Research on Emotion Analysis of Chinese Literati Painting Images Based on Deep Learning

Abstract

Starting from a pure-image perspective, using machine learning in emotion analysis methods to study artwork is a new cross-cutting approach in the field of literati painting and is an effective supplement to research conducted from the perspectives of aesthetics, philosophy, and history. This study constructed a literati painting emotion dataset. Five classic deep learning models were used to test the dataset and select the most suitable model, which was then improved upon for literati painting emotion analysis based on accuracy and model characteristics. The final training accuracy rate of the improved model was 54.17%. This process visualizes the salient feature areas of the picture in machine vision, analyzes the visualization results, and summarizes the connection law between the picture content of the Chinese literati painting and the emotion expressed by the painter. This study validates the possibility of combining deep learning with Chinese cultural research, provides new ideas for the combination of new technology and traditional Chinese literati painting research, and provides a better understanding of the Chinese cultural spirit and advanced factors.