Residential Style Recognition Simulation Application Based on Computer Vision

Nowadays, the traditional rural residential style is gradually disappearing, and protecting and inheriting these cultural heritages with historical value has become an urgent task. This study uses computer vision technology to extract and classify the architectural styles of rural dwellings through deep learning models, thereby achieving automated recognition and analysis of traditional architectural styles. The specific methods include using image recognition and pattern recognition techniques to collect a large amount of image data of rural dwellings, extracting key features through pre trained CNN models, and effectively classifying them using SVM. During feature recognition, the performance of the SVM model significantly decreases with increasing noise, from 0.85 to 0.68, indicating its high sensitivity to noise; the performance degradation of the VGG-16 (Visual Geometry Group-16) model is relatively smooth, decreasing from 0.80 to 0.65, demonstrating strong noise robustness. ResNet-50 surpasses VGG-16 in rural residential style recognition with its deep residual network structure. Through this research, it can effectively assist relevant departments in making more scientific decisions in cultural heritage protection, planning, and revitalization, while providing data support and theoretical basis for subsequent related research, promoting the development of traditional residential protection work towards a more systematic and scientific direction.