Intelligent voice control and keyword extraction method for sheep disease diagnosis

Abstract

The invention relates to an intelligent voice control and keyword extraction method for sheep disease diagnosis, and aims to improve the efficiency and accuracy of sheep disease diagnosis through an intelligent voice recognition technology. The method comprises the steps of awakening a diagnosis system, extracting symptom keywords, processing a cross-domain request and performing disease diagnosis through an intelligent voice instruction. According to the method, a big data technology and an intelligent voice recognition technology are utilized, a user inputs symptom description through voice, a system automatically recognizes and extracts keywords, then disease information in a database is matched, a diagnosis result is rapidly given, and the disease diagnosis ability and efficiency of basic-level farmers are improved.

Description

Technical Field

[0001] The invention relates to the technical field of animal disease diagnosis, and in particular to an auxiliary diagnosis method for sheep diseases based on intelligent speech recognition. Background Art

[0002] In the field of animal husbandry, sheep disease diagnosis has always been an important part of ensuring healthy breeding and economic benefits. More and more disease diagnosis systems have emerged, but the traditional sheep disease expert system diagnosis method mainly relies on the professional knowledge and experience of veterinarians, and has problems such as low diagnostic efficiency and susceptibility to human factors.

[0003] With the development of artificial intelligence technology, intelligent voice control and keyword extraction technology are increasingly used in various fields. In the medical field, these technologies have been used to assist doctors in diagnosis and treatment. However, in animal husbandry, especially in the diagnosis of sheep diseases, the application of related technologies is still immature.

[0004] At present, there is a lack of a method that can combine intelligent voice control and keyword extraction technology to achieve efficient and accurate diagnosis of sheep diseases on the market. Therefore, it is urgent to develop a new technical means to improve the efficiency and accuracy of sheep disease diagnosis and provide strong support for the development of animal husbandry.

[0005] The following problems exist in previous expert systems: ① High threshold for use: The disease diagnosis expert system serves large-scale sheep farms and individual farmers. Before use, it is necessary to have a certain understanding of sheep diseases. Grassroots sheep farm personnel who have not been exposed to expert systems cannot easily master this platform. ② Single diagnostic logic: If the user selects fewer symptoms on the diagnosis page, so that the cumulative weight of these symptoms fails to reach the minimum diagnostic threshold of goat diseases set by the system, the system will not be able to provide users with specific disease diagnosis information. ③ The description of the disease is too brief: It is often difficult to fully display the detailed information of the disease by relying solely on simple text descriptions, which may lead to misdiagnosis and bring unnecessary economic losses to the sheep farm.

[0006] The present invention conducts a large amount of data mining on structured data and unstructured data in the sheep disease database, including pathological data, symptom data, laboratory test results, etc. After preliminary standardization of the data, the sheep disease attribute value is extracted. It can accurately reflect the weight of the disease symptoms, that is, the diagnostic value. Each keyword corresponds to an attribute value. The size of the attribute value is determined by the frequency of occurrence of the symptom in all diseases. The lower the frequency of occurrence, the larger the attribute value. If the symptom is unique to the corresponding disease, the greater its diagnostic value.

[0007] The voice diagnosis technology uses the most advanced end-to-end voice recognition model, which has a recognition accuracy of up to 98% through powerful cloud computing capabilities and training with a large text database. In addition, the model uses deep learning and autonomous optimization to ensure accurate voice recognition. After returning the text, the system will remove unnecessary words and match all symptom keywords with the keywords in the database, and finally automatically jump to the diagnosis page to realize an intelligent operation process.

[0008] Based on computer technology, the present invention conducts in-depth mining of massive structured and unstructured disease data, extracts attribute values ​​of keyword sets that can accurately describe disease symptoms, and thus achieves more scientific and accurate disease judgment. Combined with advanced speech recognition technology, it achieves an intelligent diagnosis process, improving the practicality, convenience and accuracy of the diagnosis system. Summary of the invention

[0009] In order to solve the above problems, the present invention provides an intelligent voice control and keyword extraction method for sheep disease diagnosis, which realizes rapid and accurate auxiliary diagnosis of sheep diseases through intelligent voice recognition technology.

