Linear pattern detection of building groups by integrating dynamic snake convolution with YOLO11

Accurately detecting the patterns of building groups is the premise and foundation of building generalization. Due to the fuzziness and uncertainty of building patterns, it is difficult to describe them with unified rules, making this issue a key and challenging research focus in the field of cartographic generalization, which limits the level of automation in building generalization. With the development of artificial intelligence, object detection models have made significant progress in image classification and segmentation. This paper introduces the YOLO11 object detection model to achieve the detection of building groups with linear patterns by integrating the dynamic snake convolution (DSC) which is used to enhance the feature extraction capability. Experimental results show that the improved YOLO11-DSC model has better performance compared to the original YOLO11 and another two commonly used strategies (improving with CBAM, AKConv). At last, a typification example is given based on the detected linear patterns which demonstrates the usability in generalization.