Image information recognition method and device, storage medium, and computer equipment

Abstract

The present invention discloses a method and apparatus for recognizing image information, a storage medium, and a computer device, relating to the field of data processing technology. The method primarily aims to address the low efficiency of image information recognition in existing medical invoices. The method comprises: obtaining image information of a medical invoice from different block nodes of a blockchain network; extracting text data based on the image information, and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained for feature extraction based on principal component analysis; determining a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; and determining the medical term as the recognition result of the image information when the quantization parameter meets a preset quantization range. The method is primarily used for recognizing image information.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for recognizing image information, a storage medium, and a computer device. Background Art

[0002] As medical services gradually improve, medical invoices play an important role in reimbursement for different needs. Therefore, the auditor needs to verify and process a large number of medical invoices. In order to reduce the waste of manpower and material resources, medical invoices will be entered into the computer for identification, so that the identification results can be directly entered into the system for verification.

[0003] The inventors of the present application discovered during their research that, in the existing invoice recognition process, since paper invoices need to be scanned, the image information of the medical invoice obtained is usually blurry due to operational defects, paper factors, etc., and when the image information is recognized using text recognition, the text content in the image information cannot be accurately recognized. Moreover, due to the particularity of the text content in medical invoices, direct natural language processing using the Chinese character library cannot meet the recognition requirements of the medical content in the medical invoices, thereby greatly reducing the recognition efficiency of the image information in the medical invoices. Summary of the Invention

[0004] In view of this, the present invention provides a method and device for recognizing image information, a storage medium, and a computer device, the main purpose of which is to solve the problem of low efficiency in recognizing image information in existing medical invoices.

[0005] According to one aspect of the present invention, a method for recognizing image information is provided, comprising:

[0006] Obtain image information of medical invoices from different block nodes of the blockchain network;

[0007] Extracting text data based on the image information, and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis;

[0008] Determining a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information;

[0009] When the quantization parameter meets a preset quantization range, the medical term is determined as a recognition result of the image information.

[0010] Furthermore, before performing recognition processing on the text data based on the trained recognition model, the method further includes:

[0011] Acquire a training sample set from the preset text resource library containing at least one medical term;

[0012] Performing word vector conversion on the training samples in the training sample set according to a word vector conversion algorithm to obtain a vector matrix;

[0013] extracting eigenvectors matching the vector matrix from the training samples based on principal component analysis;

[0014] The convolutional neural network is trained based on the feature vector to obtain a trained recognition model.

[0015] Furthermore, the quantitative parameters include edit distance and similarity, and the quantitative parameters between the recognized text data and at least one corresponding medical term in the preset text resource library include:

[0016] Calculating the similarity between the recognized text data and at least one corresponding medical term in a preset text resource library;

[0017] When the calculated similarity is greater than a preset similarity threshold, calculating the edit distance between the text data and the medical term;

[0018] When the calculated similarity is less than or equal to a preset similarity threshold, the quantization parameter is configured to be empty.

[0019] Furthermore, extracting text data based on the image information includes:

[0020] Recognizing text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier;

[0021] Performing a mark replacement process on the text data carrying the mark, wherein the mark replacement process is used to replace the mark with text content that matches the text data;

[0022] The text data that has completed the replacement process is converted into word vectors for recognition processing.

[0023] Furthermore, before performing the mark replacement process on the text data carrying the mark, the method further includes:

[0024] Establishing a replacement mapping relationship between different text contents based on the preset text resource library, wherein the replacement mapping relationship includes a mapping relationship between the preceding text content and the succeeding text content of the text content to be replaced;

[0025] The process of performing mark replacement on the text data carrying the mark includes:

[0026] Compare the preceding and following text contents of each identifier in the text data carrying the mark in turn, determine the replacement text content that matches the preceding and following text contents according to the replacement mapping relationship, and replace the determined replacement text content with the mark to obtain the replaced text content.

[0027] Furthermore, before using the image recognition technology OCR to recognize the text data in the image information in a row-by-row iterative manner, the method further includes:

[0028] Acquire a straight line in the image information, and determine a difference between an inclination angle of the straight line and a reference angle of the image information;

[0029] The image information is rotated using the difference.

[0030] Furthermore, the method further comprises:

[0031] When the quantization parameter does not meet the preset quantization range, generating processing request information including the image information and the text data;

[0032] Sending the processing request information to the block node based on the public key of the blockchain network to request recognition of the processing request information;

[0033] Receive the recognition result of the processing request information, and update the recognition result to the training sample set of the recognition model.

