Abstract: The detection of mechanical external damage hazards of transmission lines faces challenges such as insufficient extraction of single-characteristic image information and insufficient feature discrimination in scenarios with occlusion and low contrast. Traditional methods fail to effectively integrate multi-source characteristic information, resulting in high rates of missed detection and false detection. Therefore, a multi-characteristic image interactive learning method is proposed for detection. A tri-channel complementary representation system, which comprises grayscale, pseudo-color, and line-type modalities, is constructed from the original image. Subsequently, a cross-characteristic attention fusion module coupled with a channel-space collaborative weighting strategy is devised to facilitate feature interaction and enhancement. Furthermore, integrating a deformable convolution module into the YOLOv8 architecture accommodates target deformations, while the optimization of the loss function further refines detection accuracy. Experiments show that, compared with three comparative methods, the mean average precision (mAP) of this method is increased by 17.0, 8.1, and 3.7 percentage points, respectively. In complex working conditions such as occlusion and low contrast, the detection performance of this method is significantly superior to that of the comparative methods.