Optical preprocessing and lightweight DenseNet for high-accuracy recognition of Longxiu embroidery images

High-resolution optical imaging and multi-spectral acquisition techniques provide fundamental support for high-precision image recognition of intangible cultural heritage such as Longxiu embroidery. By integrating advanced optical preprocessing methods—including high dynamic range imaging, optical edge enhancement, and multiscale texture feature extraction—this paper constructs a lightweight recognition model based on DenseNet and unstructured post-training pruning. The model achieves efficient knowledge transfer on a small-scale Longxiu embroidery dataset, significantly reducing the number of parameters while maintaining high recognition accuracy. Experimental results demonstrate that the proposed method not only improves recognition performance but also enables rapid inference, making it suitable for deployment on resource-constrained edge devices.