[0034] According to another aspect of the present invention, there is provided an apparatus for recognizing image information, comprising:

[0035] The acquisition module is used to obtain the image information of medical invoices from different block nodes of the blockchain network;

[0036] a processing module, configured to extract text data based on the image information and perform recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis;

[0037] A first determination module is configured to determine a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information;

[0038] The second determining module is configured to determine the medical term as a recognition result of the image information when the quantization parameter meets a preset quantization range.

[0039] Furthermore, the device further comprises: a conversion module, an extraction module, a training module,

[0040] The acquisition module is further configured to acquire a training sample set from the preset text resource library containing at least one medical term;

[0041] The conversion module is used to perform word vector conversion on the training samples in the training sample set according to the word vector conversion algorithm to obtain a vector matrix;

[0042] The extraction module is configured to extract eigenvectors matching the vector matrix from the training samples based on principal component analysis;

[0043] The training module is used to train the convolutional neural network based on the feature vector to obtain a trained recognition model.

[0044] Furthermore, the quantization parameters include edit distance and similarity, and the first determination module includes:

[0045] a calculation unit, configured to calculate a similarity between the recognized text data and at least one corresponding medical term in a preset text resource library;

[0046] a determining unit, configured to calculate an edit distance between the text data and the medical term when the calculated similarity is greater than a preset similarity threshold;

[0047] The configuration unit is configured to configure the quantization parameter to be empty when the calculated similarity is less than or equal to a preset similarity threshold.

[0048] Furthermore, the processing module includes:

[0049] a recognition unit, configured to recognize text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier;

[0050] a first processing unit, configured to perform a mark replacement process on the text data carrying the mark, wherein the mark replacement process is configured to replace the mark with text content matching the text data;

[0051] The conversion unit is used to convert the text data that has completed the replacement processing into a word vector for recognition processing.

[0052] Furthermore, the processing module further includes: an establishing unit,

[0053] The establishing unit is configured to establish a replacement mapping relationship between different text contents based on the preset text resource library, wherein the replacement mapping relationship includes a mapping relationship between the preceding text content and the succeeding text content of the text content to be replaced;

[0054] The first processing unit is specifically used to compare the preceding text content and following text content of each identifier in the text data carrying the mark in turn, determine the replacement text content that matches the preceding text content and following text content according to the replacement mapping relationship, and replace the determined replacement text content with the mark to obtain the text content after replacement processing.

[0055] Furthermore, the processing module further includes:

[0056] an acquisition unit, configured to acquire a straight line in the image information and determine a difference between an inclination angle of the straight line and a reference angle of the image information;

[0057] The second processing unit is configured to perform rotation processing on the image information using the difference.

[0058] Furthermore, the device further comprises:

[0059] A generating module, configured to generate processing request information including the image information and the text data when the quantization parameter does not meet a preset quantization range;

[0060] A sending module, configured to send the processing request information to the block node based on the public key of the blockchain network to request recognition of the processing request information;

[0061] A receiving module is used to receive the recognition result of the processing request information and update the recognition result to the training sample set of the recognition model.

[0062] According to another aspect of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned image information recognition method.

[0063] According to another aspect of the present invention, there is provided a computer device comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0064] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the above-mentioned image information recognition method.

[0065] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0066] The present invention provides a method and apparatus for recognizing image information, a storage medium, and a computer device. An embodiment of the present invention obtains image information of a medical invoice from different block nodes of a blockchain network; extracts text data based on the image information, and performs recognition processing on the text data based on a trained recognition model, wherein the recognition model is obtained by feature extraction training based on principal component analysis; determines a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; when the quantization parameter meets a preset quantization range, determines the medical term as the recognition result of the image information, thereby achieving accurate recognition of medical content based on a combination of machine learning and recognition correction, greatly improving the depth of recognition, and thus meeting the requirements for efficient recognition of medical content in medical invoices.

[0067] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0069] Figure 1 A flow chart of a method for recognizing image information provided by an embodiment of the present invention is shown;

[0070] Figure 2 A flow chart of another method for recognizing image information provided by an embodiment of the present invention is shown;

[0071] Figure 3 A block diagram showing the composition of an image information recognition device provided by an embodiment of the present invention is shown;

[0072] Figure 4 A block diagram showing another image information recognition device provided by an embodiment of the present invention is shown;

[0073] Figure 5 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0074] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0075] In order to solve the problem of low efficiency in recognizing image information in existing medical invoices and accurately identify the specific text content of the medical invoice from the image information obtained by scanning the medical invoice, an embodiment of the present invention proposes a method for recognizing image information for identifying the content of medical invoices in digital medicine.

[0076] The embodiment of the present invention provides a method for recognizing image information. Figure 1 As shown, the method includes:

[0077] 101. Obtain image information of medical invoices from different block nodes of the blockchain network.

