【摘要】 针对当前模型，对配电设备在复杂背景下存在识别准确度低的问题，构建了配电设备红外图像数据集，提出了一种基于深度学习的配电设备红外目标检测模型。基于YOLOv8s模型进行改进，在Neck部分将Concat替换为Concat＿BiFPN，促进多尺度特征融合；在Neck部分加入上下文聚合模块Context Aggregation，助力小目标检测；使用Wise-IoU v3代替原来的损失函数，聚焦训练过程中难以拟合的锚框的预测回归；最后，在YOLOv8s分类器与回归器部分增加小目标检测层，提升小目标的识别能力。研究结果表明，改进后的模型与原模型相比，准确度、召回率、MAP和F1分数分别提升了4.5%、3%、2.1%和3.7，可有效应用于配电设备的部件检测。更多还原

【Abstract】 In response to the problem of low recognition accuracy of current models for distribution equipment in complex backgrounds, this paper constructs an infrared image dataset of distribution equipment and proposes a deep learning based infrared target detection model for distribution equipment. This article is based on the YOLOv8s model for improvement, replacing Concats with Concat＿BiFPN in the Neck section to promote multi-scale feature fusion; Add the Context Aggregation module to the Neck section to assist in small object detection; Using Wise IoU v3 instead of the original loss function, focus on predicting regression for anchor boxes that are difficult to fit during the training process; Finally, add a small object detection layer to the YOLOv8s classifier and regressor section to enhance the recognition ability of small objects. The research results show that compared with the original model,the improved model has improved accuracy, recall, MAP, and F1 scores by 4.5%, 3%, 2.1%, and 3.7, respectively,and can be effectively used for component detection of distribution equipment.更多还原