【摘要】 为解决港口内供电设备漏油故障因油渍形态不规则、尺度多变导致的油渍检测精度低的问题，该文提出一种基于深度学习的检测方法。首先，采用并行多分支空洞卷积提取多尺度漏油特征，以应对目标尺度变化。其次，设计聚合特征融合结构，加强高层语义与低层细节的融合，提升对微小漏油的识别能力。此外，引入MPD-IoU损失函数，通过优化边界框角点距离回归，进一步提高定位精度。实验结果表明，该方法在漏油数据集上mAP@0.5达到84.6%，参数量与计算量分别为2.2 M和6.5 GFLOPs，有效平衡检测精度与模型复杂度，消融实验与热力图分析验证了各模块的有效性与协同性。更多还原

【Abstract】 To address the issue of low detection accuracy caused by the irregular shapes and varying scales of oil stains in power equipment oil leakage faults at ports,this paper proposes a deep learning-based detection method.First,parallel multi-branch dilated convolutions are employed to extract multi-scale oil leakage features,addressing the challenge of target scale variation. Second,an aggregated feature fusion structure is designed to enhance the integration of high-level semantic information and low-level details,thereby improving the recognition capability for minor oil leaks. Additionally,the MPD-Io U loss function is introduced to further enhance localization accuracy by optimizing the regression of bounding box corner distances. Experimental results demonstrate that the proposed method achieves an mAP@0.5 of 84.6% on the oil leakage dataset,with parameter count and computational cost of 2.2 M and 6.5 GFLOPs,respectively,effectively balancing detection accuracy and model complexity. Ablation studies and heatmap analysis validate the effectiveness and synergy of each module.更多还原