[0078] The image information of the medical invoice is stored in the block node by scanning the paper medical invoice in the digital medical system. When the image information is recognized, the image information is obtained from the block node.

[0079] In addition, since each hospital needs to keep its financial information confidential, that is, the image information obtained by scanning medical invoices confidential, in order to ensure the security of identifying different image information, when scanning and collecting image information, the image information of different hospitals is stored in different block nodes in the blockchain network, and a secret key is issued for the server to perform identification processing. In this embodiment of the present invention, the key includes a public key and a private key. Each block node corresponds to a block node identifier. Each hospital is assigned a private key and corresponds to a block node identifier to represent its identity. The hospital can obtain the image information in the block node corresponding to the block node identifier based on the assigned private key. The public key is used to decrypt the image information identified in the block node corresponding to the hospital, thereby ensuring the security and privacy of medical invoices and improving the efficiency of image information recognition.

[0080] 102. Extract text data based on the image information, and perform recognition processing on the text data based on a trained recognition model.

[0081] In an embodiment of the present invention, in order to recognize image information, it is first necessary to extract text data from the image information. Since medical invoices are invoices from various hospitals, the text data extracted from the image information includes text data such as the name and quantity of the drug, and the name of the hospital, for example, hydrochloric acid solution, acid solution, First Municipal Hospital, Second Municipal Hospital, etc., which are not specifically limited in this embodiment of the present invention. In addition, the recognition model is used to recognize text data to determine the specific text content. The recognition model is a machine learning model, such as a neural network algorithm, which is not specifically limited in this embodiment of the present invention. The recognition model is obtained by feature extraction training based on principal component analysis, so that features with high recognition efficiency are extracted from multiple training data to be trained based on principal component analysis with efficient feature extraction, and the recognition model is trained to recognize and process the text data based on the trained recognition model. This is not specifically limited in this embodiment of the present invention.

[0082] 103. Determine a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library.

[0083] The preset text resource library is created in the block node that matches the image information, and the medical terms are hospital names, drug names, and other contents that have been determined by different hospitals. The quantitative parameters include edit distance and similarity. Specifically, the edit distance and similarity are calculated based on the recognized text data and resource medical terms. For example, the edit distance and similarity between the recognized "hydrochloric acid solution" and the corresponding "hydrochloric acid alkali solution", "acetic acid solution", "carbonic acid solution", and "hydrochloric acid solution" in the preset text resource library are calculated.

[0084] It's important to note that in natural language processing, the edit distance is a quantitative measure of the difference between two strings, specifically the number of times one string is processed to transform it into another. Similarity is the vector distance between one string and another. Therefore, determining quantitative parameters that include the edit distance and similarity can improve the accuracy of text recognition in image information.

[0085] 104. When the quantization parameter meets a preset quantization range, determine the medical term as a recognition result of the image information.

[0086] In this embodiment of the present invention, when the quantization parameter falls within a preset quantization range, it indicates that the text data matches the corresponding medical term in the preset text resource library. Therefore, the medical term being compared is used as the recognition result of the image information, resulting in a highly accurate recognition result. For example, if the edit distance between the recognized text data "hydrochloric acid solution" and the medical term "hydrochloric acid solution" in the quantization parameter is less than 0.3 and the similarity is greater than 0.8, the medical term "hydrochloric acid solution" is used as the recognition result.

[0087] The present invention provides a method for recognizing image information. An embodiment of the present invention obtains image information of a medical invoice from different block nodes of a blockchain network; extracts text data based on the image information, and performs recognition processing on the text data based on a trained recognition model, wherein the recognition model is obtained by feature extraction training based on principal component analysis; determines a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; when the quantization parameter meets a preset quantization range, determines the medical term as the recognition result of the image information, thereby achieving accurate recognition of medical content based on a combination of machine learning and recognition correction, greatly improving the depth of recognition, and thus meeting the requirements for efficient recognition of medical content in medical invoices.

[0088] The embodiment of the present invention provides another method for recognizing image information. Figure 2 As shown, the method includes:

[0089] 201. Obtain a training sample set from the preset text resource library containing at least one medical term.

[0090] In an embodiment of the present invention, in order to meet the need for recognizing image information, a recognition model is used to recognize text data, which requires pre-training of the recognition model. Therefore, in order to train a recognition model with high recognition efficiency, a training sample set in a preset text resource library is obtained. The preset text resource library is created in the block node that matches the image information and records at least one medical term. The medical term is a term such as a hospital name or a drug name that has been determined by different hospitals, and the embodiment of the present invention does not specifically limit this.

