Agricultural structured data acquisition method and system based on voice intelligence

Abstract

The invention relates to the field of artificial intelligence, in particular to an agricultural structured data acquisition method and system based on voice intelligence, and the method comprises the steps: receiving the voice input of the agricultural field, determining an acquisition scene, and calling a predefined field set; the method comprises the following steps: intelligently matching a speech recognition text with a field set by utilizing the deep semantic understanding capability of a large language model, and performing accurate mapping and standardized conversion on spoken expressions, terminologies and numerical units to generate initial structured data; and furthermore, in combination with a preset agricultural field knowledge rule, logic verification and rationality analysis are performed on a conversion result driven by the large model, data confirmation or correction is realized through voice interaction, and finally, high-quality structured data is output. According to the method, the cognitive reasoning ability of a large language model is deeply integrated into an agricultural data acquisition full link, intelligent conversion from non-professional speech to standardized data is realized, and the problems that a traditional mode is low in efficiency and general speech recognition lacks field semantic understanding are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an agricultural structured data acquisition method and system based on voice intelligence. BACKGROUND

[0002] In current agricultural research and production management, it is necessary to systematically collect multiple types of information such as crop traits, environmental parameters, and farming operations, and organize them into agricultural structured data, i.e., table-type data with clear field definitions, standard data types, and standard units, which can be directly used for statistical analysis and intelligent decision-making. Currently, the generation of such structured data mainly relies on two ways: one is for workers to record in the field using paper forms and then manually enter the computer after returning, which is a cumbersome process, prone to errors, and low in timeliness; the other is to use electronic forms on mobile terminals such as smartphones for on-site point selection or input, which improves the level of digitization, but in typical agricultural environments such as strong light, rain, and muddy hands, screen operation is difficult, and actual input efficiency is still not ideal.

[0003] In the prior art, at the data entry and preliminary structuring level, the prior art cannot achieve the automatic conversion of "semantic understanding" and "field adaptation". Specifically, general voice recognition technology can only convert voice into text, but cannot understand the specific business scenarios and data connotations of agriculture. For example, it cannot associate "piglet" with the standard field "plant height", and the existing method is limited to the format or simple range of the field itself, lacking cross-field logical verification based on agricultural domain knowledge, such as the ability to identify contradictions such as "sowing period plant height 3 meters" that violate common sense of farming.

[0004] Secondly, at the agricultural professional application level, the prior art lacks adaptability to the complexity and dynamics of agricultural production, resulting in insufficient data rationality and usability. The data of crops or livestock and poultry changes dynamically with growth stages, breed characteristics, and environmental conditions. The existing rigid data verification model may not be able to distinguish between "measurement errors" and "environmentally induced reasonable abnormalities", and may misjudge valuable stress physiology data as invalid data and filter them out. SUMMARY

[0005] In order to solve one or several problems in the prior art, the main purpose of the present application is to provide an agricultural structured data acquisition method and system based on voice intelligence.

[0006] In order to achieve the above-mentioned purpose of the application, the present application provides an agricultural structured data acquisition method based on voice intelligence, which comprises: receiving user input of agricultural field-related voice data; According to the voice data, a current data collection scene is determined, and a collection field set predefined for the collection scene is called, wherein the collection field set comprises a standard name, a data type, and a value constraint of a field; Text information recognition is performed on the voice data. According to a recognition result, the recognized text information is matched with the collection field set, expression in the text is mapped to a corresponding standard field name, and a numerical value in the text is standardized and converted with a unit according to the data type and the value constraint, to generate initial structured data. The initial structured data is verified based on the value constraint in the collection field set and a preset agricultural knowledge rule. Final structured data is determined based on a verification result.

[0007] Embodiments of the present application also provide an agricultural structured data collection system based on voice intelligence, comprising: A receiving module is configured to receive voice data related to an agricultural field input by a user. A calling module is configured to determine a current data collection scene according to the voice data, and call a collection field set predefined for the collection scene, wherein the collection field set comprises a standard name, a data type, and a value constraint of a field. An identification module is configured to perform text information recognition on the voice data. A generation module is configured to, according to a recognition result, match the recognized text information with the collection field set, map expression in the text to a corresponding standard field name, and standardized and convert a numerical value in the text with a unit according to the data type and the value constraint, to generate initial structured data. A verification module is configured to verify the initial structured data based on the value constraint in the collection field set and a preset agricultural knowledge rule. A determination module is configured to determine final structured data based on a verification result.

[0008] The present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements steps of the method according to any one of the preceding embodiments when executing the computer program.

[0009] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements steps of the method according to any one of the preceding embodiments when executed by a processor.

[0010] The method and system for collecting agricultural structured data based on voice intelligence provided in the embodiments of the present application deeply integrate the knowledge in the field of agriculture into the whole process of voice processing, and fundamentally solve the disconnection problem between voice recognition and the demand for structured agricultural data in the prior art. The method first liberates the user's hands and eyes through pure voice interaction, greatly improving the operation adaptability and input efficiency in the field under poor environment. More importantly, the scenario-based predefined field set is used as a "blueprint" to guide the system to complete intelligent mapping from free spoken language to standard fields and numerical unit standardization, realizing automatic and accurate conversion of unstructured voice to structured data, and solving the problems of semantic understanding and format unification. On this basis, a verification mechanism based on agricultural knowledge rules is further introduced to conduct deep quality review of the data including logical relationships, thereby ensuring the agronomic rationality and direct usability of the output data. Overall, the scheme not only realizes the leap of collection efficiency, but also ensures the high quality and high credibility of agricultural data from the source collection link through the empowerment of domain knowledge. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 FIG. 1 is a flowchart of a method for collecting agricultural structured data based on voice intelligence according to an embodiment of the present application; Figure 2 FIG. 1 is a flowchart of a method for collecting agricultural structured data based on voice intelligence according to an embodiment of the present application; Figure 3 FIG. 2 is a structural schematic block diagram of a system for collecting agricultural structured data based on voice intelligence according to an embodiment of the present application; Figure 4 FIG. 3 is a structural schematic block diagram of a computer device according to an embodiment of the present application.

