[download app for free]ineukernel OCR image data recognition and acquisition principle and product application

Objective        record

1..... Application Overview ... 2

2..... Free download trial ... 2

3..... Video introduction ... 2

4..... iNeuLink.Ocr Image data acquisition application ... 2

5..... Data upload to iNeuOS Industrial Internet operating system ... 4

6..... Ocr Basic concepts ... 7

1.   Application Overview

　　 In the industrial 、 Military industry or aerospace and other fields , Some equipment and software systems are relatively old , But the cost of replacement is relatively high , Data cannot be completely recorded in the process of experiment or production , It has caused great obstacles to data application and analysis , It can't solve the labor intensity of experiment and production personnel .

　　 adopt OCR Image recognition technology can well solve the above problems , But there is no complete production software to realize operation level applications . Even though OCR The technology is relatively mature , Except for specific application scenarios , The popularity rate is not high .

　　iNeuKernel.Ocr Image data acquisition software , Realized Get screenshots or application screenshots automatically in real time 、 Dynamically add data identification label information 、 Background tasks automatically slice and collect data and forward data in real time , And iNeuOS Industrial Internet operating system interconnection , Complete data collection 、 transmission 、 Storage 、 View modeling 、 Analysis and other application processes . The diagram is as follows ：

2.   Free download trial

　　 link ：https://pan.baidu.com/s/1joGfBefaBKiFJ1l08N3KZg

　　 Extraction code ：v242

3.   Video introduction

4.   iNeuLink.Ocr Image data acquisition application

（1） Main interface of operation , The application is relatively simple . Here's the picture ：

  （2） Configure image source , You can use shortcut keys , Full screen screenshot or program screenshot , In order to locate the relative position of the coordinates of the data label , Data extraction . Here's the picture ：

  （3） Add or edit identification tags , single click 【 Data identification 】 You can use red boxes to mark other data location information on the map source , stay 【 Data identification results 】 You can see the result of identifying data , And fill in the data label name . Here's the picture ：

  （4） Forward data , The identified data results can be forwarded to other systems , Fill in the station number 、 long-range IP、 long-range Port Forwarding cycle and whether it is enabled . Here's the picture :

5.   Data upload to iNeuOS Industrial Internet operating system

　　 download iNeuOS Offline installer , For the installation process, see ： One key deployment . After entering the system , Select desktop 【 Device model 】, Click the small plus sign Icon in the upper left corner , Add a new device . choice 【 Service instance 】 hinder 【 edit 】 Button link , To configure 【 Basic settings 】 and 【Socket】,【Socket】 The default listening port is 6699, Other information generally does not need to be modified , Here's the picture ：

     Choose the configured 【 Service instance 】, And configure the communication type 、 Application protocol and selection driver , The configuration is as follows ：

     Select the rotation icon in the upper left corner , Will restart the background service , Apply the currently configured information . Hardware gateway and iNeuOS When the platform communicates for the first time, it will synchronize the equipment and data point information of the gateway , In the current iNeuOS Generate sub devices driven by devices , There will be... On the sub device iNeuKernel Data point information of hardware gateway . Here's the picture ：

 　　 iNeuOS Industrial Internet trial ： Trial address

6.   Ocr Basic concepts

（1）OCR Application scenarios of

　　 according to OCR In terms of application scenarios , We can roughly divide it into identifying special objects in specific scenes OCR As well as identifying common features in a variety of scenarios OCR. As far as the former is concerned , Certificate recognition and license plate recognition are special OCR The typical case . Design for specific scenes 、 Optimize to achieve the best effect display in specific scenes . That's universal OCR Is used in more 、 In more complex situations , Have better universality . In this process, due to the uncertainty of the scene , such as ： The background of the picture is extremely rich 、 Uneven brightness 、 Uneven illumination 、 Incomplete occlusion 、 Text distortion 、 Various fonts and so on , It will bring great challenges .

（2）OCR Technical route of

　　 Typical OCR The technical route is shown in the figure below ：

　　 among OCR The key path of recognition is text detection and text recognition , This is also where deep learning technology can give full play to its effectiveness .PaddleHub The network structure of the open source pre training model is Differentiable Binarization+ CRNN, be based on icdar2015 Training under the data set .

　　 First ,DB It is a text detection algorithm based on segmentation . Among various text detection algorithms , The segmentation based detection algorithm can better deal with irregular shaped text such as bending , Therefore, it can often achieve better detection results . However, the process of transforming the segmentation result into detection frame in the post-processing step of segmentation method is complex , Time consuming . Therefore, the author proposes a differentiable binary module （Differentiable Binarization, abbreviation DB）, The binary threshold is added to the training to learn , More accurate detection boundary can be obtained , So as to simplify the post-processing process .DB The algorithm ends in 5 On data sets state-of-art The effect and performance of . Reference paper ：Real-time Scene Text Detection with Differentiable Binarization

　　 The picture below is DB Structure diagram of the algorithm ：

　　 next , We use CRNN（Convolutional Recurrent Neural Network） Convolution recurrent neural network , yes DCNN and RNN The combination of , It is specially used to identify sequential objects in images . And CTC loss In combination with , Do character recognition , You can learn directly from text word level or line level annotation , No detailed character level annotation is required . Reference paper :An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition

　　 The picture below is CRNN Network structure diagram ：

---

The Internet of things & Big data technology QQ Group ：54256083

The Internet of things & Big data projects QQ Group ：727664080
QQ：504547114
WeChat ：wxzz0151
Blog ：https://www.cnblogs.com/lsjwq
WeChat official account ：iNeuOS