针对自动泊车过程中复杂环境所导致的模型目标检测精确度低、漏检率高等问题，提出一种基于YOLOv5（you only look once version 5）改进的自动泊车场景多目标检测模型：PSMD（parking scene multi-target detection）-YOLOv5。采用混合池化的方法对SPPF（spatial pyramid pooling- fast）进行改进，提升模型对全局信息的提取能力，从而增强模型检测的性能；引入BiFPN（bidirectional feature pyramid network）网络结构，调整不同尺度特征的重要性程度，加强模型的特征融合能力；加入结合了大卷积核的CA（coordinate attention）注意力机制，增强感兴趣的目标区域，进一步提升模型的特征提取能力；将CIoU（complete intersection over union）的损失函数替换为MPDIoU（minimum point distance based intersection over union）损失函数，加快模型的收敛以及提升定位能力。实验结果表明，PSMD-YOLOv5模型在自动泊车场景数据集上检测的查准率、召回率和mAP分别为81.8%、80.3%和85.3%，相比于原始YOLOv5模型整体的查准率、召回率和mAP分别提升了1%、6.5%和5.2%。并通过与其他目标检测模型的比较，验证了本文模型的有效性。

Detection accuracy in automatic parking is often compromised by complex environments, leading to high missed detection rates. To address these challenges, an improved multi-target detection model for parking scenes based on YOLOv5, named PSMD-YOLOv5 (parking scene multi-target detection YOLOv5), was proposed. A hybrid pooling method was used to improve the spatial pyramid pooling fast (SPPF) module, enhancing the model's ability to extract global information and improving detection performance. The bidirectional feature pyramid network (BiFPN) structure was introduced to adjust the importance of multi-scale features, strengthening feature fusion capability. A coordinate attention (CA) mechanism with large convolutional kernels was incorporated to enhance target regions of interest, further improving feature extraction. The complete intersection over union (CIoU) loss function was replaced with the minimum point distance based intersection over union (MPDIoU) loss function to accelerate model convergence and enhance localization accuracy. Experimental results show that on the automatic parking scene dataset, the PSMD-YOLOv5 model achieves precision, recall, and mAP values of 81.8%, 80.3%, and 85.3%, respectively. Compared to the original YOLOv5 model, overall precision, recall, and mAP are improved by 1%, 6.5%, and 5.2%. Furthermore, the effectiveness of the proposed model is verified through comparisons with other target detection models.