Multi-Scale Feature Extraction Image Dehazing Networks

Our paper introduces an end-to-end multi-scale feature fusion dehazing network designed specifically for special robots operating on high-voltage transmission towers. The network utilizes a feature pyramid structure for multi-level feature extraction and combines various feature attention mechanisms to eliminate the effects of fog in images. Experimental validation shows that this dehazing network can effectively improve image quality, particularly under adverse weather conditions such as heavy fog, significantly enhancing contrast and detail clarity. This, in turn, enhances the visual perception capabilities of the robot when performing inspection and maintenance tasks. Additionally, the method demonstrates good robustness and adaptability under different lighting and fog density conditions.