CN114187600A - An Auxiliary System for Intelligent Management of Measuring Assets - Google Patents

Abstract

An auxiliary system for intelligent management of metering assets relates to the technical field of OCR identification. →Feature extraction and dimensionality reduction →Classifier design, training and actual recognition →OCR recognition post-processing →Output results. The beneficial effects of the present invention are: the OCR technology allows everyone to reduce equipment configuration, reduce labor costs, and improve work efficiency.

Description

Auxiliary system for intelligent management of metering assets

Technical Field

The invention relates to the technical field of OCR (optical character recognition), in particular to an auxiliary system for intelligent management of metering assets.

Background

The weight of the transformer, the position of the nameplate and the installation space limit the way of looking over the nameplate information, and in the prior working mode, when the transformer of a user is received and handled in the early stage, when the staff need to make a consignment checking order, transact the procedures of warehousing and ex-warehouse, record and maintain the paper, manually input the information of the electronic version of the mutual inductor, and check, accept, meter installation and power supply at the later stage, in order to match the field mutual inductor with the early-stage transformer for inspection, the mutual inductor information is required to be checked and recorded, and the system can be communicated with a worker handling the procedures of receiving and sending the mutual inductor by telephone to confirm whether the mutual inductor is matched or not and whether a certificate is verified or not, the work is complicated, the office must be supported by the information of the worker, and the electricity stealing prevention is realized during the work of checking the power mutual inductor of the electricity consumer, the paper data which are difficult to store and query are required to be checked for many years ago; the tool for intelligently managing and metering assets solves the problems of difficulty in checking mutual inductor information and data in work, long waiting time of a client in the process of checking mutual inductor information, acceptance and power supply user mutual inductor information and the like of a user.

Disclosure of Invention

The invention provides an auxiliary system for intelligent management of metering assets, which is characterized in that: the operation steps are beginning → login interface → input account password → check information → correctly enter the system → upload photos → OCR character recognition → recognition preprocessing → feature extraction and dimension reduction → classifier design, training and actual recognition → OCR recognition post-processing → output result; and if the input account password is incorrect, returning to the login interface.

The identification pretreatment is graying, and the color image is subjected to the substeps of noise reduction, binarization, character segmentation and normalization; after binarization, the image only has two colors, namely black and white, wherein one color is an image background, and the other color is a character to be identified; the noise reduction algorithm influences the feature extraction, characters in an image are divided into single characters by character segmentation, the characters are recognized word by word during recognition, if lines of the characters are inclined, inclination correction is usually carried out, normalization is carried out to regulate single character images to the same size, and a unified algorithm is applied under the same specification.

The characteristic extraction and dimension reduction: the characters are information for identifying characters, and each different character can be distinguished from other characters through the characteristics; the feature extraction of the numbers and the English letters is easy, because the numbers are only 10, the English letters are only 52, and all the letters are small character sets, the feature extraction is difficult for Chinese characters, the Chinese characters are large character sets, and 3755 first-level Chinese characters which are most commonly used in the national standard and Chinese light are used; the Chinese character has a complex structure and many characters with similar shapes, after determining which kind of characteristics are used, the characteristics are subjected to characteristic dimension reduction, the characteristics are represented by a vector, the dimension is the component number of the vector, if the dimension of the characteristics is too high, the efficiency of the classifier can be greatly influenced, the dimension reduction needs to reduce the dimension, and the characteristic vector after the dimension reduction also reserves enough information content to distinguish different characters.

The classifier design, training and actual recognition: the classifier is used for identification, such as feature extraction and dimension reduction, for a character image, features are extracted and sent to the classifier, the classifier classifies the character image, which character the features are identified into is determined, and before actual identification, the classifier is trained to supervise learning cases.

