Lightweight insulator target detection algorithm based on improved YOLOX

To address the challenges of high parameter volume, insufficient detection accuracy, and high false positive rates in traditional insulator target detection algorithms, this paper proposes a lightweight detection algorithm based on an enhanced YOLOX framework. First, Depthwise Separable Convolutions (DWConv) are employed to replace traditional convolutions, simplifying the model structure and reducing parameter count. Second, the Spatial Pyramid Pooling Fast-N (SPPF-N) module is introduced to substitute the traditional SPP module, and the last CSPLayer is removed, further compressing parameters while improving detection speed. Meanwhile, an Enhanced Convolutional Block Attention Module (E-CBAM) is embedded between the backbone and neck networks to significantly enhance feature extraction capabilities, overcoming the performance limitations of lightweight networks. Additionally, Coordinate Attention (CA) is integrated into the detection head to improve the precision of small target detection. Finally, the σWIoU loss function replaces the traditional IoU to accelerate convergence and optimize overall performance. Experiments show that compared to the original YOLOX, the proposed algorithm reduces parameters by 53.41% to 4.164 M and computational load to 12.975G, with mAP increased by 1.3%, detection accuracy reaching 98.81%, and recall achieving 100%. This algorithm balances high precision with lightweight design, meeting mobile deployment requirements and providing an efficient and reliable solution for insulator detection.