Current laser cleaning effect detection technologies are limited by high costs, operational complexity, and insufficient feedback information, restricting their widespread industrial application. To address these challenges, a lightweight semantic segmentation network based on deep learning is proposed for detecting paint removal effects. The model is optimized through techniques like model pruning and cross-platform inference acceleration to reduce parameter count and computational complexity. Finally, a combination of corner detection and edge detection algorithms is used to achieve target localization. A deployment algorithm for the detection system is also designed for experimental verification. Experimental results demonstrate that the proposed segmentation model not only effectively identifies the cleaning status but also provides accurate feedback on the residual paint layer information, facilitating parameter adjustment for continued cleaning, thus meeting the real-time detection requirements in the industrial sector.