The OCR recognition post-processing: the classification result is optimized by post-processing, the classification of the classifier occasionally makes mistakes, such as the recognition of Chinese characters, because of the existence of the shape-close characters in the Chinese characters, a character is easily recognized as the shape-close character, and the problem can be solved in the post-processing, such as the correction is carried out through a language model, if the classifier recognizes the 'where' to store ', the' where 'to store' is found to be wrong through the language model, and then the correction is carried out; the OCR recognition image has a large amount of characters which have complex conditions of typesetting, font size and the like, and the recognition result is formatted in the post-processing and output according to the typesetting arrangement in the image.

The uploading photo is an uploading mutual inductor nameplate photo.

The core of the invention is to use artificial intelligence OCR (character recognition) technology, which refers to the process of analyzing and recognizing the image file of the text data to obtain characters and layout information, and also recognizes the characters in the image and returns the characters in the form of text.

Through the accurate mutual-inductor data plate information of different models of different producers of discernment of using this technique for the mutual-inductor data plate information direct conversion who shoots the scanning can edit the characters type and record in this instrument, exports the form of selected data with the Excel form simultaneously, has not only solved the problem of the information that the operational environment caused under the complicacy and check the difficulty, record check consuming time of a specified duration through the scanning of shooing.

The invention has the beneficial effects that: through the development of the invention, the papery record and the secondary manual electronic record when the former customer inspects the power transformer are changed into the electronic record which is changed along with the scanning and the checking, and the field acceptance inspection of the power transformer, the power supply operation inspection of the power transformer, the anti-electricity-stealing inspection of the power transformer and the later maintenance of the power transformer are changed into the working mode that the user information can be mastered along with the scanning and the checking in order to take out the mobile phone which is carried along. From slowly reading to grasp at any time, from original manual work to present AI intelligence, what improve is not only work efficiency, operating mass, has changed the working method from the root, has also saved the time of handling procedures, waiting to complete the acceptance, power supply for the customer simultaneously, faster, better for user service.

OCR technology has let everybody reduce equipment configuration, has reduced the human cost, has improved work efficiency.

Description of the drawings:

FIG. 1 is a block diagram of the process of the present invention.

Detailed Description

Embodiment 1, an auxiliary system for intelligent management of a metered asset, characterized in that: the operation steps are beginning → login interface → input account password → check information → correctly enter the system → upload photos → OCR character recognition → recognition preprocessing → feature extraction and dimension reduction → classifier design, training and actual recognition → OCR recognition post-processing → output result; and if the input account password is incorrect, returning to the login interface.

The identification pretreatment is graying, and the color image is subjected to the substeps of noise reduction, binarization, character segmentation and normalization; after binarization, the image only has two colors, namely black and white, wherein one color is an image background, and the other color is a character to be identified; the noise reduction algorithm influences the feature extraction, characters in an image are divided into single characters by character segmentation, the characters are recognized word by word during recognition, if lines of the characters are inclined, inclination correction is usually carried out, normalization is carried out to regulate single character images to the same size, and a unified algorithm is applied under the same specification.

The characteristic extraction and dimension reduction: the characters are information for identifying characters, and each different character can be distinguished from other characters through the characteristics; the feature extraction of the numbers and the English letters is easy, because the numbers are only 10, the English letters are only 52, and all the letters are small character sets, the feature extraction is difficult for Chinese characters, the Chinese characters are large character sets, and 3755 first-level Chinese characters which are most commonly used in the national standard and Chinese light are used; the Chinese character has a complex structure and many characters with similar shapes, after determining which kind of characteristics are used, the characteristics are subjected to characteristic dimension reduction, the characteristics are represented by a vector, the dimension is the component number of the vector, if the dimension of the characteristics is too high, the efficiency of the classifier can be greatly influenced, the dimension reduction needs to reduce the dimension, and the characteristic vector after the dimension reduction also reserves enough information content to distinguish different characters.

The classifier design, training and actual recognition: the classifier is used for identification, such as feature extraction and dimension reduction, for a character image, features are extracted and sent to the classifier, the classifier classifies the character image, which character the features are identified into is determined, and before actual identification, the classifier is trained to supervise learning cases.