[0012] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail below with reference to the embodiments and the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0014] Referring to Figure 1 The method for collecting agricultural structured data based on voice intelligence provided in the embodiments of the present application comprises: S1, receiving voice data related to the field of agriculture input by a user; S2, determining a current data collection scene according to the voice data, and calling a collection field set predefined for the collection scene, wherein the collection field set comprises a standard name, a data type, and a value constraint of a field; S3, performing text information recognition on the voice data; S4, according to a recognition result, matching the recognized text information with the collection field set, mapping an expression in the text to a corresponding standard field name, and standardizing a value and a unit in the text according to the data type and the value constraint, to generate initial structured data; S5, based on a value constraint in the collection field set and a preset agricultural knowledge rule, performing verification on the initial structured data; S6, based on a verification result, determining final structured data.

[0015] As described in steps S1-S3, step 1 captures a voice signal orally spoken by a user in an agricultural field through an audio input device such as a microphone, as a raw input for data processing. This fundamentally changes the data collection method, and changes the interaction mode from a touch screen or a pen and paper that relies on “hand-eye coordination” to pure voice interaction. This enables the user to conveniently input data even when hands are occupied, the environment is harsh, or in a mobile state, greatly improving the on-site adaptability of data collection. Step 2 performs preliminary analysis on the voice content and identifies keywords related to specific agricultural activities, thereby determining which scene the current task belongs to. Then, the system calls a structured collection field set bound to the scene in advance as a “blueprint” for subsequent processing. This realizes the “scenarization” and “standardization” of collection. Through the predefined field set, the uniformity and completeness of the data structure are ensured, and the missing or confusion of fields caused by the user’s arbitrary oral description is avoided. Step 3 converts continuous voice signals into corresponding text sequences using voice recognition technology. This process can be completed based on a general or optimized agricultural voice corpus recognition engine. Converting audio signals that are difficult to calculate into text information that can be processed by natural language processing technology is a prerequisite for all subsequent intelligent understanding and conversion steps.

[0016] As described in steps S4-S6 above, step 4 is a core conversion step. The system maps the words in the text to the corresponding fields according to the standard names in the field set and the associated glossary. At the same time, according to the data type and value constraint of the field, the numerical value and unit in the text are intelligently converted and standardized. Directly aiming at the existing technical defect one. It solves the problem of general speech recognition "only translation, not understanding", and realizes: semantic error correction and normalization: automatically map colloquial or error words such as "pig lamb" and "seedling height" to the standard field "plant height". Unstructured to structured crossing: extract multiple field values from a sentence, and convert descriptions such as "two meters and three" to standard numerical value "230" and unit "cm", generate initial structured data that can be directly processed, fill the key fault between speech and database. Step 5, after preliminary structuring, the system does not directly adopt the data, but applies two layers of verification: one is based on the value constraint of the field itself; The second is to use the preset agricultural knowledge rules to verify the logical relationship across fields. The introduction of domain knowledge driven quality control not only checks the data format, but also finds logical contradictions such as "sowing period plant height 3 meters" which violate agricultural common sense, significantly improving the reliability and agronomic reasonableness of the data. Step 6 confirms the data that passes the verification, and marks or triggers the subsequent correction process for the data that does not pass the verification, and finally outputs a set of structured data set after quality control. Ensure the credibility and usability of the output data. The final generated data set is format specification, logically self-consistent, and can be directly imported into the database or analysis software, realizing the data ready state of collection and analysis. In this embodiment, the scheme combines "scenario field set calling" and "intelligent mapping and standardization based on field definition", and creatively injects agricultural domain knowledge into the speech conversion process. This is not just "recognizing words", but also "understanding intentions and formatting", realizing the automatic and high-accuracy conversion from free speech to standard field values. By introducing the "verification based on agricultural knowledge rules" step, the data quality control is upgraded from simple value range check to associated review containing agricultural logic. This makes it possible for the system to distinguish between "wrong data" and "special farming conditions".

[0017] As described above, the deep integration of agricultural field knowledge into the whole process of voice processing fundamentally solves the disconnection between voice recognition and the need for structured agricultural data in the prior art. The method first liberates the user's hands and eyes through pure voice interaction, greatly improving the operation adaptability and input efficiency in the harsh environment of the field. More importantly, using a scenario-based predefined field set as a "blueprint" to guide the system to complete intelligent mapping from free speech to standard fields and numerical unit standardization, it realizes the automatic and accurate conversion of unstructured voice to structured data, and solves the problems of semantic understanding and format unification. On this basis, further introduce the verification mechanism based on agricultural knowledge rules, implement deep quality review of data including logical relationships, and thus ensure the agronomic rationality and direct usability of the output data. Overall, this scheme not only realizes the leap in collection efficiency, but also ensures the high quality and high credibility of agricultural data from the source collection link through the empowerment of domain knowledge.

[0018] Reference Figure 2 In one embodiment, the step of verifying the initial structured data based on the value constraints in the collection field set and the preset agricultural knowledge rules comprises: S51, obtaining the preset agricultural knowledge rules, wherein the agricultural knowledge rules define the logical constraint relationship between different collection fields; S52, substituting the values of the related fields in the initial structured data into the corresponding logical constraint relationship for calculation; S53, judging whether the initial structured data satisfies the logical constraint relationship according to the calculation result; S54, in response to the judgment of satisfaction, determining that the initial structured data passes the verification; S55, in response to the judgment of not satisfying, marking or triggering the correction of the initial structured data.

