With the rapid development of the digital era, the inheritance of intangible cultural heritage is facing the crisis of fading, and this problem is particularly prominent in Hainan Li brocade culture. Aiming at these problems, this paper constructs a set of non-heritage pattern recognition and reconstruction system based on computer vision, and innovatively designs a deep learning architecture. The architecture adopts a coding and decoding network structure, which effectively retains the high-level semantic information and the underlying feature information contained in the pattern through the deep fusion of multi-scale features between different layers. An asymmetric convolutional layer is introduced in the feature extraction process to avoid additional consumption of computational resources and significantly enhance the model’s ability to capture pattern details. A pattern reconstruction architecture based on generative adversarial network is also developed, and a residual multi-head self-attention mechanism is incorporated into its generator, so that the reconstructed non-heritage patterns can maintain the original cultural characteristics and present a more delicate and realistic visual effect. In the specific application in the field of cultural and creative design, a complete digital resource library of non-heritage patterns is established to realize the intelligent matching between traditional cultural elements and contemporary design needs. The study proves that the deep learning-based pattern recognition and reconstruction method breaks through the limitations of traditional techniques in detail preservation and cultural characteristics inheritance, and realizes the two-way goal of digital protection and innovative application of non-heritage patterns. This research provides novel technical ways for the digital protection of intangible cultural heritage, and also indicates a brandnew direction for the innovative development of cultural and creative industries.
