CenterUNet: a Model Fused Semantic Segmentation for Flange Center Point Detection

In the flange segment assembly process of the shipbuilding industry, the accurate detection of the flange’s center point is a key technical challenge for improving the precision and efficiency of automated assembly. Traditional methods rely on manual detection, often limited by insufficient detection accuracy, resulting in low assembly efficiency and unstable precision. To address these issues, this paper proposes a deep learning model combining image semantic segmentation with regression algorithms, called CenterUNet. The model first effectively separates tar-get objects from backgrounds through image segmentation technology, significantly reducing environmental interference. It then uses an optimized regression algorithm to precisely calculate the coordinates of the flange center points, breaking through the recognition limitations of traditional methods in environments. The main advantages of CenterUNet lie in its outstanding accuracy and efficiency. Compared to ResNet18,VGG and EfficientNetV2, our model not only achieves significant improvements in the accuracy of flange center point detection but also shows significant enhancements in stability and robustness. This holds significant practical significance and application value for advancing intelligent manufacturing and automated assembly.