OCR智能识别技术在工器具管理中的研究与应用

目前电力行业所用工器具存在管理烦琐、出入库追溯困难等现象,针对电力行业的设备编码,基于OCR智能识别技术,进行定制化的技术开发。为提高该项技术的识别速率与准确率,制定了技术路线,分别研发了dskj_device_det、dskj_director_cls、dskj_pact和dskj_quant等技术算法,这些算法涉及到数据预处理、方向分类器、设备编码检测、设备编码识别、模型推理速度优化及模型开发与部署。运用该套算法编译的OCR智能识别程序对100套工器具进行扫描识别,试验结果为用时469 s,正确个数91套,识别准确率达到90%以上,达到推广的目标。

At present,there exist some problem swith the power industry,such as cumbersome management and difficult traceability management in and out of the warehouse.For the equipment coding of the power industry,customized technology development is carried out based on OCR intelligent identification technology.The deep learning framework tensorflow and pytorch are used as the main training framework.In order to improve the recognition speed and accuracy of this technology,the technical route is developed,and the technical algorithms such as dskj_device_det,dskj_director_cls,dskj_pact and dskj_quant are developed respectively.These algorithms involve data preprocessing,data enhancement scheme,detection algorithm,recognition algorithm,as well as direction classification algorithm and model compression and pruning algorithm.The OCR intelligent recognition program compiled by this algorithm is used to scan and identify 100 sets of tools.The test results show that it takes 469 s and the correct number is 91 sets,the recognition accuracy,reached more than 90%,and achieving the goal of promotion.