[0091] It should be noted that in order to ensure privacy, when constructing the training set, the blockchain structure is used to obtain the text data of drug names, hospital names, etc. that can be used as training data from different block nodes of the blockchain network. That is, each hospital stores a corresponding preset text resource library in the block node with its own private key. When the recognition model is trained, the training sample set is obtained from all the preset text resource libraries to obtain the data to be trained.

[0092] 202. Perform word vector conversion on the training samples in the training sample set according to a word vector conversion algorithm to obtain a vector matrix.

[0093] In an embodiment of the present invention, since the recognition model is a machine learning model as a mathematical algorithm, and the training samples in the sample set are medical terms in text form, during training, in order to meet the quantitative requirements for model training, the training samples in the training sample set are converted into word vectors using a sub-vector conversion algorithm to obtain a vector matrix for model training. Among them, the word2vec algorithm is used to convert the training samples, such as the text data of the medical name and the text data of the hospital name, into word vectors. Specifically, the conversion process of the word vector can use one-hot encoding to digitize the training samples, that is, to obtain a vector containing only one 1 and the rest are 0 to uniquely represent the word, and then further obtain the word vector of the drug name and the hospital name based on the Word2vec algorithm.

[0094] It should be noted that Word2Vec is a shallow neural network that includes an input layer x, a hidden layer h, and an output layer o. The dimensions of the input layer and the output layer are N (the size of the vocabulary), and the dimension of the hidden layer is D (the dimension of the word vector). The network weight matrix between the input layer and the hidden layer is V(N×D), and the network weight matrix between the input layer and the hidden layer is V'(N×D), h=xV, o=softmax(V'h). The input layer x of the Word2vec algorithm receives the one-hot encoding of word a, and the output layer o calculates the probability of all words appearing in the context of word a, and the probability is calculated using softmax. The Word2vec algorithm is trained in a corpus so that the conditional probability of the word appearing is as consistent as possible with the distribution in the corpus. After training, the weight vector matrix V obtained is the final word vector matrix, and V' is the vector matrix when the word is used as a prediction output, which is not specifically limited in the embodiment of the present invention.

[0095] 203. Extract eigenvectors matching the vector matrix from the training samples based on principal component analysis.

[0096] In the embodiment of the present invention, in order to utilize the recognition model for recognition training and improve the accuracy of model training, feature vectors are extracted from the training samples converted into vector matrices. Specifically, the feature vectors of the vector matrix are extracted based on the principal component analysis method, and the extracted feature vectors are used as input to the convolutional neural network to be trained.

[0097] In the embodiment of the present invention, the specific feature extraction process of the vector matrix by principal component analysis includes: A. Calculating the sample data set {(x1, y1), (x2, y2), (x N ,y N)} is the n-dimensional principal component of the covariance matrix of the sample set, which is the n-dimensional eigenvector matrix W, where x is the data eigenvector value containing N dimensions, and y is the verification result of this data. For each xi, center it. B. Calculate the sample covariance matrix; C. Perform eigenvalue decomposition on the covariance matrix; D. Take the eigenvectors corresponding to the largest n eigenvalues ​​(w1, w2,,, w n ), after standardizing all eigenvectors, we get the eigenvector matrix W; E, transform each xi into feature z i =W T x i , and get the filtered feature set D'={z1,z2,,,z m}.

[0098] 204. Train a convolutional neural network based on the feature vector to obtain a trained recognition model.

[0099] In the embodiment of the present invention, in order to recognize the text data extracted from the image information, a feedforward neural network including convolution calculation and having a deep structure, i.e., a convolutional neural network, is used for training. The training steps include: the first stage: forward propagation stage: a. Take a sample (X, Y) from the sample set p ), input X into the network; b, calculate the corresponding actual output O p In this phase, information is transformed from the input layer to the output layer. This process is also the process performed when the network is running normally after training. In this process, the network performs calculations, that is, the input is multiplied by the weight matrix of each layer to obtain the final output result: p =F n (…(F2(F1(X p W(1))W(2))…)W(n)). The second stage, the backward propagation stage: a. Calculate the actual output O p and the corresponding ideal output Y p The difference between the two groups is obtained; b. Back propagation is used to adjust the weight matrix according to the method of minimizing the error. After completing the entire training process, the improved CNN prediction and recognition model based on principal component analysis is obtained.

[0100] 205. Obtain image information of medical invoices from different block nodes of the blockchain network.

[0101] 206. In a row-by-row iteration manner, utilize optical character recognition (OCR) technology to recognize text data in the image information.

