Application of painting style recognition and imitation technology based on deep learning in art education

Painting recognition in the field of computer vision remains a dynamic and significant research challenge, primarily because art style recognition does not depend on clearly defined or specific features. Instead, it requires the extraction of subtle and often non-deterministic stylistic patterns from visual artworks. Furthermore, utilising a deep convolutional neural network (CNN) overfits the model, whereas using a shallow CNN model results in underfitting. To overcome these challenges, we developed a painting style recognition-based Light Weight Improved CNN (PSR-LWICNN) model. To extract painting information and identify relevant art styles, we built the recommended model with two successive convolutional layers and one fully connected layer. We trained and tested the painting style recognition task using the Wikipaintings dataset. In this. Our proposed model surpasses many current state-of-the-art models, with an accuracy of 98.8 percent. The AUC, which classifies the model's performance, indicates that the model is stable, with an accuracy of 0.95. In the context of art education, this method offers a powerful solution to the limitations of traditional techniques particularly in recognizing and teaching styles such as Chinese painting by integrating intelligent and data-driven analysis into the learning process.