Scenic Handwritten Poetry Recognition Based on Fusion of Multi-scale Features and Knowledge Graph

Handwritten poetry in scenic areas has complex background texture, diverse font size and style, which makes it difficult for tourists in scenic areas to recognize handwritten poetry. Therefore, the recognition scene of handwritten poetry in scenic areas was first analyzed and studied, the detection of poetry in scenic areas-network (DPSA-Net) was designed to extract the characteristics of handwritten poetry in scenic areas at different scales, and the handwritten poetry detection in scenic areas combined with the link dependency between handwritten poetry characters was realized. Second, a convolution recurrent aggregation network (CRA-Net) was designed to recognize handwritten poetry in scenic areas, which combines convolutional neural networks (CNN) and bi-directional long short-term memory network to extract sequence features of handwritten poetry images, and the transformation from features to text by aggregation cross-entropy (ACE) was realized. Finally, combining the scenic areas knowledge graph, the output of CRA-Net was corrected, so as to improve the recognition accuracy of handwritten poetry in scenic areas. Results show that after correcting the recognition results of CRA-Net by handwritten poetry correction technology in scenic areas, the recognition accuracy reaches 79.04%. At the same time, this technology has good anti-interference ability and good application prospect.