[0102] In an embodiment of the present invention, to meet the requirements for identifying text data in image information, image recognition technology (OCR) is used to identify the text data. Specifically, when performing OCR recognition, recognition is performed based on the content of each line, that is, the text in each line of the image is recognized. During this process, due to factors such as scanning pixels and scanning luminosity, the text in each line may not be clearly displayed in the image, resulting in incomplete recognized text data. For example, the text in the image is "First Central Hospital", but the recognized text data is only "First Hospital", and the two words "Center" are missing, which affects the recognition accuracy of the text data recognition model. Therefore, in an embodiment of the present invention, during the recognition process, if unrecognized text data is detected, the unrecognized text content in the text data is marked, that is, text data containing the mark is obtained. The unrecognized text content in the text data is marked with a preset mark. Different text content can be marked with different preset marks or with the same mark. The preset mark is not specifically limited in the embodiment of the present invention. For example, if "Center" is not recognized, it is marked in the recognized text data, resulting in the marked text data "First ** Hospital".

[0103] Furthermore, for the purpose of specific explanation and limitation, before the text data in the image information is recognized by using the image recognition technology OCR in a row-by-row iterative manner, the method also includes: obtaining a straight line in the image information, and determining the difference between the inclination angle of the straight line and the reference angle of the image information; and rotating the image information using the difference.

[0104] In the embodiment of the present invention, in order to avoid the text data to be recognized in the image information from being unable to be recognized in line form and to meet the requirement of recognizing the predetermined straight line form, it is necessary to perform rotation correction on the tilted image information. Specifically, since the scanned medical invoice in the embodiment of the present invention can be an invoice, and the invoice contains a straight line that can represent the tilt angle of the invoice, the straight line in the image information is obtained and the tilt angle of the straight line is determined so as to be corrected with the reference angle of the reference line for OCR recognition. The difference between the tilt angle of the straight line and the reference angle of the image information is calculated, and the rotation processing is performed based on the difference to complete the correction, thereby improving the recognition efficiency of the OCR recognition technology.

[0105] 207. Perform label replacement processing on the text data carrying the label.

[0106] For the embodiments of the present invention, in order to improve the accuracy of recognition, so that the obtained text data can be accurately matched to the corresponding recognition results, thereby meeting the all-round and high-efficiency recognition of the text data in the image information, the marked text data is subjected to identification replacement processing, that is, the identification is replaced with the text content. Specifically, the identification replacement processing is used to replace the marked text content that matches the text data, that is, the text content is replaced to the identification in the text data. For example, in "The First ** Hospital", the text content "Municipal" is replaced to the "**" place. The embodiments of the present invention do not make specific limitations.

[0107] Further, for further illustration and limitation, in order to improve the accuracy of replacing the text content, thereby effectively improving the recognition of the text data, before the marked text data is subjected to identification replacement processing, the method further includes: establishing a replacement mapping relationship of different text contents based on the preset text resource library.

[0108] In the embodiments of the present invention, in order to accurately determine the text content for replacing the identification for effective replacement, a replacement mapping relationship of different text contents is established based on the preset text resource library. When performing text replacement processing, the text content to be replaced is obtained according to the replacement mapping relationship. Among them, the replacement mapping relationship includes the mapping relationship between the pre-text content and the post-text content of the text content to be replaced. Each text content to be replaced corresponds to a pre-text content and a post-text content, so that when replacing, the replacement is performed based on the pre- and post-text contents of the identification in the text data.

[0109] In the embodiments of the present invention, for further limitation and illustration, the identification replacement processing of the marked text data includes: sequentially comparing the pre-text content and the post-text content of each identification in the marked text data, determining the replacement text content that matches the pre-text content and the post-text content according to the replacement mapping relationship, and replacing the determined replacement text content to the marked place to obtain the text content after replacement processing.

[0110] For example, the replacement mapping relationship is that the pre-text content to be replaced with the text content "liquid" is "dissolved", and the post-text content is "substance". Then for the text data "hydrochloric acid diss*", the "*" is replaced with the text content "liquid", thereby improving the accuracy and effectiveness in the recognition process.

[0111] 208. Perform word vector conversion on the text data after the replacement processing to perform recognition processing.

[0112] 209. Perform recognition processing on the text data based on the trained recognition model.

[0113] Specifically, the trained recognition model is a convolutional neural network model.

[0114] 210. Calculate the similarity between the recognized text data and at least one corresponding medical term in a preset text resource library.

[0115] In the embodiment of the present invention, since the quantization parameters include similarity and edit distance, in order to use the quantization parameters for the final verification of recognition, the similarity between the recognized text data and the medical terms in the preset text resource library is calculated, that is, by the formula Among them, q is the recognition result and d is the medical term.

[0116] 211. When the calculated similarity is greater than a preset similarity threshold, calculate the edit distance between the text data and the medical term.

