Acquiring sensor nameplate text images in emission chimney environments often results in blurred features and unsatisfactory recognition rates due to complex and variable lighting conditions caused by unevenly distributed dust suspensions generated during production. In order to enhance the detailed feature expression of the image, improve the information content of the nameplate image, solve the problem of low recognition rate brought about by complex lighting, especially in the case of non-uniform lighting, and enhance the performance of text recognition, a Retinex-based non-uniform luminance self-adjustment enhancement algorithm is proposed to preprocess the captured text image of the nameplate. The image effects before and after processing of the classical image processing algorithms and the algorithm proposed in this paper are subjectively compared qualitatively and objectively analysed quantitatively, and the image processing effect is better, which proves the advancement of the preprocessing algorithm proposed in this paper.