[0019] As described above, Step 1 retrieves specific agricultural business rules from a pre-built knowledge base. These rules are not isolated field requirements, but rather define the correlation conditions that multiple field values ​​should satisfy in the form of logical expressions. This transforms abstract domain knowledge into concrete logical relationships that can be recognized and executed by a computer. For example, the experience that "the ear position should generally not exceed 70% of the plant height" is transformed into a calculable rule, providing a clear basis for automated and in-depth verification. Step 2 extracts the corresponding specific values ​​from the generated initial structured data based on the fields involved in the rule, substitutes them into the logical expression for calculation, and obtains a Boolean value or quantitative result. This achieves the connection between knowledge and data. Through calculation, actual observation data is placed in the agricultural knowledge system for quantitative verification, making implicit experience explicit. Step 3 analyzes the calculation results of Step 2 and, based on preset judgment criteria, concludes whether the data passes the current rule's verification. This completes the automated decision-making process from calculation to judgment. The system can automatically identify data that is valid within a single field but violates cross-field agricultural logic, such as identifying contradictory records where the sowing period has the plant height at maturity. Steps 4 and 5 perform a data splitting operation based on the judgment conclusion: if the conditions are met, the data validity is confirmed; if not, an anomaly marker is added to the data, or a subsequent interactive correction process is triggered. This forms a closed-loop verification and data governance mechanism. It not only identifies problems but also drives their resolution, ensuring that all data entering the final dataset has passed the knowledge rule review or that abnormal data has been clearly marked, thus guaranteeing the overall quality and credibility of the dataset. In this embodiment, existing technologies typically only perform independent field-level verification and cannot capture the agronomical logical contradictions inherent between data fields. This solution, by introducing the definition and calculation judgment mechanism of "logical constraint relationships," can discover deep-seated errors such as "ear position higher than plant height" or "specific traits appearing at impossible growth stages," errors that traditional value range verification cannot reach.

[0020] In one embodiment, the steps of matching the identified text information with the collection field set, mapping the expressions in the text to corresponding standard field names, and standardizing the numerical values ​​and units in the text according to the data type and value constraints to generate initial structured data include: Based on the standard names of each field in the collection of fields and the predefined word list, the words in the text information are matched with the corresponding fields; Extract the numerical and unit descriptions corresponding to each field from the matched text information; Based on preset conversion rules, the numerical description is converted into a standard numerical format that conforms to the data type; According to the unit defined for each field in the collection of fields, the unit description is converted or converted into the defined unit; According to the standard numerical format and the converted or converted unit, the value of each field in the initial structured data is generated.

[0021] As described above, step 1 uses a pre-constructed association dictionary containing field standard names and their common colloquial expressions, aliases, and even common recognition error words to scan and semantically compare the recognized text, mapping the text to the standard field defined in the collection of fields. The field. Realize the understanding and correction of the field semantics. Make the system understand that the real intention of expressions such as "piglet" and "seedling height" is the "plant height" field, solve the fundamental obstacle of general speech recognition in professional terminology mapping, and complete the key alignment from the original text to the business field. Step 2, after determining the field to which the text segment belongs, the system uses natural language processing technology to identify and separate the words or phrases that describe the size of the value and the measurement unit from the segment or adjacent context. Realize the accurate positioning and separation of key information. It structures the effective data components mixed in natural language and extracts them as raw materials for subsequent standardization processing. Step 3 calls the numerical conversion rule library designed for the agricultural scene to process the extracted numerical description. This includes converting Chinese numbers to Arabic numbers, quantifying fuzzy quantifiers, and unifying decimal and percentage formats. Solve the problem of non-standardized values in agricultural colloquial language. Whether the user says "one thousand two hundred" or "one thousand two", the system can output the standard "1200", ensuring the calculability and consistency of the data in mathematics. Step 4 takes the standard unit defined for each field in the collection of fields as the basis to identify the extracted unit description and convert it when necessary. For example, convert the user's "mu" to the field-defined "square meter" according to the conversion rule. Realize the unity and comparability of the measurement unit. Eliminate the confusion caused by different user habits, so that all data are based on the same measurement standard. Step 5 combines the completely standardized numerical value and unit as the final value of the field, and fills it into the corresponding position of the initial structured data object. Complete the final assembly from unstructured text to machine-readable structured data. At this point, a data record with a standard format, clear meaning, and direct use for database operations or business logic is generated.

[0022] In an embodiment, after the step of determining the current data collection scene, the method further comprises: According to the growth stage or growth cycle of the target crop or target livestock and poultry corresponding to the collection scene, a physiological index mapping relationship matching the growth stage or growth cycle is established; According to the physiological index mapping relationship, the fields related to physiological or growth state in the initial structured data are analyzed for rationality; According to the result of rationality analysis, if the analysis result is unreasonable, candidate correction data is generated based on the physiological index mapping relationship, and the user is guided to confirm or correct through voice interaction.

