A insulator defect detection network based on improved YOLOv7 for UAV aerial images

The failure of insulators can compromise the safety of the entire power transmission system. Therefore, insulator identification and defect detection from massive UAV aerial image data has become an urgent problem to be solved in the field of operation and maintenance . To evaluate insulator integrity, we propose a lightweight insulator defect detection network model based on improved Yolo v7. First of all, it is proposed to use the long side definition method to add angle information to the detection frame, preprocess the aerial image of the UAV, and realize the target rotation frame detection, which helps to accurately identify and locate the insulator. Secondly, combining the deformable convolution with the ELAN module as the backbone network , a DCN-ELAN module is proposed to enable the model to learn more defect shapes, and a multi-scale bidirectional feature fusion module (MBFF) is added to adaptively adjust the weight of each feature map to enhance the adaptability to various targets. Especially improve the detection ability of small targets. Finally, RefConv convolution is used as the core to build a lightweight information representation network. This technique improves the network’s capability to handle image boundary details, simplifies the model’s complexity, and reduces computational requirements. The generalization of the model is verified on VOC and IDID datasets, and the experimental results show that the proposed method achieves better insulator defect detection performance than other methods. • Enhanced YOLOv7-based network for UAV images with precise rotated box detection. • Deformable convolutions and a fusion module improve defect and target detection. • Refocusing convolution reduces parameters, improving efficiency and accuracy. • Achieves superior defect detection performance validated on VOC and IDID datasets. datasets.