Abstract: With the growing demand for personalized customization and the prevalence of small-batch production, the shoe manufacturing industry is encountering increasing challenges in terms of efficiency and cost. Manual gluing remains the primary method for bonding soles and uppers; however, it suffers from low efficiency, inconsistent quality, and potential harm to workers’ health, making it a key bottleneck in industrial automation. To address these issues, in this paper, SoleSeg, a deep learning-based 2D contour segmentation algorithm designed for automatic extraction of gluing trajectories, was proposed. Built upon the YOLOv8n-seg framework, SoleSeg incorporates several key improvements: a bidirectional feature pyramid network (BiFPN) was introduced to enhance multi-scale feature fusion, a spatial-channel interactive attention (SCIA) was developed to strengthen feature extraction, and a lightweight fusion convolution module (LFConv) was proposed to reduce computational complexity. Experimental results demonstrate that compared with the original model, SoleSeg significantly reduced both computational overhead and the number of model parameters, making it well-suited for deployment in resource-constrained embedded devices. This advancement lays a solid technical foundation for achieving low-cost, high-precision recognition of 3D gluing trajectories.