Feature Extraction and Classification of Traditional Architectural Styles Based on Convolutional Neural Network

With the diversification of architectural culture and the increase in protection needs, the automatic recognition and classification of traditional architectural styles has become an important topic in the research field. At present, traditional CNN (Convolutional Neural Networks) has difficulty in capturing the spatial characteristics and adaptive weights of different buildings when extracting traditional architectural features, and the classification accuracy is low. This paper studies the extraction and classification of traditional architectural style features based on the VGG19 (Visual Geometry Group 19) model and the channel-spatial attention mechanism. The study first used the selective search algorithm to generate and extract the main candidate regions of traditional buildings, and input the candidate regions of traditional architectural styles into VGG19 as features to extract features such as window decorations and carvings, wall textures and materials, and architectural layout. Then, combined with CSAM (Channel Spatial Attention Module), the correlation between the extracted features was used to adaptively assign corresponding weights to style features with high importance, and the spatial features of components of different architectural styles were extracted. Finally, the Softmax classifier is used to construct the mapping relationship between feature descriptors and classification labels, and the classification labels and their probabilities are predicted for traditional architectural style images. The experiment is based on the public dataset of traditional Chinese architecture, and features are extracted and classified for traditional architectural style images. The results show that the accuracy of the VGG19-CSAM (Visual Geometry Group 19-Channel Spatial Attention Module) model reaches 0.98, which is 0.03 higher than the VGG19 model, and the F1 reaches 0.96. The test outcomes indicate that the combination of convolutional neural network VGG19 and CSAM significantly improves the performance of feature extraction and classification of traditional architectural styles, and facilitates the conservation process of traditional architectural styles.