The OCR recognition post-processing: the classification result is optimized by post-processing, the classification of the classifier occasionally makes mistakes, such as the recognition of Chinese characters, because of the existence of the shape-close characters in the Chinese characters, a character is easily recognized as the shape-close character, and the problem can be solved in the post-processing, such as the correction is carried out through a language model, if the classifier recognizes the 'where' to store ', the' where 'to store' is found to be wrong through the language model, and then the correction is carried out; the OCR recognition image has a large amount of characters which have complex conditions of typesetting, font size and the like, and the recognition result is formatted in the post-processing and output according to the typesetting arrangement in the image.

Example 2; the utility model provides an auxiliary system of intelligent management of measurement mutual-inductor which characterized in that: the operation steps are beginning → login interface → input account password → check information → correct entering system → upload mutual inductor information card photo → OCR character recognition → recognition preprocessing → feature extraction and dimension reduction → classifier design, training and actual recognition → OCR recognition post-processing → output result; and if the input account password is incorrect, returning to the login interface.

Claims (6)

1. An auxiliary system for intelligent management of metering assets, characterized in that: the operation steps are: start → login interface → input account password → check information → enter the system correctly → upload photos → OCR character recognition → recognition preprocessing → feature extraction and Dimension reduction → classifier design, training and actual recognition → OCR recognition post-processing → output results; if the input account password is incorrect, return to the login interface. 2 . The auxiliary system for intelligent management of metering assets according to claim 1 , wherein the identification preprocessing is grayscale, and the color image is processed by noise reduction, binarization, character segmentation and normalization. 3 . These sub-steps are normalized; after binarization, there are only two colors left in the image, namely black and white, one of which is the image background, and the other color is the text to be recognized; the noise reduction algorithm affects the feature extraction, and the character cuts Divide the text in the image into a single text, and recognize it one word at a time. If the text line is skewed, it is often necessary to perform skew correction, and normalize the single text image to the same size. Under the same specification, a unified algorithm is applied. 3. The auxiliary system for the intelligent management of measuring assets according to claim 1, wherein: the feature extraction and dimensionality reduction: the feature is the information used to identify characters, and each different character can pass the feature To distinguish it from other characters; it is easier to extract features for numbers and English letters, because there are only 10 numbers and only 52 English letters, all of which are small character sets. For Chinese characters, feature extraction is difficult, and Chinese characters are large characters Set, there are 3755 first-level Chinese characters in the national standard alone; Chinese characters are complex in structure, and there are many similar characters. After determining which feature to use, the feature dimension reduction is carried out. The feature is represented by a vector, and the dimension is The number of components of the vector, if the dimension of the feature is too high, the efficiency of the classifier will be greatly affected. Dimension reduction not only reduces the dimension, but also retains enough information in the feature vector after reducing the dimension. to distinguish between different texts. 4. The auxiliary system for intelligent management of metering assets according to claim 1, characterized in that: the classifier is designed, trained and actually identified: the classifier is used for identification, such as feature extraction and dimensionality reduction , for a text image, extract the features to the classifier, the classifier will classify it, and determine which text the feature should be recognized as. Before the actual recognition, the classifier should be trained to supervise the learning case. 5. The auxiliary system for intelligent management of metering assets according to claim 1, characterized in that: the post-processing of the OCR recognition: the post-processing optimizes the classification results, and the classification of the classifier occasionally makes mistakes, such as for Chinese characters. Because of the existence of close characters in Chinese characters, it is easy to recognize a word as its close characters, and this problem can be solved in post-processing, such as correction through language model - if the classifier will "where" If it is recognized as "where to store", the language model will find that "where to store" is wrong, and then correct it; the OCR recognition image will have a lot of text, and these texts have complex situations such as typesetting and font size, and the recognition results will be checked in post-processing. Format the output according to the layout in the image. 6 . The auxiliary system for intelligent management of metering assets according to claim 1 , wherein the uploaded photo is an uploaded photo of the nameplate of the transformer. 7 .

Images