Oil leak detection in substation equipment based on PFDAL-DETR network

Although deep-learning detection methods are widely used in power systems, there is still potential for improvement in the efficiency and automation of oil leak detection in substation equipment. This study proposes an online oil leak detection algorithm for substation equipment based on the PFDAL-DETR model. The optimization of feature sharing among convolution kernels through the perceptual field dual attention mechanism (PFDAM) enhances feature extraction efficiency in deep-learning models. A thin-head feature extraction network incorporating a lightweight model (LM) is proposed to preserve the gradual transmission of spatial feature information to the channels. A loss function based on shape intersection over union (Shape-IoU) is used to address the issue of class imbalance and improve model performance. Experimental results demonstrate that the proposed model achieves a mean average precision (MAP@0.5) of 91.20%. This surpasses current state-of-the-art detection models such as YOLO and DETR, providing meaningful research implications for the detection of oil leaks in substation equipment.