Abstract: In complex environments such as heavy dust and ore adhesion at the feed inlet of open−pit mine crushing stations, acquired ore images are prone to quality degradation and blurred contours. Conventional segmentation algorithms struggle to accurately extract ore contours, affecting the accuracy and real−time performance of ore particle size analysis. To improve ore image segmentation accuracy in complex environments and reduce the computational load on edge devices, a lightweight ore image segmentation method, ALEC−YOLO11, is proposed. First, to address false and missed detections caused by ore image degradation, a two−stage image preprocessing workflow combining enhancement and dehazing was designed. CLAHE and Unsharp Mask were used to improve image contrast and edge details, and the dark channel prior optimized by guided filtering was employed to suppress interference from dust and water mist, resulting in higher−quality images. Second, the Zoom_cat module was utilized to perform cross−scale feature alignment and aggregation through adaptive pooling and upsampling, improving the model's ability to recognize ores of different particle sizes. Third, the ADown module was introduced in the bottom−up path to replace strided convolution, thereby enhancing the model's ability to preserve edge information. Finally, an asymmetric lightweight decoupled segmentation head (ASLD−Segment) was designed at the output end to compress the classification and regression branches and enhance the mask branch, thereby achieving effective decoupling of multi−task features. Experimental results show that, compared with the YOLO11n−seg model, the proposed method achieves an mAP₅₀ of 0.918 without reducing mAP₅₀₋₉₅. Meanwhile, the model parameter count is reduced by approximately 13.1%, and the end−to−end processing speed reaches 145 FPS, indicating that lightweight deployment is achieved while maintaining stable accuracy, which satisfies the real−time processing demands of industrial scenarios.