmentation accuracy(mIoU) improves by 0.3%,thereby significantly enhancing the detection precision of key objects in this task.更多还原