阅读说明：本技术 电力专业的设备铭牌识别算法 (Electric power professional equipment nameplate recognition algorithm ) 是由 曾纪钧 龙震岳 温柏坚 刘晔 张金波 蒋道环 梁哲恒 沈桂泉 张小陆 沈伍强 邓 于 2021-07-16 设计创作，主要内容包括：本发明提供一种电力专业的设备铭牌识别算法,以多个神经网络搭建模型框架,利用Faster-RCNN和BiLSTM对铭牌图像数据进行文本检测,利用DenseNet和CTC对铭牌图像数据进行文本识别。经过对电力现场作业中的设备铭牌图像的数据训练后,得到识别算法模型。该识别算法对水平横排文本信息识别率高,能排除特殊符号的干扰。该识别算法可以使用于黑底白字、白底黑字、白底红字、黄底黑字文本检测和识别,字体的类型是黑体、宋体和微软雅黑时效果最佳。(The invention provides an electric power professional equipment nameplate recognition algorithm, which is characterized in that a model frame is built by a plurality of neural networks, text detection is carried out on nameplate image data by using fast-RCNN and BilSTM, and text recognition is carried out on the nameplate image data by using Densenet and CTC. And obtaining a recognition algorithm model after data training of the equipment nameplate image in the power field operation. The recognition algorithm has high recognition rate on horizontal text information and can eliminate the interference of special symbols. The recognition algorithm can be used for detecting and recognizing black-matrix white characters, white-matrix black characters, white-matrix red characters and yellow-matrix black characters, and the type of the characters is the best effect when the characters are black bodies, Song dynasty bodies and Microsoft elegant black.)

1. An electric power professional equipment nameplate recognition algorithm is characterized by comprising the following steps:

acquiring data of a nameplate image;

secondly, carrying out text detection and recognition on nameplate image data by using a neural network framework consisting of CTPN, DenseNet and CTC to obtain training image data;

training the training image data to obtain an equipment nameplate recognition algorithm model of the electric power specialty;

inputting real image data to test the model, collecting the identification accuracy of the model, and optimizing the model;

and fifthly, carrying out live deployment debugging on the model and releasing the model.

2. The electric power professional device nameplate identification algorithm of claim 1 wherein in step one, the nameplate image data includes a background picture of the usage scene, type and size of the identification font, and special characters.

3. The electric power professional device nameplate identification algorithm of claim 1 wherein in step two, the text detection of the nameplate image data is performed using fast-RCNN and BiLSTM.

4. The electric power professional device nameplate identification algorithm of claim 1 wherein in step two, a densinet and a CTC are used for text recognition of nameplate image data.

Technical Field

The invention relates to the field of intelligent identification, in particular to an equipment nameplate identification algorithm for the electric power specialty.

Background

With the rapid development of economy, electric power construction projects are increasing day by day, and electric power operation relates to a lot of fields, and the situation of safe production is still severe. The problems of heavy operation tasks, tight time, multiple dangerous sources and wide operation range generally exist in an operation field due to the fact that electric power such as electric power production, construction and overhaul and the like exist, the traditional mode completely depends on manual supervision, the efficiency is low, the management means is single, the whole-process and all-around management is difficult to achieve, and management holes easily exist. The establishment of an intelligent identification method for electric power field operation is an important means for improving the safety supervision of various operation fields and ensuring the safe production of electric power.

The text in the natural picture is generally called scene text (scene text), and the scene text detection recognition is the continuation and upgrade of the traditional OCR on the natural picture. At present, in the field operation of electric power, the identification of character information in pictures is realized by electronic identification means such as NFC electronic tag identification, radio frequency identification, bar code identification and the like. According to the electronic identification technology, a database is established for the power equipment by a worker, then the power equipment is coded, the early-stage workload is huge, and the maintenance of the database consumes manpower. Therefore, aiming at the defects in the prior art, the invention provides an electric power professional equipment nameplate recognition algorithm, which is necessary to overcome the defects in the prior art.

Disclosure of Invention

The invention aims to avoid the defects of the prior art and provides an electric power professional equipment nameplate recognition algorithm which has the characteristics that a complex recognition system is not required to be established for equipment, the quantity of algorithm models is small, and the character recognition is accurate.

The above object of the present invention is achieved by the following technical measures.

