Regarding the issues of unclear target contrast, weak edges, and severe noise interference caused by strong lighting in images collected by coal mine monitoring systems, an improved de-hazing model for strong light image is proposed based on multi-scale Taylor-Former and long-term contextual attention in this paper. In this method, Taylor expansion is used to approximate softmax function in the attention mechanism, thereby linear computational complexity can be achieved. At the same time, multi-scale attention and multi branch architecture are introduced to obtain features of images under strong lighting. Finally, a long-term contextual attention mechanism is added to the improved TaylorFormer to enhance useful target features while weakening the interference information. Based on the coal mine monitoring image dataset, the improved method in this paper is compared with several other methods, and ablation experiments are conducted with the basic TaylorFormer and multi-scale TaylorFormer. The experimental results have verified that the proposed method has achieved good improvements in peak signal-to-noise ratio, structural similarity, feature similarity, and information entropy, and the dehazing visual effect is superior.