The surface defects of mosaic ceramics on building walls not only compromise aesthetics but also pose structural safety and high-altitude falling hazards. Therefore, accurate defect detection is essential. Leveraging drone image data, this paper proposes a mosaic ceramic wall surface defect detection method based on an improved YOLO11 technique. Specifically, the standard convolution module of YOLO11 is replaced by the morphological-dilated-attention coupled convolution proposed in this study. Additionally, the large separable kernel attention module is fused with the channel and position self-attention module, and the spatial grouping enhancement technique is introduced into the backbone network. Together, these modifications establish the YOLO11-MLS model, which achieves both higher accuracy and model lightweighting. Experimental results demonstrate a detection accuracy of 92.9%, representing a 2.3% improvement over the original YOLO11. Moreover, the model achieves a single-image detection time of 22.8 ms, while reducing parameters and computation by 23.3% and 20.6%, respectively, realizing efficient, accurate, and lightweight detection of mosaic ceramic wall surface defects.
Dong et al. (Sat,) studied this question.