Lightweight Detection Algorithm Model Tailored for Transmission Line Inspection Images

The quality of images collected during transmission line inspection is often affected by complex environmental factors, making it difficult to clearly present key information. To address the shortcomings of existing algorithms in detection accuracy and parameter redundancy, we propose a lightweight improved algorithm model based on the YOLOV8. Firstly, a lightweight downsampling module (Adown) is introduced into the backbone to reduce the number of model parameters and computational load, while efficiently extracting and preserving key information from the images. Secondly, the feature fusion network (TFNet) in the neck is redesigned, utilizing a multi-scale feature focusing module and a special feature diffusion mechanism to enable the generated feature maps to comprehensively capture cross-scale feature information, thereby facilitating more effective target detection and classification tasks. Finally, a lightweight dynamic detection head (TADDH) is adopted, which not only significantly reduces the number of parameters to make the model more portable, but also achieves dynamic feature selection through inter-task interaction mechanisms, enabling the acquisition of more useful information. Experimental results show that the improved model exhibits a 2.2 % increase in average detection accuracy compared to the original model, while reducing the number of parameters by 27.4 %, demonstrating superior performance. This achievement provides practical feasibility for deploying the model on mobile devices.