[0010] The present invention provides a process for intelligent diagnosis of sheep diseases, which improves the accuracy and practicality of the diagnosis system. The method for building the diagnosis system is as follows:

[0011] Step 1: Use big data technology to establish a sheep disease database. Consult domestic and foreign sheep disease-related literature books and authoritative websites, and use big data technology to extract knowledge from the collected structured data (see ①) and unstructured data (see ②), and store the standardized data in the diagnostic system database. Diseases are divided into eight categories: infectious diseases, nutritional metabolic diseases, reproductive diseases, internal medicine diseases, lamb diseases, parasitic diseases, surgical diseases, and toxic diseases. Each disease is described in detail, including: a brief overview of the disease, pathogens and epidemiology, clinical symptoms, pathological changes, laboratory tests, diagnostic points, differential diagnosis, prevention and control measures, etc.

[0012] Step 2: Extract disease attribute values. The extraction process is based on standardized processing of disease data, involving two tables. The specific method is as follows:

[0013] 1. Sheep disease data table: The collected sheep disease data are standardized and classified into eight categories: infectious diseases, nutritional metabolic diseases, reproductive diseases, internal medicine diseases, lamb diseases, parasitic diseases, surgical diseases, and toxic diseases. The description of each disease consists of fields such as a brief overview of the disease, pathogens and epidemiology, clinical symptoms, pathological changes, laboratory tests, diagnostic points, differential diagnosis, and prevention and control measures.

[0014] 2. Sheep disease symptom table: It is the specific symptoms and physiological characteristics shown by sheep when they are sick (see symptom extraction ③). Including changes in body temperature, respiratory rate, abnormal behavior, weight changes, and loss of appetite. By identifying and analyzing these attribute values, it is possible to accurately determine what disease the sheep is suffering from. ① Structured data: They have a clear fixed format and are stored in the database in a unified and regular manner, which is easy for computers to process and identify. The structured sheep disease data collected includes link data from authoritative websites and database data in the background. ② Unstructured data: In contrast to structured data, there is no clear format, and it is relatively complicated for computers to process. However, it plays an important role in disease diagnosis and auxiliary treatment, including video diagnosis and treatment processes, pathological pictures, text information, etc. for each disease. ③ Attribute value extraction: Analyze and summarize the original sheep disease data, and estimate the weights of all keywords that accurately describe the clinical symptoms of each disease.

[0014] Step 3: Construction of intelligent speech recognition system. The specific process is as follows: (1) Establish an intelligent speech control module. This module is used to convert speech into text to facilitate subsequent operations. This module needs to use a suitable speech-to-text model to perform efficient and accurate speech recognition. The Baidu Cloud speech recognition model is compared with the browser's built-in speech recognition model. Due to the powerful cloud computing resources and a large amount of training data of the Baidu Cloud speech recognition model, the Mandarin recognition accuracy rate is as high as 98%. In addition, deep learning, model self-optimization and other features make speech recognition faster and more accurate. Therefore, the Baidu Cloud speech recognition model is selected. (2) Establish a wake-up diagnosis system module. This module is used for the system to grant diagnosis authority after the user inputs the predetermined wake-up word by voice. After selecting a suitable speech recognition model, the speech recognition function needs to be added to the disease diagnosis system in order to achieve intelligent operation steps, that is, the user does not need to click manually during use to complete the diagnosis process. (3) A symptom keyword extraction module. This module is used to extract key information from the user's voice input. After waking up the diagnosis system, the user's input speech is analyzed and recognized, and after being converted into text, the attribute value of the symptom description in the text is extracted using a word segmenter. (4) A disease diagnosis matching module. This module is used to match the extracted keywords with the disease information in the database and output the diagnosis results.

[0015] 2. The significant feature of the application of the above-mentioned intelligent auxiliary diagnosis system in the expert system is that the user only needs to input the symptoms by voice, and the voice recognition model can convert it into text and extract the symptom keywords. The keywords are then compared with the content in the database, and the auxiliary diagnosis system will give priority to recommending diseases with greater diagnostic value. The user can perform voice-assisted diagnosis again based on other symptoms to complete the final comparison results and display the diagnosed disease.

