基于神经网络的金属零件表面字符检测与识别技术研究

Research on Part Surface Character Detection and Recognition Technology Based on Neural Network

摘要: 基于神经网络的零件表面字符检测与识别技术研究是通过对产品表面字符特征进行提取，再对提取的特征图像进行分割训练，从字符检测和字符识别两方面进行研究。字符检测方面，首先构建CTPN神经网络模型对原始图像进行特征分析检测，再用贝塞尔曲线进行改进。字符识别方面，首先搭建CRNN神经网络模型对于已处理的特征图像进行识别分析，将特征图像在CNN神经网络中进行特征提取，其次，将已提取的图像输入到RNN神经网络中进行分析预测，最后，再输入到CTC层进行字符转化并输出结果。经过试验测试验证，可以应用于对于产品表面字符识别。

Abstract: The character detection and recognition technology of parts surface based on neural network is studied by the character feature of product surface. Then the extracted image is segmented and trained, and the character detection and character recognition are studied. In terms of detection, firstly, the neural network is constructed to detect the original image and harvest improvements with Bezier curves. In character recognition, the CRNN neural network model is first built to recognize the processed feature images. Then, feature images are extracted in CNN neural network. Secondly, extracted images are input to RNN. Finally, it is input to CTC layer for character conversion and output results. It can be applied to product surface character recognition.