[0117] In an embodiment of the present invention, in order to improve the correction of the recognition results and thus improve the accuracy of recognition, when the similarity is greater than a preset similarity threshold, it indicates that based on the similarity level, the text data and the medical term are matched. Therefore, the edit distance between the text data and the medical term is further calculated.

[0118] It should be noted that the edit distance is the minimum number of single-character editing operations required to convert one word into another between two words. In the embodiment of the present invention, the text data is vectorized when performing pattern recognition, and the calculation of the edit distance requires determining the number of editing operations between characters. Therefore, when calculating the edit distance, the recognized text data that has been vectorized can be converted into a character string to calculate the edit distance. This is not specifically limited in the embodiment of the present invention.

[0119] 212. When the calculated similarity is less than or equal to a preset similarity threshold, configure the quantization parameter to be empty.

[0120] For the embodiment of the present invention, if the similarity is less than or equal to the preset similarity threshold, it means that the text data does not match the medical term based on the similarity level, and the recognition result obtained is incorrect. The quantization parameter is configured to be empty so that it is directly excluded when the quantization parameter is judged, thereby improving the correction accuracy of the recognition result.

[0121] 213a. When the quantization parameter meets a preset quantization range, determine the medical term as a recognition result of the image information.

[0122] 213b. When the quantization parameter does not meet the preset quantization range, generate processing request information including the image information and the text data.

[0123] For the embodiment of the present invention, in order to meet the needs of text data recognition, for the situation where the quantization parameters do not meet the preset quantization range of text data recognition, processing request information containing image information and text data can be further generated so that it can be processed manually or by a third-party server to achieve diversity in image information recognition.

[0124] The steps following embodiment 213b of the present invention also include: sending the processing request information to the block node based on the public key of the blockchain network.

[0125] In an embodiment of the present invention, since the image information is obtained from the block nodes of each hospital, in order to meet the recognition requirements and confidentiality requirements of the image information, the processing request information is sent to the block node based on the public key of the blockchain network to request identification of the processing request information.

[0126] For example, image information a is obtained from the block node of the First Central Hospital. If the quantization parameters obtained after image information recognition do not meet the preset quantization range, the processing request information corresponding to the image information a is sent to the block node of the First Central Hospital using the public key to request the First Central Hospital to identify it.

[0127] Furthermore, the embodiment of the present invention further includes: receiving a recognition result of the processing request information, and updating the recognition result to a training sample set of the recognition model.

[0128] In an embodiment of the present invention, in order to make the recognition of image information more intelligent, the recognition result of the processing request information is received, and the recognition result is updated to the training sample set so that the recognition model is retrained to obtain a recognition model with more recognition accuracy, thereby improving the recognition accuracy.

[0129] The present invention provides another method for recognizing image information. An embodiment of the present invention obtains image information of a medical invoice from different block nodes of a blockchain network; extracts text data based on the image information, and performs recognition processing on the text data based on a trained recognition model, wherein the recognition model is obtained by feature extraction training based on principal component analysis; determines a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; when the quantization parameter meets a preset quantization range, determines the medical term as the recognition result of the image information, thereby achieving accurate recognition of medical content based on a combination of machine learning and recognition correction, greatly improving the depth of recognition, and thus meeting the requirements for efficient recognition of medical content in medical invoices.

[0130] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a device for recognizing image information, such as Figure 3 As shown, the device includes:

[0131] An acquisition module 31 is used to obtain image information of medical invoices from different block nodes of the blockchain network;

[0132] a processing module 32 for extracting text data based on the image information and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis;

[0133] A first determination module 33 is configured to determine a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information;

[0134] The second determining module 34 is configured to determine the medical term as the recognition result of the image information when the quantization parameter meets a preset quantization range.

[0135] The present invention provides an image information recognition device. An embodiment of the present invention obtains image information of a medical invoice from different block nodes of a blockchain network; extracts text data based on the image information, and performs recognition processing on the text data based on a trained recognition model, wherein the recognition model is obtained by feature extraction training based on principal component analysis; determines a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; when the quantization parameter meets a preset quantization range, determines the medical term as the recognition result of the image information, thereby achieving accurate recognition of medical content based on a combination of machine learning and recognition correction, greatly improving the depth of recognition, and thus meeting the requirements for efficient recognition of medical content in medical invoices.

[0136] Furthermore, as a response to the above Figure 2 In order to realize the method shown in FIG, the embodiment of the present invention provides another image information recognition device, such as Figure 4 As shown, the device includes:

[0137] An acquisition module 41 is used to obtain image information of medical invoices from different block nodes of the blockchain network;

[0138] a processing module 42 for extracting text data based on the image information and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis;

[0139] A first determination module 43 is configured to determine a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information;

[0140] The second determining module 44 is configured to determine the medical term as the recognition result of the image information when the quantization parameter meets a preset quantization range.

