Abstract: A visible light and infrared image-matching algorithm based on improved SuperPoint and LightGlue was proposed to address the difficulty and high mismatch rate of image matching in regions with significant lighting differences and deformation in photovoltaic power station inspection robots. First, owing to the shortcomings of the native SuperPoint feature point extraction network with a large number of parameters and weak global feature extraction ability, a lightweight SuperPoint encoder based on the Conv2Former structure with a self-attention mechanism was adopted, and a strong global feature expression ability was obtained. Second, owing to the inability of native SuperPoint to maintain the spatial invariance of input data, STN modules were alternately inserted into the SuperPoint encoder to provide spatial invariance for the improved SuperPoint. Finally, in the feature-matching module, LightGlue was used to match the feature points extracted by the improved SuperPoint. Experimental results show that, compared with existing algorithms, the proposed algorithm achieves better matching performance, higher matching efficiency, and stronger robustness to lighting variations and deformation.