[0023] As described above, step 1 determines the collection scene, and further determines the specific growth stage of the organism according to the current time, variety information or user specification. Then, a knowledge model corresponding to the stage is called, which defines the reasonable value range or typical characteristics of each physiological state field in this stage. This is essentially a dynamic standard template. It realizes the fundamental change from static to dynamic of the verification standard. It makes the system understand that the reasonable value of "plant height" in the seedling stage, elongation stage and mature stage is completely different, thereby providing a time-varying judgment benchmark in line with biological laws for subsequent analysis. Step 2 compares and evaluates the values of the fields in the initial structured data reported by the user, which describe the growth traits, with the dynamic standard established in step 1 for the current growth stage. This analysis aims to determine whether the observed value deviates significantly from the normal trajectory of its life cycle. The introduction of life cycle-based longitudinal rationality review enables the system to discover "early maturity" or "developmental retardation" phenomena that may be normal in isolation but are abnormal when placed on the growth timeline, achieving deeper data insights. Step 3, when the data is determined to be unreasonable, the system does not simply reject it, but rather infers a more likely reasonable value as a candidate suggestion based on the physiological index mapping relationship. Then, the suggestion is presented to the user through voice dialogue, guiding them to review and make decisions on site. The data quality control is upgraded from one-way alarm to two-way collaborative intelligent assistance. It not only finds problems, but also provides solutions using knowledge, and reduces user operation burden through the most natural voice interaction, achieving efficient human-machine collaborative error correction in the complex field environment. In this embodiment, the quality control dimension of agricultural data collection is expanded from static, field-by-field relationship verification to dynamic, time-series growth and development rule verification. It solves the defects pointed out in the background art: the existing technology lacks adaptability to the complexity and dynamics of agricultural production, leading to possible misjudgment of rationality. The performance of crop traits is the orderly unfolding of its internal genetic program in the time dimension. By establishing a "growth stage-physiological index" mapping relationship, the system first gains the ability to simulate this cognitive process. It makes the verification not only focus on the horizontal contradictions between data points, but also examine whether a single data point is "misplaced" in the vertical growth curve. An intelligent verification layer with time awareness is constructed. It improves the depth of the system's judgment of data rationality and provides a core framework for the handling of more complex scenarios, such as the identification of overall growth curve shifts under environmental stress.

[0024] In an embodiment, the step of performing rationality analysis on the fields related to physiological or growth state in the initial structured data according to the physiological index mapping relationship comprises: extracting reference value intervals or reference value characteristics of each physiological state field corresponding to the growth stage or growth cycle from the physiological index mapping relationship; comparing actual values of the same physiological state field in the initial structured data with corresponding reference value intervals or reference value characteristics; calculating a deviation degree between the actual values and corresponding reference standards according to the comparison results; judging a rationality level of the actual values based on preset deviation degree thresholds, wherein the rationality level comprises at least reasonable, doubtful and unreasonable.

[0025] As described above, step 1 retrieves and obtains quantified reference standards of each physiological state field in the current growth stage from the dynamic physiological index mapping relationship. This standard is not a fixed value, but a reasonable value range or a typical characteristic description. Abstract agricultural growth rules are converted into specific and operable data benchmarks. For example, the specific interval of “30 to 80 centimeters” is extracted for the “wheat plant height in the jointing stage” field, which provides accurate and objective basis in line with agricultural time for subsequent quantitative comparison. Step 2 compares the actual observation values reported by the user with the corresponding reference standards extracted in step 1 in a mathematical or logical manner. This is a process of positioning specific data points in a reasonable background interval. It realizes the first intersection of data and knowledge. Through comparison, it can be directly judged whether an observation value is within the normal interval, at the edge or outside, laying a foundation for subsequent fine evaluation. Step 3 uses a preset algorithm to quantify the degree of deviation of the actual value from the reference standard. This can be the distance from the interval boundary or the relative error from the typical value, and finally output a value representing the size of the deviation. An accurate quantitative evaluation mechanism is introduced. It converts the vague sense of “too big” or “too small” into the precise measurement of “deviating by 20 percent”, so that the judgment of rationality changes from qualitative to quantitative, supporting more fine decision-making. Step 4 forms a hierarchical judgment standard by presetting multiple deviation degree thresholds. According to which threshold interval the calculated deviation degree falls into, the rationality of the data is divided into “reasonable”, “doubtful” or “unreasonable” and other levels. It realizes flexible quality evaluation. It changes the binary judgment mode of “right or wrong”, allows the system to identify and distinguish different situations such as “completely normal”, “slightly suspicious but possibly reasonable” and “extremely likely to be wrong”, and provides key decision-making basis for differentiated subsequent processing.

[0026] In an embodiment, the step of extracting the reference value interval or the reference value feature of each physiological state field corresponding to the growth stage or the growth cycle from the physiological indicator mapping relationship comprises: obtaining current environmental state information associated with the target crop or the target livestock and poultry; judging whether there is a predefined stress environment type according to the current environmental state information, the stress environment type being used to represent a set of environmental conditions that have a specific mode of influence on the growth and development of the target crop or the target livestock and poultry; if there is, extracting a reference value interval or a reference value feature corresponding to the growth stage or the growth cycle and the stress environment type from the physiological indicator mapping relationship; if there is not, extracting a reference value interval or a reference value feature corresponding to the growth stage or the growth cycle and under a normal environment from the physiological indicator mapping relationship.

[0027] As mentioned above, step 1 can obtain the environmental data of the collection site in real time or near real time through Internet of Things devices such as weather stations, soil sensors, or by brief oral description by the user. These data constitute the key context for evaluating the growth background. The system is endowed with environmental perception capability, so that the judgment is no longer limited to the biological data itself, but can place the observation value in the specific environmental background for consideration, providing a crucial input for scenario-based analysis. Step 2 matches and analyzes the acquired environmental data with various stress condition models defined in the knowledge base. These models describe patterned environmental conditions that have a specific adverse effect on growth and development, such as continuous drought, persistent low temperature, or nutrient deficiency. Automatic identification and classification of adverse conditions are achieved. The system can identify the type of environmental data that may cause substantial stress to crops, just like an agronomist, and thus initiate the corresponding intelligent processing logic. Steps 3 and 4 are a key branch logic. If a specific stress is identified, the system calls the reference standard for the growth stage and under the stress type from the knowledge base. This standard reflects the reasonable range of trait performance under this adversity. If there is no stress, the standard under normal environment is called. Dynamic adaptation and refinement of the verification benchmark are achieved. It recognizes that the reasonable performance of the same growth stage under different environments is different, thus providing the most practical judgment scale for subsequent rationality analysis. In this embodiment, an environmental context perception mechanism is introduced, and based on this, dynamic and differentiated selection of judgment standards is achieved. The background technology solves the problem that the existing rigid data verification model may misjudge valuable adversity physiological data as invalid data due to the neglect of environmental interaction. The phenotype of the crop in this scheme is the result of the interaction between genotype and environment. Through the process of "perceiving the environment-identifying the stress-switching the standard", this idea is engineeringized. The system can understand that "shorter plant height" under drought stress may not be an error that needs to be corrected, but rather a key farm condition response that should be faithfully recorded.