[0016] The present invention has the following beneficial effects: (1) Easy operation: through voice control, the diagnosis process is simplified and the operation difficulty is reduced; (2) High diagnostic efficiency: intelligent voice recognition and keyword extraction technology greatly improve the diagnosis speed. (3) High accuracy: combined with big data and expert system, the accuracy of diagnosis is improved. (4) Wide applicability: It is suitable for sheep farms of various sizes, especially in remote areas or when there is a lack of professional veterinarians. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above aspects of the present invention will be more easily understood in conjunction with the accompanying drawings. Figure 1 What is shown are the recognition results of different speech models; Figure 2 The wake-up voice system is shown; Figure 3 What is shown is a complete schematic diagram of the working of intelligent auxiliary diagnosis software; Figure 4 What is shown is the diagnostic process of intelligent auxiliary diagnosis software; Figure 5 What is shown is a continued speech diagnosis indication; Figure 6 It shows the details of the disease; Figure 7 Video prevention and control measures are shown. Specific embodiments

[0018] The embodiments of the present invention are described in detail below, and the present invention is further described in detail in conjunction with the embodiments (website: cms.cars38.com). The embodiments herein are only used to illustrate the present invention and cannot be understood as limiting the present invention.

[0019] 1. Disease data standardization Data on sheep diseases were collected from authoritative websites, books, documents and field investigations. The structured and unstructured data were extracted using big data technology, and the extracted sheep disease data were stored in the auxiliary diagnosis system "Medical Chain". A total of 195 sheep disease data were extracted and classified into infectious diseases, nutritional metabolic diseases, reproductive diseases, internal medicine diseases, lamb diseases, parasitic diseases, surgical diseases, and toxic diseases.

[0020] 2. Extraction of attribute values A disease database table was established (named disease in the MySQL database). The data table contains 195 records including disease identity code, disease code, disease name, disease alias, pathogen and epidemiology, clinical symptoms, pathological changes, laboratory tests, diagnostic points, video links, and prevention and control measures. The corresponding fields are: id, dis_code, dis_name, dis_alias, dis_reason, symptom, pathology, check, video, and treatment. Create a disease symptom data table (named symptoms in the MySQL database), which includes disease id, symptom code, symptoms, internal short code, disease code, attribute value, etc. The corresponding fields are: id, symcode, symp, innercode, discode, weight. Symptom internal short code: It consists of two digits, starting with 01, and will increase as the number of symptoms in the disease increases. Symptom code: It consists of the disease code and the symptom internal code. The "1" in front of the disease code 101 represents an infectious disease, and the "1" behind it represents the first disease in the infectious disease. For example, the infectious disease pasteurellosis is labeled 103, and its symptom "bloody stool" has an internal short code of 03, then the internal short code of "bloody stool" in sheep diseases is 103-03. In the database after data standardization, if the description of a certain disease is too complicated, the description of the disease is extracted, and the keyword that can best summarize the disease is used as the feature value. For example, the description of the infectious disease tetanus is "In the early stage of the disease, the symptoms of adult sheep may not be obvious, specifically manifested as the inability to lie down or stand up independently. As the disease progresses into the middle stage, the sheep's ears are upright, the mouth is tightly closed, drooling, and swallowing is difficult." Then, its symptoms are extracted as "unable to stand", "mouth tightly closed", "drooling", "ears upright", "difficulty in swallowing", so its attribute value is {unable to stand, mouth tightly closed, drooling, ears upright, difficulty in swallowing}. In this way, 1621 records were extracted from 195 diseases. The logic of the auxiliary diagnosis system is based on the matching degree of attribute values. The system will recommend related diseases based on the symptom keywords entered by the user, and display the symptom keyword set corresponding to each disease. After the user clicks on a disease, he can view the relevant pathological symptom pictures and video content containing detailed prevention and treatment measures.

