Substation Meter Defect Detection Based on Efficient Local Attention and Large Kernel Separation Convolution

A meter defect detection method based on efficient local attention and large kernel separation convolution was proposed to solve the problems of poor meter defect detection accuracy and low detection efficiency in substations. First, by adding efficient local attention mechanism to the neck structure of YOLOv8 model, the model feature fusion ability and model localization accuracy are improved; then, the spatial pyramid structure based on large kernel separation convolution is designed and used to efficiently process high-level features, reduce both computational complexity and parameter count, and enhance overall computational efficiency. Comparison experiments are conducted with the base model and the commonly used single-stage target detection model on the substation meter defect dataset. The experimental results demonstrate that the model considerably enhances the precision, recall, and average accuracy of meter defect detection, achieving improvements of $4.9 \%, 4.7 \%$, and $2 \%$, respectively over the baseline model, which indicates that the detection method based on efficient local attention and large kernel separation convolution ensures that it can efficiently deal with the high-level features while accurately identifying the region of interest. And the model can enhance the detection of substation meter defects, leading to improved performance and more reliable results.