[0028] In an embodiment, the step of judging the rationality level of the actual value based on the preset deviation threshold value comprises: According to the identification information of the target crop or target livestock and poultry, a pre-defined stress elasticity coefficient associated with the current stress environment type is obtained; Based on the current environmental state information, the stress intensity of the stress environment type is determined; According to the stress elasticity coefficient and the stress intensity, the preset deviation threshold value is dynamically adjusted to obtain an adjusted deviation threshold value adapted to the current environment and variety; Based on the adjusted deviation threshold value, the rationality level of the actual value is judged.

[0029] As described above, step 1 queries the pre-set resilience coefficient for the identified stress environment type from the variety or species characteristic knowledge base according to the specific identification of the current collection object. The coefficient quantifies the inherent tendency or tolerance of the organism to have its trait performance vary when facing a specific adversity. The variety precision and individualization of the judgment basis are realized. The system can distinguish the response differences of different genetic backgrounds to adversity, such as knowing that a certain wheat variety is strong in drought resistance and another is weak. Step 2 analyzes and calculates the obtained original environmental data to convert it into a quantitative indicator representing the severity of the current stress. For example, the intensity level of drought is divided according to the number of consecutive days when the soil water content is below the threshold. The quantitative evaluation and grading of environmental impact are realized. It pushes the qualitative judgment of "there is drought" to the quantitative description of "moderate drought" or "severe drought", so that the system can perceive the urgency of adversity and provide accurate input for dynamic adjustment. Step 3 combines the resilience coefficient representing the internal tolerance with the stress intensity representing the external pressure, and dynamically calculates a new deviation threshold suitable for the current specific situation through a pre-set algorithm model. When the tolerance is poor and the stress is strong, the threshold may be significantly relaxed. The intelligent self-adaptation and flexibility of the judgment standard are realized. It enables the fault tolerance of the system to change in real time with the variety characteristics and environmental pressure, thereby logically acknowledging that "it is normal for a fragile variety to grow poorly under severe adversity", avoiding the mechanical application of fixed standards. Step 4 finally uses the dynamically adjusted threshold value according to the situation, rather than the fixed initial threshold value, to evaluate the deviation degree of the actual observed data and determine its rationality level accordingly. The ultimate rationality decision of situational awareness is completed. This makes the final judgment result consider not only the observed value itself, but also the variety background and environmental context in which the value is generated, so as to obtain a conclusion that best fits the actual situation of agriculture. In this embodiment, the core idea of "genotype x environment" interaction is further introduced, and through the combination and calculation of the two quantifiable parameters of "resilience coefficient" and "stress intensity", the dynamic and individualized adaptation of the judgment threshold is realized. The background technology solves the problem that the existing rigid verification model completely ignores the quantitative interaction of variety differences and environmental influences, which may misjudge the reasonable abnormal data caused by environmental stress as invalid errors, which have important biological significance, thereby causing the loss of key agricultural information. By constructing an agricultural intelligent decision-making model, it is recognized that the judgment of data rationality cannot be separated from the basic premise of "who is growing in what environment". By dynamically adjusting the threshold, it is essentially performing an advanced anomaly detection, not simply detecting deviations, but after understanding the potential reasons for the deviation, it is judged whether the deviation is within the expected range. Therefore, the invention is a key evolution from rule-based verification to model-based intelligent situational evaluation.It ensures that the system can both sensitively capture real errors and intelligently accept and retain "reasonable anomalies" that record the stories of the interactions between organisms and the environment in the complex and changeable field reality, thereby greatly improving the scientific value and practical efficiency of the collected data.

