Shape Matching Method for Steel Plate Scrap Based on Siamese Neural Networks

This paper proposes a two-dimensional shape matching method for steel plate remnants based on a Siamese Neural Network (SNN). A custom dataset comprising 20,000 polygon images is constructed, and a shared-parameter convolutional architecture is employed to optimize feature distances using contrastive loss. Experimental results demonstrate that the proposed SNN outperforms both traditional and existing deep learning methods across multiple evaluation metrics, exhibiting superior matching performance and generalization capability. This approach provides a reliable foundation for the intelligent identification of steel plate remnants.