SGooTY: A Scheme Combining the GoogLeNet-Tiny and YOLOv5-CBAM Models for Nüshu Recognition

With the development of society, intangible cultural heritage Chinese Nüshu is in danger extinction. To promote research and popularization traditional culture, we use deep learning to automatically detect recognize handwritten characters. address difficulties such as creation a character dataset, uneven samples, recognition, first build large-scale HWNS2023, by using various data augmentation methods. This dataset contains 5500 images 1364 labeled samples. Second, this paper, propose two-stage scheme model combining GoogLeNet-tiny YOLOv5-CBAM (SGooTY) for recognition. In stage, five basic models including AlexNet, VGGNet16, GoogLeNet, MobileNetV3, ResNet are trained tested on structure improved enhance accuracy recognising second combine an object detection re-recognize misidentified characters ensure overall system. Experimental results show that achieves highest 99.3% characters, which significantly improves recognition rate After integrating model, reached 99.9%.