[0021] 1. Speech to text The voice-assisted diagnosis system uses automatic speech recognition technology, which is a very natural human-level interaction method. The task of automatic speech recognition is to recognize the sound input sequence X = {x1,…,x T}, the tag sequence L of length T = {L1,…,L N} (usually letters, words or Chinese characters), length is N. Here, is the D-dimensional speech input vector (such as a Mel filter bank) corresponding to the T-th speech frame, is the vocabulary of labels, is the label of the position u in L. We use To indicate that The task of ASR is to find the most likely tag sequence given X. formal: This technology is used and integrated with speech database, acoustic model training, dictionary program, language model training, and the database of this article to form a relatively mature intelligent speech recognition model. The present invention selects the browser's own speech recognition model and the Baidu Cloud speech recognition model, both of which are equipped with the most advanced speech recognition technology. The browser's own speech recognition model can be directly called, which is more convenient. The Baidu Cloud recognition model needs to call the interface, upload to Baidu Cloud, and then return the text. (1) Browser implementation of speech-to-text conversion: Use JavaScript to access the browser's speech recognition function. First, initialize the speech recognition object and set the recognition language to Chinese. When you click on record, start recognition, monitor the speech recognition results, obtain the search box element, and put the recognition results into the search box (as shown below). Figure 1 (shown) (2) Baidu Cloud Speech Recognition Model: Download the configuration file of Baidu Cloud Speech Recognition Model, create a speech service application in Baidu Cloud Console, debug the model online, enter APP_ID, API_KEY, SECRET_KEY in baidu.php, set the format to "wav", the frequency to "16000", the recognized speech to "Chinese", create upload.php to upload the speech to Baidu Cloud, click Record to start recognition, and the recognition result is returned to the search box. (As shown below Figure 1 (shown) The recognition results are as follows Figure 1 As shown in the figure, the key to speech recognition is the speech recognition model, which is related to the accuracy of the subsequent diagnosis results. Although the browser speech recognition model is more convenient to use, it is not accurate when the recognized speech file is too long, which will cause the failure of keyword extraction and thus diagnosis failure. The Baidu Cloud speech recognition model uses the powerful cloud computing resources and a large amount of training data to achieve a Mandarin recognition accuracy of up to 98%. In addition, deep learning, model self-service optimization and other features make speech recognition faster and more accurate.

[0022] 2. Wake up the diagnostic system After selecting an appropriate speech recognition model, the speech recognition function needs to be added to the disease diagnosis system. To achieve an intelligent operation process, that is, the user does not need to click manually during use to complete the diagnosis process. Use JavaScript code to control the start and end of recording. When we enter the diagnosis page and start speaking, the power level variable in the JavaScript code will change. It will change its size according to the sound of our speech. If it is greater than 1, recording will start. If it is less than 1 for three consecutive seconds, recording will end. If not speaking, there will be no recording and recognition. After the recording ends, the speech will be converted into text for the next diagnosis operation. To prevent users from misoperating, the diagnosis system needs to be awakened before each diagnosis. Only when the text converted from speech recognition contains the three words "Doctor Yang" can the next diagnosis be carried out (for the specific process, see Figure 2 ).

[0023] 3. Extract symptom keywords After recognizing the three words "Doctor Yang", the diagnosis process begins. The diagnosis requires symptom keywords. However, what we say is a paragraph, and we need to find the corresponding disease symptom keywords in the database from this paragraph. Therefore, a tokenizer component composer is needed to break a paragraph into individual words, and then eliminate the unnecessary words, only retaining the symptom keywords. Create a.text and b.text in composer. a.text is a dictionary created by yourself, and enter the words that the tokenizer cannot handle in it. For example, for the disease symptom name "winter and spring epidemic", the tokenizer gives two words "winter and spring" and "epidemic", and the database cannot retrieve them. Enter "winter and spring epidemic" in a.text to handle it. b.text is a dictionary for eliminating unnecessary keywords. For example, for a recognized paragraph "Doctor Yang, my sheep has diarrhea", enter "Doctor Yang", "my sheep", and "has" in b.text to eliminate the unnecessary keywords.

[0024] 4. Compare symptoms When the keywords are extracted, put the symptom keywords into the database and compare them with the symptoms one by one. After successful comparison, jump to the diagnosis page. However, when extracting keywords, if the user says an irrelevant word, for example: the extracted words are "cough, I eat, subcutaneous hemorrhage", the comparison is unsuccessful because "I eat" has no corresponding disease in the database, so the diagnosis is unsuccessful. To avoid this phenomenon, in the returned text, list the useless keywords or the keywords that cannot be found in the database separately, and then restore and delete them, and then carry out the diagnosis.

