Authenticity Assessment of Industrial Heritage Villages Based on Multimodal Deep Learning

With the acceleration of urbanization and continuous cultural tourism involvement, industrial heritage villages have gradually exposed weakened authenticity during renewal and reuse. Some regions emphasize landscape effects and commercial functions in renovation while ignoring original production and living logic, making village spaces increasingly homogeneous. Other regions adopt conservative protection methods that restrict usage behaviors, causing heritage spaces to separate from real life. Current authenticity assessment methods rely heavily on manual investigation and experience-based judgment. Assessment results are easily influenced by subjective factors and fail to support long-term, dynamic management requirements. To address these problems, this paper proposes an authenticity assessment method for industrial heritage villages based on multimodal deep learning. Starting from spatial form, production activities and cultural use scenarios, this research constructs a multimodal dataset integrating image and text information. Deep learning models extract features related to living heritage transmission. On this basis, a comprehensive authenticity assessment model is established. Experimental results show this method can accurately identify key living heritage elements and form stable authenticity level classification. The research provides an operable technical approach for intelligent protection and assessment of industrial heritage villages.