Enhancing Deep Learning Model Performance by Integrating CBAM and CondConv Technologies
Xinqiang Wu, Guangyu Li, Shichen Li, Qi Ping Cao, Song Sen Yang, Haolin Li, Ming Liu · 2024
In the field of transmission line inspection, accurately identifying the status of insulators is crucial for ensuring the stable operation of the power system. To address the issue of low recognition rates under complex backgrounds associated with traditional methods, this paper proposes an improved model based on the YOLOv8 algorithm, integrating the Convolutional Block Attention Module (CBAM) and Conditional Convolution (CondConv) to enhance the detection accuracy and adaptability for transmission line insulators. The CBAM enhances the model's ability to identify specific insulator features by strengthening the representation of key spatial and channel features, while CondConv dynamically adjusts convolutional kernels to optimize the model's response to complex background variations. Experimental results demonstrate that, compared to the traditional YOLOv8 model, this architecture significantly improves the detection accuracy of insulators in varying environments while maintaining real- time processing speed. This research provides robust technical support for the development of automated transmission line inspection technologies.