Crop full life cycle traceability system based on big data

Abstract

The invention, which relates to the technical field of intelligent agriculture, discloses a crop full-life-cycle traceability system based on big data, comprising an edge sensing calculation unit, a cloud service platform and an alliance block chain network. The edge unit collects multi-modal data, determines a phenological stage, selects an optimal AI model to analyze and identify farming behaviors, triggers compliance verification, generates structured traceability event data and links the structured traceability event data with a hash value; the cloud stores a phenological knowledge graph and an SOP rule base; the block chain solidifies the operation rule through the intelligent contract, and realizes the credible traceability of the crop in the whole life cycle. Through deep fusion and innovative design of edge calculation, artificial intelligence, a block chain and a multi-mode sensing technology, adaptive accurate monitoring and resource optimization are realized, traceability information is improved from behavior records to compliance proof, and objective and quantitative quality connotation is given to traceability data.

Description

Technical Field

[0001] The present invention relates to the field of smart agricultural technology, and in particular to a crop life cycle traceability system based on big data. Background Art

[0002] With consumers' increasing concern for agricultural product quality and safety, and businesses' growing demand for standardized management of agricultural production, full-lifecycle traceability technology for crop production has become a research hotspot in smart agriculture. As a key cash crop, traceability of tea production is particularly crucial. Existing tea traceability technologies typically combine technologies such as the Internet of Things, artificial intelligence, and blockchain to collect data on tea production environments, identify pests and diseases, and record and trace information.

[0003] However, existing technologies still have many flaws. The AI recognition models currently used are often general-purpose and cannot be adjusted according to the morphological characteristics of tea leaves in different phenological periods and the types of susceptible pests and diseases, resulting in poor identification targeting, waste of resources and low efficiency. Agricultural behavior monitoring can only record the behavior itself and cannot determine whether it meets production standards, resulting in a lack of in-depth verification of the quality of the production process in traceability information. In the complex environment of tea gardens, the lightweight models on edge computing devices have poor robustness and large fluctuations in recognition accuracy, while uploading tasks to the cloud will cause network congestion and high latency. In addition, while uploading all raw data to the chain can ensure credibility, it will bring huge storage and transaction pressure to the blockchain, which is costly and impractical. Summary of the Invention

[0004] The purpose of the present invention is to provide a crop life cycle traceability system based on big data, which solves the problems existing in the background technology.

[0005] To solve the above technical problems, the present invention provides a crop life cycle traceability system based on big data, comprising: an edge sensing computing unit for collecting multimodal data of target crops; the multimodal data includes current environmental parameters, historical meteorological data, and tea tree image data; A cloud service platform for storing preset crop phenology knowledge graphs and preset standard operating procedure rule bases; A consortium blockchain network for solidifying operating rules derived from the standard operating procedure rule base in the form of smart contracts; The edge sensing computing unit is also used to: Based on the collected current environmental parameters and the historical meteorological data, and by querying the crop phenology knowledge graph, determining the current phenological stage of the target crop; Based on the current phenological stage, the optimal AI model is selected and loaded from the local model library using the model suitability scoring method; Using the loaded optimal artificial intelligence model, the collected tea tree image data is analyzed online to identify farming behaviors; When the agricultural behavior is identified, a compliance check of the agricultural behavior is triggered; The compliance verification is to call the smart contract in the alliance blockchain network to verify the operator identity, operation time, related materials and geographical location of the agricultural behavior in real time to generate a compliance verification result; It is also used to generate structured traceability event data based on the compliance verification results; The structured traceability event data is associated with the hash value of the multimodal data and recorded together in the alliance blockchain network.

[0006] Preferably, the specific implementation of the model suitability scoring method includes: Determine the preset phenological correlation coefficient; Determine the preset model normalization resource consumption; Multiplying the phenological correlation coefficient by a preset first weight coefficient to obtain a phenological correlation weighted value; Multiplying the model normalized resource consumption by a preset second weight coefficient to obtain a resource consumption weighted value; The resource consumption weighted value is subtracted from the phenological relevance weighted value to generate a model suitability score for deciding which artificial intelligence model to load.

[0007] Preferably, The phenological correlation coefficient is a preset value used to characterize the importance of each artificial intelligence model in different phenological stages; The preset values are predefined for each phenological stage based on the knowledge of agricultural experts; The model normalized resource consumption is obtained by performing offline benchmark tests on each artificial intelligence model on the target edge hardware and normalizing the test results; The first weight coefficient and the second weight coefficient are adjustable parameters preset according to business requirements.

[0008] Preferably, the specific implementation of the compliance check includes: Calling the operator verification function in the smart contract to determine whether the operator's identity is in the preset authorized personnel list; Calling the time verification function in the smart contract to determine whether the operation time is within the preset compliance time window; Calling the material verification function in the smart contract to determine whether the associated material is in the preset allowed use list; If and only if all verification functions return positive results, the compliance verification result is determined to be true; otherwise, the compliance verification result is determined to be false.

[0009] Preferably, the multimodal data collected by the edge sensing computing unit also includes spectral data.

[0010] Preferably, it further comprises a tea quality online evaluation module for extracting a plurality of original quality characteristic values based on the loaded optimal artificial intelligence model and the collected spectral data; The original quality characteristic values include: the bud-to-leaf ratio representing the morphology of the tea buds, the color uniformity representing the color of the leaves, and the spectral absorption peak representing the contents of the tea leaves; Normalizing each of the original quality characteristic values, multiplying them by their corresponding quality weight coefficients, and then summing all the products to generate a tea quality evaluation score; The structured traceability event data is also generated based on the tea quality assessment score.

