Transformer Oil Leakage Detection Based on Improved AFT

In this paper, we propose an improved model based on Attention Fusion Transformer (AFT) network, which focuses on improving transformer oil leakage detection in substation scenarios. Conv-Attention(Conv-ATT) module is introduced to improve AFL network for dealing with the challenge of target detection in complex background, which enhances the accurate recognition ability of the model to the target in complex environment. Through the improved method, the detection ability mAP(Mean Average Precision, represents the mean value of Average Precision under different Recall.) is increased by 5.2 %, and the transformer oil leakage detection with higher accuracy is realized.