When components of non-material cultural heritage walk into modern artistic creation, they often experience at the same time pattern recombination, color re-arrangement, the shifting of handicraft marks, and the movement of cultural meanings. Methods which only depend on image similarity or category identification have difficulty in steadily catching these deep translational connections. To address this issue, this paper constructs a source-work paired corpus based on source ICH images, images of creative works, and their associated textual descriptions, and proposes the H-DFA hierarchical deep feature analysis framework.This method establishes a comprehensive representation across three dimensions-form retention, color-craft coupling, and semantic transfer-and converges the final deep feature scores through creation-adaptive weighting to identify the degree of retention, translation intensity, and cultural consistency of ICH elements across different works. The experimental section focuses on unified data partitioning, comparison methods, and evaluation interfaces, comparing methods such as HOG-SVM, CLIP, MICMLF, and WuMKG-guided, and verifying the model’s stability under occlusion perturbations and cross-scenario creative conditions.The results show that H-DFA achieves 90.8%, 89.9%, 0.844, and 0.810 for Accuracy, Macro-F1, mAP@10, and Spearman ρ, respectively, outperforming all comparison methods overall. Layer-by-layer ablation results indicate that semantic correction and creative adaptation weighting are critical components for improving ranking quality and expert consistency; Results under different scenarios show that this method can keep a leading level of performance in six creative work items: China traditional painting, decoration painting, picture drawing, wall picture design, electronic painting and mixed material art. This research makes clear that to build a stable corresponding relationship between ICH source elements and creative outcomes, and to arrange visual, technical, semantic information by using a layered deep feature framework, is able to more effectively support the quantitative analysis, creative assessment, digital utilization of ICH cultural elements within the creation of fine art.