[0141] Furthermore, the device further comprises: a conversion module 45, an extraction module 46, a training module 47,

[0142] The acquisition module 41 is further configured to acquire a training sample set from the preset text resource library containing at least one medical term;

[0143] The conversion module 45 is used to perform word vector conversion on the training samples in the training sample set according to the word vector conversion algorithm to obtain a vector matrix;

[0144] The extraction module 46 is configured to extract eigenvectors matching the vector matrix from the training samples based on principal component analysis;

[0145] The training module 47 is used to train the convolutional neural network based on the feature vector to obtain a trained recognition model.

[0146] Furthermore, the quantization parameters include edit distance and similarity, and the first determination module 43 includes:

[0147] A calculation unit 4301 is configured to calculate a similarity between the recognized text data and at least one corresponding medical term in a preset text resource library;

[0148] a determining unit 4302 for calculating an edit distance between the text data and the medical term when the calculated similarity is greater than a preset similarity threshold;

[0149] The configuration unit 4303 is configured to configure the quantization parameter to be empty when the calculated similarity is less than or equal to a preset similarity threshold.

[0150] Furthermore, the processing module 42 includes:

[0151] The recognition unit 4201 is configured to recognize text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier;

[0152] A first processing unit 4202 is configured to perform a tag replacement process on the text data carrying the tag, wherein the tag replacement process is configured to replace the tag with text content matching the text data;

[0153] The conversion unit 4203 is used to convert the text data that has completed the replacement processing into a word vector for recognition processing.

[0154] Furthermore, the processing module further includes: an establishing unit 4204,

[0155] The establishing unit 4204 is configured to establish a replacement mapping relationship between different text contents based on the preset text resource library, wherein the replacement mapping relationship includes a mapping relationship between the preceding text content and the succeeding text content of the text content to be replaced;

[0156] The first processing unit 4202 is specifically used to compare the preceding text content and following text content of each identifier in the text data carrying the mark in turn, determine the replacement text content that matches the preceding text content and following text content according to the replacement mapping relationship, and replace the determined replacement text content with the mark to obtain the text content after replacement processing.

[0157] Furthermore, the processing module 42 further includes:

[0158] an acquisition unit 4205 for acquiring a straight line in the image information and determining a difference between an inclination angle of the straight line and a reference angle of the image information;

[0159] The second processing unit 4206 is configured to perform rotation processing on the image information using the difference.

[0160] Furthermore, the device further comprises:

[0161] A generating module 48 is configured to generate processing request information including the image information and the text data when the quantization parameter does not meet a preset quantization range;

[0162] A sending module 49 is configured to send the processing request information to the block node based on the public key of the blockchain network to request recognition of the processing request information;

[0163] The receiving module 410 is configured to receive the recognition result of the processing request information and update the recognition result into the training sample set of the recognition model.

[0164] The present invention provides another device for recognizing image information. An embodiment of the present invention obtains image information of a medical invoice from different block nodes of a blockchain network; extracts text data based on the image information, and performs recognition processing on the text data based on a trained recognition model, wherein the recognition model is obtained by feature extraction training based on principal component analysis; determines a quantization parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node that matches the image information; when the quantization parameter meets a preset quantization range, determines the medical term as the recognition result of the image information, thereby achieving accurate recognition of medical content based on a combination of machine learning and recognition correction, greatly improving the depth of recognition, and thus meeting the requirements for efficient recognition of medical content in medical invoices.

[0165] According to one embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction. The computer executable instruction can execute the image information recognition method in any of the above method embodiments.

[0166] Figure 5 A schematic structural diagram of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0167] like Figure 5 As shown, the computer device may include: a processor (processor) 502 , a communication interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

[0168] The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0169] The communication interface 504 is used to communicate with other devices such as clients or other servers.

[0170] The processor 502 is configured to execute the program 510 , and specifically to execute the relevant steps in the above-mentioned embodiment of the method for recognizing image information.

[0171] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0172] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a computer device may be of the same type, such as one or more CPUs, or may be of different types, such as one or more CPUs and one or more ASICs.

[0173] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0174] The program 510 may be specifically configured to enable the processor 502 to perform the following operations:

[0175] Obtain image information of medical invoices from different block nodes of the blockchain network;

[0176] Extracting text data based on the image information, and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis;

[0177] Determining a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information;

[0178] When the quantization parameter meets a preset quantization range, the medical term is determined as a recognition result of the image information.

