Abstract: To address the issue of weak light and strong reflections within the gas insulated switchgear (GIS) cavity environment, we proposed an algorithm to further improve the detection accuracy of the internal inspection robot in the cavity by introducing a highlight removal method based on the dark channel prior and dual-threshold segmentation. Initially, an approximate pseudo-mirror reflection image is generated using the dark channel prior. Subsequently, a proposed dual-threshold segmentation function identifies the image's highlight regions as a MASK, precisely optimizing the pseudo-mirror reflection image using this MASK to prevent erroneous pixel separation in the diffuse reflection component areas. Finally, an alternating direction method of multipliers (ADMM) algorithm optimizes and solves the L₁-L₂ mixed variational model constrained with the pseudo-mirror reflection image to accomplish the removal of image highlights. Qualitative and quantitative experiments are conducted on public datasets and GIS chamber datasets demonstrate that, compared to traditional classic highlight removal algorithms, the proposed algorithm effectively separates mirror reflection components and diffuse reflection components, efficiently removes highlights and avoids image distortion and has a certain practical significance in improving the accuracy of GIS cavity fault detection.