Research on rapid classification of artistic paintings based on an improved LSTM algorithm

With the rapid advancement of digitization in the arts, the need to develop intelligent systems capable of efficiently analyzing and managing artistic painting collections has become increasingly important. These systems must handle large-scale visual data while accounting for diverse artistic styles and complex visual patterns. This research proposes a deep learning approach that combines an improved Long Short-Term Memory (LSTM) model with feature optimization techniques to enhance the classification of artistic paintings. Initially, the LSTM model is employed to extract temporal patterns from the paintings' visual data, focusing on recurrent relationships between key visual features, such as color gradients, textures, and brushstroke sequences. To further refine the model, the input data is pre-processed using convolutional layers to extract spatial features before feeding it into the LSTM. To improve the performance of the LSTM model, enhancements like attention mechanisms and adaptive learning rate schedules, including Adam and cyclical learning rates, are introduced. These mechanisms help the model better capture complex visual features, avoid overfitting, and ensure faster convergence. The effectiveness of the improved LSTM model was validated through comprehensive tests on multiple datasets. Additionally, experiments conducted on the SemArt dataset reveal that the proposed method accurately classifies paintings by type, genre, and period, outperforming conventional approaches.