[0179] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing device, centralized on a single computing device, or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than that shown, or can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0180] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for recognizing image information, characterized in that: include: Obtain image information of medical invoices from different block nodes of the blockchain network; Extracting text data based on the image information, and performing recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by feature extraction based on principal component analysis; Determining a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information; When the quantization parameter meets a preset quantization range, determining the medical term as a recognition result of the image information; The step of extracting text data based on the image information includes: Recognizing text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier; Performing a tag replacement process on the text data carrying the tag, wherein the tag replacement process is used to replace the tag with text content that matches the text data, the text content being found based on a replacement mapping relationship, wherein the replacement mapping relationship includes a mapping relationship between text content preceding and following the text content to be replaced; Convert the replaced text data into word vectors for recognition; The quantitative parameters include edit distance and similarity. The quantitative parameters between the recognized text data and the corresponding at least one medical term in the preset text resource library include: Calculating the similarity between the recognized text data and at least one corresponding medical term in a preset text resource library; When the calculated similarity is greater than a preset similarity threshold, calculating the edit distance between the text data and the medical term; When the calculated similarity is less than or equal to a preset similarity threshold, configuring the quantization parameter to be empty; Before performing the mark replacement process on the text data carrying the mark, the method further includes: Establishing a replacement mapping relationship of different text contents based on the preset text resource library; The process of performing mark replacement on the text data carrying the mark includes: Compare the preceding and following text contents of each identifier in the text data carrying the mark in turn, determine the replacement text content that matches the preceding and following text contents according to the replacement mapping relationship, and replace the determined replacement text content with the mark to obtain the replaced text content.

2. The method according to claim 1, characterized in that Before performing recognition processing on the text data based on the trained recognition model, the method further includes: Acquire a training sample set from the preset text resource library containing at least one medical term; Performing word vector conversion on the training samples in the training sample set according to a word vector conversion algorithm to obtain a vector matrix; extracting eigenvectors matching the vector matrix from the training samples based on principal component analysis; The convolutional neural network is trained based on the feature vector to obtain a trained recognition model.

3. The method according to claim 1, characterized in that Before using the image recognition technology (OCR) to recognize text data in the image information in a row-by-row iterative manner, the method further includes: Acquire a straight line in the image information, and determine a difference between an inclination angle of the straight line and a reference angle of the image information; The image information is rotated using the difference.

4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: When the quantization parameter does not meet the preset quantization range, generating processing request information including the image information and the text data; Sending the processing request information to the block node based on the public key of the blockchain network to request recognition of the processing request information; Receive the recognition result of the processing request information, and update the recognition result to the training sample set of the recognition model.

5. A device for recognizing image information, characterized in that: include: The acquisition module is used to obtain the image information of medical invoices from different block nodes of the blockchain network; a processing module, configured to extract text data based on the image information and perform recognition processing on the text data based on a trained recognition model, wherein the recognition model is trained by performing feature extraction based on principal component analysis; A first determination module is configured to determine a quantitative parameter between the recognized text data and at least one corresponding medical term in a preset text resource library, wherein the preset text resource library is created in a block node matching the image information; a second determining module, configured to determine the medical term as a recognition result of the image information when the quantization parameter meets a preset quantization range; Wherein, the processing module includes: a recognition unit, configured to recognize text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier; a first processing unit configured to perform a tag replacement process on the text data carrying the tag, wherein the tag replacement process is configured to replace the tag with text content that matches the text data, wherein the text content is found based on a replacement mapping relationship, wherein the replacement mapping relationship includes a mapping relationship between text content preceding and following the text content to be replaced; a conversion unit, configured to convert the replaced text data into word vectors for recognition processing; The quantization parameters include edit distance and similarity, and the first determination module includes: a calculation unit, configured to calculate a similarity between the recognized text data and at least one corresponding medical term in a preset text resource library; a determining unit, configured to calculate an edit distance between the text data and the medical term when the calculated similarity is greater than a preset similarity threshold; a configuration unit, configured to configure the quantization parameter to be empty when the calculated similarity is less than or equal to a preset similarity threshold; The processing module includes: a recognition unit, configured to recognize text data in the image information using optical character recognition (OCR) technology in a row-by-row iterative manner, wherein unrecognized text content in the text data is marked with a preset identifier; a first processing unit, configured to perform a mark replacement process on the text data carrying the mark, wherein the mark replacement process is configured to replace the mark with text content matching the text data; The conversion unit is used to convert the text data that has completed the replacement processing into a word vector for recognition processing.

6. A storage medium storing at least one executable instruction, wherein the executable instruction enables a processor to execute an operation corresponding to the image information recognition method according to any one of claims 1 to 4.

7. A computer device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the image information recognition method according to any one of claims 1 to 4.

Images