[0025] IV. Application effect The sheep disease intelligent auxiliary diagnosis system uses segmented voice intelligent inquiry, which is compared with the disease name table, disease alias table, disease symptom table and the Internet in turn to obtain the disease diagnosis result. In the intelligent auxiliary diagnosis system, the user inputs the symptoms by voice, and the voice recognition model converts the speech to text. The composer component analyzes the converted text to extract the symptom keywords, and puts the symptom keywords into the database for comparison. After the comparison is successful, it automatically jumps to recommend the corresponding disease and all corresponding symptoms to the user. The user can choose the corresponding symptom of the disease or say the observed symptoms again for the final comparison, and finally outputs the diagnosis result (see the specific diagnosis process). Figure 4 ). Users can click on the corresponding disease to view its detailed diagnosis and treatment methods (as shown below Figure 6 If you have any doubts, you can watch the relevant videos to learn more about the specific prevention process and implementation points, so as to fully grasp the scientific knowledge of disease prevention and control (as shown below) Figure 7 In addition, the user can also say multiple symptoms to complete the matching process at one time, and can continue to wake up the system for diagnosis after the diagnosis is completed (the diagnosis process is as shown in Figure 4 Figure 5 shown). The sheep disease diagnosis expert system with intelligent voice operation has further promoted the development of intelligent expert systems. It has mainly solved the following problems and has the following significance: 1) The previous expert system knowledge base is old, the diagnosis of diseases depends on the experience of experts, and the diagnostic logic is not scientific enough. Nowadays, big data technology has improved the knowledge base, established a picture library and a short video library, and the scientific identification method has made disease diagnosis more scientific, reasonable, intuitive and accurate, providing reliable disease prevention and treatment solutions for grassroots animal husbandry breeders, and promoting the circular development of animal husbandry. 2) Adding an intelligent voice recognition model realizes the process of voice operation of the sheep disease diagnosis expert system, so that the expert system can meet the needs of breeders with different needs and no longer single application scenarios. 3) The intelligent expert system has promoted the development of smart animal husbandry, improved the efficiency of breeding production, solved the problem of lack of veterinary diagnosis and untimely diagnosis, and avoided a lot of economic losses.

Claims

1. An intelligent voice control module, characterized in that: This module is used to convert speech into text for subsequent operations. The intelligent voice control module includes: (1) a speech recognition unit, which is used to select and apply an appropriate speech-to-text model to achieve efficient and accurate speech recognition; (2) The speech recognition unit adopts the Baidu Cloud speech recognition model. By comparing it with the built-in speech recognition model of the browser, the Mandarin recognition accuracy rate reaches 98% due to the advantages of the Baidu Cloud speech recognition model's cloud computing resources and training data. The recognition speed and accuracy rate are improved through deep learning and model self-optimization technology.

2. A wake-up diagnostic system module, characterized in that: This module is used to grant the system the authority to perform diagnosis after the user enters a predetermined wake-up word. The wake-up diagnosis system module includes: (1) a wake-up unit, used to monitor and recognize a predetermined wake-up word; (2) A disease diagnosis activation unit, which is used to activate the disease diagnosis system after voice recognition, so that the user can complete the diagnosis process through voice without manual operation.

3. A symptom keyword extraction module, characterized in that: This module is used to extract symptom-related keywords from the user's voice input. The symptom keyword extraction module includes: (1) a speech analysis unit, used to convert a speech signal input by a user into text; (2) Word segmentation units are used to extract attribute values ​​of symptom descriptions in the text for subsequent disease diagnosis.

4. A disease diagnosis matching module, characterized in that: This module is used to match symptom keywords with disease information in the database and output diagnosis results. The disease diagnosis matching module includes: (1) a keyword matching unit, used to compare the extracted symptom keywords with the disease information in the database; (2) A diagnosis output unit, which is used to output disease information with diagnostic value based on the comparison results as auxiliary diagnosis suggestions.

5. Application of the intelligent auxiliary diagnosis system according to any one of claims 1 to 4 in an expert system, characterized in that: The system allows users to input symptom descriptions through voice. The speech recognition model converts the speech into text and automatically extracts symptom keywords, which are further compared with the database content. Diseases with higher diagnostic value are recommended first. Users can input more symptom comparison information through voice again, and finally output the diagnosis results.

Images