[0030] In a specific embodiment, taking the field investigation of a certain corn breeding base as an example, Zhang, a breeding technician, is in the test field to investigate the heading stage traits of corn material numbered "B73". Zhang holds the device and speaks out the observation results: "B73, No. 5 field, plant height is about 2.4 meters, ear height is 90 centimeters." The system obtains real-time information through the Internet of Things interface: there is no effective rainfall in the current field for 15 consecutive days, the soil moisture sensor shows that the water content is only 45% of the field water holding capacity, and the air temperature is 35°C. Step A: Determine the scene and establish a dynamic mapping relationship. The system identifies keywords such as "corn", "heading stage", "B73" in the voice, and determines that the scene is "corn heading stage trait investigation". Then, according to the "B73" variety and "heading stage", combined with the current environmental state information (continuous drought, high temperature), the system establishes a dynamic physiological index mapping relationship from the knowledge base. This mapping relationship not only contains the conventional heading stage reference value, but also specially associates the expected trait range under the "moderate water-high temperature combined stress" model. Step B: Intelligent conversion and initial data generation. The system converts the voice into text and maps "plant height" and "ear height" to standard fields. Convert "2.4 meters" to "240 centimeters" and confirm "90 centimeters" as "90 centimeters" to generate initial structured data: {variety: "B73", growth stage: "heading stage", plant height: 240, unit: cm, ear height: 90, unit: cm}. Step C: Environmental perception and differentiated standard extraction. The system analyzes the acquired soil moisture and air temperature data and determines that the current environment is of the pre-defined "water-high temperature combined stress" type. Therefore, the system does not extract from the conventional standard, but from the mapping relationship, the reference value range of "B73" variety at the heading stage under this specific stress environment. For example: the conventional environment reference interval of plant height at the heading stage is [220, 260] cm. The reference interval of plant height under moderate water-high temperature stress is [200, 240] cm (expected growth is inhibited). Step D: Quantitative analysis and grading judgment. The system compares the actual value with the differentiated standard: plant height: actual value 240 cm, within the stress environment reference interval [200, 240] cm. Calculate the deviation: for plant height, the value has reached the upper limit of the stress interval, and the deviation is low. Reasonable level judgment: based on the calculation, the system determines that the plant height data is "reasonable" considering the environmental stress. If this 240 cm data is also reasonable under the conventional standard, the system's understanding depth at this moment is different: it recognizes that this value is achieved under adversity and has potential value in indicating the drought resistance of the variety. The system confirms the data is valid, generates the final structured record, and automatically adds the environmental label: {environmental context: "water-high temperature combined stress", data confidence state: "reasonable-inhibited by environment"}.Contrastive display: if rigid verification is used without considering the environment, the system will only use the conventional standard [220, 260] cm for judgment, although no error will be reported, but the key agricultural information that the plant height is measured under drought conditions will be completely lost, and the data value will be greatly discounted. Advanced scene (if data is abnormal): suppose Zhang Gong describes the plant height as "one meter and eight" (180 cm). The system calculates that the value is significantly lower than the lower limit of the stress interval, and the deviation is large, so it is determined to be "doubtful" or "unreasonable". At this time, the system can infer based on the physiological mapping relationship: under such drought, is the extremely low plant height accompanied by other visible symptoms? Further, through voice active guidance, Zhang Gong confirms: "the system detects that the current drought is severe, and the plant height you reported is 180 cm, which is significantly lower than the expected value. Please confirm whether you have observed that the leaves are severely curled or the lower leaves are yellow and withered?" This realizes knowledge-based interactive review.

[0031] In a feasible embodiment, in the field of intelligent collection of agricultural data, the existing data verification technology through environmental perception and preset rules can adapt to the influence of a single environmental factor to a certain extent, but when facing the more common and complex multi-dimensional and dynamic scenes in real agricultural production, there are still significant limitations. In addition to the defect of insufficient modeling of the nonlinear superposition effect of multiple stress factors, the existing method also faces other key challenges in actual application. For example, the system highly depends on the "typical" knowledge model constructed by historical data, which is difficult to effectively identify and process breakthrough phenotypes that exceed the existing cognitive range in the breeding process, and there is a risk of misjudging "innovation" as "error". At the same time, in new planting areas or rare crop varieties with sparse data, the system faces the "cold start" dilemma due to the lack of sufficient prior knowledge, resulting in the failure or significant reduction of the reliability of its intelligent verification function. In addition, the existing framework lacks consideration of the time lag effect and historical influence of environmental stress, which may judge reasonable current phenotypes caused by past events as abnormal.

[0032] In summary, the method further comprises: judging whether there is a pre-defined stress environment type according to the current environment state information, comprising: Parsing at least two coexisting stress factors and their respective intensity values from the current environment state information; Querying a pre-defined multi-factor interaction effect knowledge base to obtain interaction effect coefficients between at least two stress factors; Generating a comprehensive stress index by weighted calculation based on the intensity values of the stress factors and the interaction effect coefficients; Matching the comprehensive stress index with a pre-set threshold range to determine the corresponding stress environment type.

[0033] As mentioned above, the comprehensive analysis of environmental sensor data (such as soil moisture, air temperature, nutrient concentration) identifies multiple environmental factors that simultaneously exceed their respective normal thresholds and pose potential stress to crop growth, and quantifies the degree of deviation for each factor. The complex environmental conditions in the field are finely deconstructed and quantitatively perceived. Instead of looking for a "dominant stress", the system can comprehensively identify and quantify the specific intensity of multiple coexisting factors such as "drought", "low temperature", "nitrogen deficiency", etc., providing accurate input data for understanding complex stress. Access a knowledge base that stores agricultural research results, which defines the possible interactions (such as synergy, antagonism) between different combinations of stress factors in the form of data or rules. The system queries and obtains the corresponding interaction coefficients based on the identified combination of stress factors. The introduction of the ability to recognize the nonlinear interaction between stress factors enables the system to understand that "drought superimposed with high temperature" may be much more harmful than the sum of their individual effects, while "moderate drought and low temperature" may produce some antagonistic effect, thus logically surpassing the simple juxtaposition or weighted average of the primary model. Using an embedded mathematical model, the intensity values of each factor are used as basic input, and the interaction coefficients are used to adjust their weights in the comprehensive calculation or introduce nonlinear terms, and finally a single quantitative index representing the overall stress pressure is calculated. The conversion from multiple independent environmental parameters to an integrated and comparable stress pressure measurement is achieved. The index reflects the "total stress" caused by multiple factors and their interactions on crops, providing a unified and scientific scale for subsequent judgment. The calculated comprehensive stress index is compared with the pre-set index threshold range corresponding to different stress types (such as "mild complex stress" and "severe water-heat stress") in the knowledge base, so that the continuous index value is classified into the most matching discrete stress environment type. The mapping from continuous quantitative environmental stress to discrete operational business labels is completed. This ensures that the subsequent processes (such as calling corresponding reference standards) can be based on the most accurate classification of the current complex stress situation.

