Ink painting classification method based on deep feature fusion

In response to the problem of insufficient classification accuracy caused by the neglect of multi-scale feature fusion in traditional convolutional neural networks (CNN) for ink painting classification tasks, this paper proposes a deep feature fusion based ink painting classification model. Firstly, a multi-scale feature extraction module is constructed to capture the stroke details and composition features of ink paintings through parallel convolutional kernels with different receptive fields. Secondly, a dual path attention fusion module is designed, which adopts a parallel mechanism of channel attention and spatial attention to achieve adaptive weighted fusion of cross level features, enhancing the feature expression ability of ink wash blending effect; simultaneously introducing a cross layer connection structure to promote the interaction and fusion of shallow texture features and deep semantic features. This study provides new technological ideas for intelligent appreciation in the protection of digital cultural heritage.