基于卷积神经网络的文本框识别算法在电力业务系统上的应用研究

Research on the Application of Text Box Recognition Algorithm Based on Convolutional Neural Network in Power Service System

针对在电力行业上业务办理终端信息录入效率低的问题,提出一种基于卷积神经网络(CNN)的文本框识别算法。采用Faster RCNN网络对文本框数据集进行训练与验证,结合OCR技术开发辅助录入系统。通过引入基于CNN的文本框识别算法,兼容不同系统的业务终端应用,在不改变原系统架构的情况下,提高了算法的适用性。实验结果表明,基于CNN的文本框识别算法应用于辅助录入系统上,相对于人工录入方式在信息录入速度与准确性有显著提升,在电力行业的业务办理终端上具有广泛应用前景。

A text box recognition algorithm based on a convolutional neural network (CNN) is proposed to address the low efficiency in information input of business terminals in the power industry. A Faster RCNN network is used to train and validate the text box dataset, and combined with OCR technology to develop an auxiliary input system. By introducing a CNN-based text box recognition algorithm, the algorithm’s applicability is improved for business terminal applications across different systems without changing the original system architecture. Experimental results show that the CNN-based text box recognition algorithm applied to the auxiliary input system significantly improves information input speed and accuracy compared to manual input methods and has broad application prospects in business terminals in the power industry.