Oracle character detection based on improved Faster R-CNN

Oracle is the earliest known systematic writing in our country. It mostly originated from Shang Dynasty royal divination. The records cover a wide range of contents, including national politics, social atmosphere, and military wars. Oracle bone inscriptions are also the source of modem Chinese characters in my country, and they carry the history of Chinese civilization for thousands of years. At present, oracle bone characters still rely on manual identification and then handed over to experts to decipher, and the level of information is low. Therefore, in order to effectively detect oracle bones. In this paper, an improved object detection algorithm Faster R-CNN is proposed to detect oracle bone inscriptions. The algorithm uses ResNet101 as the backbone feature extraction network, and uses multi-scale feature fusion to improve text detection capabilities at different scales. Finally, the Soft-NMS algorithm is used to suppress redundant candidate regions. Experiments show that the average accuracy of the improved algorithm reaches 88.2%, which can effectively detect oracle bones.