To address the issue where rectangular anchor boxes in object detection-based Chinese character component segmentation cannot segment semi-enclosed Chinese characters, this paper proposes a method for segmenting Chinese character components based on Chinese character structure masks. This method utilizes a U-Net encoder with ResNet as the backbone network, transforming the segmentation of Chinese character components into the generation of Chinese character structure masks. First, this study proposes a Res-CBAM module, which leverages the structural features of Chinese characters by incorporating CBAM into the residual U-Net network, effectively solving the problem of incomplete segmentation of Chinese character components. Secondly, a vector-guided supervision mechanism is designed to guide the training process of the model by designing structure vectors of Chinese characters, effectively addressing the issue of component adhesion in Chinese characters. Experimental results demonstrate that compared to traditional object detection methods, this method can achieve fast and efficient segmentation in lightweight networks by training small datasets.