Addressing the issue of low detection accuracy and high false positive and false negative rates caused by the small size of insulator targets, a transmission line insulator detection model based on the improved YOLOv7 is proposed. Firstly, the dual-branch fused channel attention mechanism is integrated with the main ELAN (encoder-local aggregation network) module to emphasize crucial channel information and suppress interference from noise and irrelevant data. Secondly, a locally self-attentive mechanism is introduced in the feature fusion section to enhance the focus on local tiny regions. Additionally, the BiFPN (Bi-directional feature pyramid network) cross-layer connection is incorporated in the Neck section to preserve edge information better and improve the detection of small targets while slightly increasing computational load. Lastly, using evaluation metrics such as precision, recall, and mean average precision, ablation experiments and comparative experiments are conducted on collected datasets. Experimental results indicate that the improved network model achieves a detection accuracy of 92.1% for transmission line insulator detection, a 3% enhancement over the traditional YOLOv7 network model. The average detection mean, recall rate, are also improved by 3.1% and 3.6% respectively. Furthermore, the improved model demonstrates significant advantages over YOLOv5-ECA and Faster R-CNN in various evaluation metrics, proving its effectiveness in detecting transmission line insulators.