[0034] In an embodiment, taking the application scenario of grading a batch of "Sunshine Rose" grapes in a fruit warehouse as an example, the application utilizes the deep understanding capability of a large language model to realize intelligent conversion and verification of complex feature descriptions into structured data in non-growth observation scenarios. User voice input: The quality inspector holds a device and orally describes the inspection results of a batch of grape samples: "The variety is Sunshine Rose, the cluster weight is about 500 grams, but the uniformity of the fruit grains is general, the sugar content is about 18 degrees, there is a little downy mildew spot, and the coloring is good." System action: After receiving the voice data, the system determines the current scenario as "post-harvest quality detection of fruits and vegetables" based on keywords such as "variety", "cluster weight", "uniformity", "sugar content", "disease spot", and "coloring" in the voice recognition text, and calls the pre-defined collection field set of this scenario. This set not only includes numerical fields such as "cluster weight (g)" and "sugar content (Brix °)", but also includes enumeration or level fields such as "fruit grain uniformity", "disease condition", and "coloring level" that need to be determined from descriptive language. AI-driven intelligent conversion and initial data generation: field mapping and semantic correction: the system uses a large language model to perform deep semantic analysis on the recognized text. For example, the colloquial "there is a little downy mildew spot" is accurately mapped to "disease type: downy mildew" and "disease degree: slight" two fields. At the same time, the model can associate the ambiguous description "general" to the standard option "fruit grain uniformity: medium". Numerical and level standardization: the system standardizes "about 500 grams" to "cluster weight: 500g" and "about 18 degrees" to "sugar content: 18.0 Brix °" according to field definitions. For the uncertain expression "coloring is good", the model can determine it as "coloring level: good" in combination with the context and mark the confidence. AI-enabled deep verification and intelligent interaction: cross-field logic verification: the system calls the built-in agricultural knowledge rules of the large language model to verify the initial structured data. For example, the model will infer that if the "disease degree" is "slight", the "overall rating" field should not be "inferior". The system finds that the user has not provided "overall rating", so it triggers a warning for incomplete logic. Reasonability analysis and candidate generation: the system analyzes the knowledge of the "Sunshine Rose" variety characteristics based on the AI model. Assuming that the sugar content of this variety is usually higher than 17 degrees, the reported 18 degrees is reasonable. At the same time, the model can generate intelligent questions for missing or questionable fields, such as: "Based on the sugar content of 18 degrees, coloring of good, and slight disease, the system infers that the overall rating can be 'excellent'. Please confirm. Please also provide the fruit stem freshness." Interactive correction and confirmation: the quality inspector can respond with voice: "confirm the overall rating as excellent, and the fruit stem is green." The system completes data completion and correction through voice interaction to form the final structured data record.In this embodiment, it is shown that in the general agricultural quality inspection scene without involving the growth stage and environmental stress, the present application realizes the accurate conversion of fuzzy, non-standard spoken language descriptions such as "general", "a little", "almost good" into standardized field values by deeply integrating large language models. Using built-in variety characteristics and quality logical rules, data contradictions and omissions are automatically discovered, and intelligent questioning is actively initiated. Through voice dialogue, data confirmation, correction and completion are quickly completed, greatly improving the efficiency and accuracy of complex trait description data entry.

[0035] With reference to Figure 3 In the embodiments of the present application, a voice intelligent-based agricultural structured data acquisition system is also provided, comprising: A receiving module 1 for receiving user-input agricultural field-related voice data; A calling module 2 for determining a current data acquisition scene according to the voice data and calling a pre-defined acquisition field set corresponding to the acquisition scene, wherein the acquisition field set includes the standard name of the field, the data type and the value constraint; An identification module 3 for performing text information identification on the voice data; A generation module 4 for matching the identified text information with the acquisition field set according to the identification result, mapping the expression in the text to the corresponding standard field name, and standardizing the conversion of the numerical value and unit in the text according to the data type and value constraint, to generate initial structured data; A verification module 5 for verifying the initial structured data based on the value constraint in the acquisition field set and the pre-set agricultural knowledge rules; A determination module 6 for determining the final structured data based on the verification result.

[0036] As described above, it can be understood that each component of the voice intelligent-based agricultural structured data acquisition system proposed in the present application can realize the function of any one of the voice intelligent-based agricultural structured data acquisition methods as described above, and the specific structure will not be described again.

[0037] With reference to Figure 4 In the embodiments of the present application, a computer device is also provided, which can be a server, and the internal structure thereof can be as shown in Figure 4The computer device includes a processor, a memory, a network interface and a database connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store monitoring data and the like. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a voice-intelligent-based agricultural structured data acquisition method.

[0038] The processor executes the voice-intelligent-based agricultural structured data acquisition method, which includes receiving user-input agricultural field-related voice data, determining a current data acquisition scene according to the voice data, and calling a predefined acquisition field set corresponding to the acquisition scene, wherein the acquisition field set includes a standard name of a field, a data type and a value constraint; performing text information recognition on the voice data; according to the recognition result, matching the recognized text information with the acquisition field set, mapping expressions in the text to corresponding standard field names, and standardizing conversion of numerical values and units in the text according to the data type and the value constraint, to generate initial structured data; based on the value constraint in the acquisition field set and a preset agricultural knowledge rule, verifying the initial structured data; and based on the verification result, determining final structured data.

[0039] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement a voice-intelligent-based agricultural structured data acquisition method, which includes the steps of receiving user-input agricultural field-related voice data, determining a current data acquisition scene according to the voice data, and calling a predefined acquisition field set corresponding to the acquisition scene, wherein the acquisition field set includes a standard name of a field, a data type and a value constraint; performing text information recognition on the voice data; according to the recognition result, matching the recognized text information with the acquisition field set, mapping expressions in the text to corresponding standard field names, and standardizing conversion of numerical values and units in the text according to the data type and the value constraint, to generate initial structured data; based on the value constraint in the acquisition field set and a preset agricultural knowledge rule, verifying the initial structured data; and based on the verification result, determining final structured data.