The method provides an electric power professional equipment nameplate recognition algorithm, and comprises the following steps:

acquiring data of a nameplate image;

secondly, carrying out text detection and recognition on nameplate image data by using a neural network framework consisting of CTPN, DenseNet and CTC to obtain training image data;

training the training image data to obtain an equipment nameplate recognition algorithm model of the electric power specialty;

inputting real image data to test the model, collecting the identification accuracy of the model, and optimizing the model;

and fifthly, carrying out live deployment debugging on the model and releasing the model.

Specifically, in the first step, the nameplate image data includes a background image of the use scene, the type and size of the identification font, and special characters.

Specifically, in the second step, the fast-RCNN and the BilSTM are adopted to carry out text detection on the nameplate image data.

Specifically, in the step two, the DenseNet and the CTC are adopted to perform text recognition on the nameplate image data.

The electric power professional equipment nameplate identification algorithm is characterized in that a model frame is built through a plurality of neural networks, text detection is carried out on nameplate image data through fast-RCNN and BilSTM, and text identification is carried out on the nameplate image data through DenseNet and CTC. And obtaining a recognition algorithm model after data training of the equipment nameplate image in the power field operation. The recognition algorithm has high recognition rate on horizontal text information and can eliminate the interference of special symbols. The recognition algorithm can be used for detecting and recognizing black-matrix white characters, white-matrix black characters, white-matrix red characters and yellow-matrix black characters, and the type of the characters is the best effect when the characters are black bodies, Song dynasty bodies and Microsoft elegant black.

Drawings

The invention is further illustrated by means of the attached drawings, the content of which is not in any way limiting.

FIG. 1 is a technical roadmap for the present invention.

Detailed Description

The invention is further illustrated by the following examples.

An electric power professional equipment nameplate recognition algorithm comprises the following steps:

acquiring data of a nameplate image;

secondly, carrying out text detection and recognition on nameplate image data by using a neural network framework consisting of CTPN, DenseNet and CTC to obtain training image data;

training the training image data to obtain an equipment nameplate recognition algorithm model of the electric power specialty;

inputting real image data to test the model, collecting the identification accuracy of the model, and optimizing the model;

and fifthly, carrying out live deployment debugging on the model and releasing the model.

The nameplate image data in the first step comprises a background picture of the using scene, the type and the size of the identification font, special characters and the like. In the embodiment, the equipment nameplate of the pine-building substation is selected as a training material. The recognition algorithm of the present embodiment basically supports recognition of Chinese and English letters and parts of special symbols. The special data training is specially carried out on some special symbols in the power related field, such as omega, partial italic characters, fancy fonts and the like. Unlike signboard recognition, nameplate recognition is more disturbing, and for this reason the recognition algorithm is trained specifically on parts of special symbols, such as "), +, \\\, |! The punctuation marks of @, #, $,%, &, # and (", etc., and the photos with such special marks were also selected for training.

Each symbol to be trained provides more than 100 samples. In order to enable the algorithm to have a more robust characteristic, the training background image is required to be rich in form and fit with an actual application scene. The inventors have collected and pre-processed training images of different colors and different sizes. The inventors also collected pictures of real scenes for use as model training data.

And step two, performing text detection on the nameplate image data by adopting fast-RCNN and BilSTM, and performing text recognition on the nameplate image data by adopting DenseNet and CTC. The mainstream deep learning text detection and identification technology at present is tesseract, CTPN + DenseNet + CTC and CRNN + CTC. The CTPN + DenseNet + CTC neural network framework used in this embodiment actually uses fast-RCNN + bilst to perform text detection on samples, and then uses DenseNet + CTC to perform text recognition on samples, so that the speed is slightly slow. The algorithm model of the embodiment adopts a plurality of neural networks in combination, and in order to improve the calculation speed of the algorithm, the GPU can be adopted to process image data. During training, training data can be stored in the solid state disk, and training speed can be improved.

The electric power professional equipment nameplate identification algorithm is characterized in that a model frame is built through a plurality of neural networks, text detection is carried out on nameplate image data through fast-RCNN and BilSTM, and text identification is carried out on the nameplate image data through DenseNet and CTC. And obtaining a recognition algorithm model after data training of the equipment nameplate image in the power field operation. The recognition algorithm has high recognition rate on horizontal text information and can eliminate the interference of special symbols. The recognition algorithm can be used for detecting and recognizing black-matrix white characters, white-matrix black characters, white-matrix red characters and yellow-matrix black characters, and the type of the characters is the best effect when the characters are black bodies, Song dynasty bodies and Microsoft elegant black. A quick and accurate identification algorithm is provided for intelligent identification of electric field operation, and smooth development of electric field operation is guaranteed.

Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting the protection scope of the present invention, and although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.