Abstract: The concentration of coal slurry in the clarification layer of thickeners is an important parameter for evaluating sedimentation and guiding reagent dosing in coal preparation plants. Traditional methods have relied on manual observation or instruments such as turbidity meters and interface meters, which often suffer from poor stability, limited real−time performance, and susceptibility to environmental disturbances. To overcome these limitations, this study took coal slurry from the Guobei Coal Preparation Plant as the research object and proposed a lightweight visual recognition method based on light transmission characteristics. Considering the low contrast and strong noise interference in coal slurry images, a dedicated preprocessing method was established. Comparative experiment results identified “bilateral filtering + CLAHE” as the optimal enhancement scheme, increasing root mean square contrast by 30.4% and image entropy by 12.8%. Grayscale processing further reduced data volume by 23% while retaining key luminance information. Subsequently, a YOLOv11−based concentration recognition model was developed, introducing C3K2 modules into the Backbone and Neck and integrating a C2PSA attention mechanism to enhance feature extraction, while maintaining a lightweight model structure. Training stabilized after approximately 150 epochs; at 200 epochs, the bounding box, classification, and distribution losses converged to 0.10, 0.53, and 0.77, respectively. Test results showed a recognition accuracy of 95.8%, an average detection speed of about 30 FPS, and Precision, Recall, and mAP95 of 92.5%, 93.2%, and 95.2%. Compared with conventional methods, the proposed approach enables continuous, stable, and real−time concentration recognition, providing a reliable technical solution for online intelligent monitoring of coal slurry clarification layer in thickeners for coal preparation plants.