[0040] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, databases, or other media in this application and in examples used herein, unless specifically stated otherwise, can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0041] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article, or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article, or method. Without more limitations, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article, or method that includes the element.

[0042] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A voice intelligence-based agricultural structured data collection method, characterized in that, The method comprises: receiving user input agricultural field related voice data; determining a current data collection scenario according to the voice data, and calling a predefined collection field set corresponding to the collection scenario, wherein the collection field set comprises standard names of fields, data types, and value constraints; performing text information recognition on the voice data; mapping the recognized text information to the collection field set, mapping expressions in the text to corresponding standard field names, and standardizing numerical values and units in the text according to the data types and value constraints to generate initial structured data; performing verification on the initial structured data based on the value constraints in the collection field set and preset agricultural knowledge rules; determining final structured data based on the verification result.

2. The voice intelligence based agriculture structured data collection method according to claim 1, characterized in that, The step of performing verification on the initial structured data based on the value constraints in the collection field set and preset agricultural knowledge rules comprises: obtaining the preset agricultural knowledge rules, wherein the agricultural knowledge rules define logical constraint relationships between different collection fields; substituting numerical values of related fields in the initial structured data into corresponding logical constraint relationships for calculation; judging whether the initial structured data satisfies the logical constraint relationships according to the calculation result; in response to a judgment that the initial structured data satisfies the logical constraint relationships, determining that the initial structured data passes the verification; in response to a judgment that the initial structured data does not satisfy the logical constraint relationships, marking or triggering correction of the initial structured data.

3. The voice intelligence based agriculture structured data collection method as claimed in claim 1, wherein, The step of mapping the recognized text information to the collection field set, mapping expressions in the text to corresponding standard field names, and standardizing numerical values and units in the text according to the data types and value constraints to generate initial structured data comprises: matching words in the text information to corresponding fields according to standard names of fields in the collection field set and a predefined association word table; extracting numerical value descriptions and unit descriptions corresponding to each field from the matched text information; converting the numerical value descriptions into standard numerical value formats conforming to the data types according to preset conversion rules; converting or calculating the unit descriptions into defined units for each field in the collection field set; generating values of each field in the initial structured data according to the standard numerical value formats and the converted or calculated units.

4. The voice intelligence based agriculture structured data collection method as claimed in claim 1, wherein, After the step of determining the current data collection scenario, the method further comprises: establishing a physiological index mapping relationship matched to a growth stage or growth cycle of a target crop or target livestock or poultry corresponding to the collection scenario; performing rationality analysis on fields related to physiological or growth states in the initial structured data according to the physiological index mapping relationship; generating candidate correction data based on the physiological index mapping relationship and guiding the user to confirm or correct based on the result of the rationality analysis if the result of the rationality analysis is unreasonable.

5. The voice intelligence based agriculture structured data collection method as claimed in claim 4, wherein, The step of performing rationality analysis on the fields related to physiological or growth state in the initial structured data according to the physiological index mapping relationship comprises: Extracting reference value intervals or reference value characteristics of each physiological state field corresponding to the growth stage or growth cycle from the physiological index mapping relationship; Comparing actual values of the same physiological state field in the initial structured data with corresponding reference value intervals or reference value characteristics; Calculating deviation degrees between the actual values and corresponding reference standards according to comparison results; Judging rationality levels of the actual values based on a preset deviation threshold, wherein the rationality levels comprise at least reasonable, doubtful and unreasonable.

6. The voice intelligence based agricultural structured data collection method as claimed in claim 5, wherein, The step of extracting reference value intervals or reference value characteristics of each physiological state field corresponding to the growth stage or growth cycle from the physiological index mapping relationship comprises: Obtaining current environmental state information associated with the target crop or target livestock and poultry; Judging whether there is a predefined stress environment type according to the current environmental state information, wherein the stress environment type is used to represent a set of environmental conditions that have a specific mode of influence on the growth and development of the target crop or target livestock and poultry; If there is, extracting reference value intervals or reference value characteristics corresponding to the growth stage or growth cycle and the stress environment type from the physiological index mapping relationship; If there is not, extracting reference value intervals or reference value characteristics corresponding to the growth stage or growth cycle and under normal environment from the physiological index mapping relationship.

7. The voice intelligence based agricultural structured data collection method as claimed in claim 6, wherein, The step of judging the rationality level of the actual value based on the preset deviation threshold comprises: Obtaining a predefined stress elasticity coefficient associated with the current stress environment type according to identification information of the target crop or target livestock and poultry; Determining a stress intensity of the stress environment type based on the current environmental state information; Dynamically adjusting the preset deviation threshold according to the stress elasticity coefficient and the stress intensity to obtain an adjusted deviation threshold adapted to the current environment and variety; Judging the rationality level of the actual value based on the adjusted deviation threshold.

8. A voice intelligence based agricultural structured data acquisition system characterized in that, The method comprises: A receiving module configured to receive agricultural field related voice data input by a user; A calling module configured to determine a current data collection scenario according to the voice data and call a collection field set predefined for the collection scenario, wherein the collection field set comprises standard names, data types and value constraints of fields; An identification module configured to perform text information identification on the voice data; A generation module configured to map identified text information to corresponding standard field names according to the results of identification, and standardize conversion of values and units in the text according to the data types and value constraints to generate initial structured data; A verification module configured to verify the initial structured data based on the value constraints in the collection field set and preset agricultural knowledge rules; determining module configured to determine the final structured data based on the result of the checking. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images