[0011] Preferably, the quality weight coefficient is obtained through a machine learning regression modeling method, which includes the following specific steps: collecting multimodal feature data of a large number of tea samples and the true sensory evaluation scores given by tea evaluation experts, and using the multimodal feature data as independent variables and the true sensory evaluation scores as dependent variables to train a multivariate linear regression model, and using the normalized regression coefficient of the multivariate linear regression model as the quality weight coefficient.

[0012] Preferably, it is also used to trigger an alarm when the compliance verification result is false and record the farming behavior that caused the verification to be false as an abnormal event.

[0013] Compared with the prior art, the present invention has the following beneficial effects: Through the phenological perception mechanism, the optimal AI model can be autonomously and intelligently selected and loaded based on the crop's biological rhythms, improving the targetedness and accuracy of identification while optimizing resource utilization on edge devices. By introducing intelligent SOP cross-validation through real-time capture of operator identity, operation time, associated materials, and other data, and solidifying rules using blockchain smart contracts, traceability information is upgraded from a simple record of behavior to proof of compliance, enhancing its value and credibility. 3. By combining the crop's biological cycle and production standards with real-time monitoring data, a digital twin is constructed, providing a data foundation for precision management, yield forecasting, and quality optimization, and improving the ability to comprehensively utilize data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can derive other drawings based on these drawings without inventive effort. Figure 1 It is a logic block diagram of the system of the present invention; Figure 2 This is a diagram showing the specific implementation steps of the model applicability scoring method of the present invention; Figure 3 This is a diagram of the specific implementation steps of the compliance verification of the present invention. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] Example 1: See also Figure 1-Figure 3 The present invention provides a crop life cycle traceability system based on big data, including: an edge perception computing unit for collecting multimodal data of target crops; the multimodal data includes current environmental parameters, historical meteorological data and tea tree image data; a cloud service platform for storing a preset crop phenology knowledge map and a preset standard operating procedure rule base; a consortium blockchain network for solidifying the operating rules derived from the standard operating procedure rule base in the form of smart contracts; the edge perception computing unit is also used to: determine the current phenological stage of the target crop based on the collected current environmental parameters and the historical meteorological data, and query the crop phenology knowledge map; based on the determined current phenological stage, A suitability scoring method is provided, which selects and loads the optimal artificial intelligence model from a local model library; uses the loaded optimal artificial intelligence model to perform online analysis on the collected tea tree image data to identify agricultural behaviors; when the agricultural behaviors are identified, triggers compliance verification of the agricultural behaviors; the compliance verification is to call the smart contract in the alliance blockchain network to perform real-time verification on the operator identity, operation time, related materials and geographical location of the agricultural behaviors to generate compliance verification results; the system is also used to generate structured traceability event data based on the compliance verification results; the structured traceability event data is associated with the hash value of the multimodal data, and they are recorded together in the alliance blockchain network.

[0017] A big data-based crop lifecycle traceability system operates through a fully trusted closed loop, from perception and analysis to recording. Edge sensing computing units deployed in tea gardens continuously collect multimodal data, which comprehensively depicts the tea tree's growing environment. These data include current environmental parameters such as real-time air temperature, humidity, and light intensity acquired through environmental sensors, combined with historical meteorological data synchronized from a cloud service platform and tea tree image data captured by high-definition cameras. Furthermore, the cloud service platform, serving as the system's central knowledge and rule base, pre-stores a tea phenology knowledge graph constructed by agricultural experts and a standard operating procedure (SOP) rule base developed by the company. The consortium blockchain network solidifies the core rules in the SOP rule base in the form of smart contracts, ensuring the immutability and transparency of operational specifications. The operating logic of the system starts with the edge perception computing unit, integrates the collected current environmental parameters and historical meteorological data, and queries and compares them with the phenological knowledge map, so as to accurately determine the specific phenological stage of the current tea tree, such as the budding period or the picking period; this judgment forms the basis for all subsequent intelligent analysis; the system does not use a static and fixed analysis model, but immediately starts a model suitability scoring method after determining the phenological stage, and dynamically selects and loads an optimal model from multiple alternative artificial intelligence models stored locally based on the monitoring priority of the current stage; when the loaded optimal artificial intelligence model, such as an image analysis model for bud identification, identifies the agricultural behavior of picking from the real-time image data stream, the system will immediately trigger the compliance verification process; this verification process It is not completed locally or in a centralized cloud. Instead, it is done by calling smart contracts deployed on the alliance blockchain network to conduct strict real-time verification of key information such as the operator identity, current operation time, materials used, and geographic location of the picking operation. After the verification is passed, the system will generate a structured traceability event data containing elements such as verification results, timestamp, and operator information. To ensure the integrity and non-repudiation of the original data, the system further cryptographically associates the hash value of the original multimodal data with this structured traceability event data. This complete record containing event details and data fingerprints is eventually written into the alliance blockchain network, forming a highly reliable, traceable, and information-rich traceability chain, realizing in-depth, dynamic, and reliable monitoring of the entire life cycle of tea production.

[0018] Example 2: The specific implementation method of the model suitability scoring method includes: determining a preset phenological correlation coefficient; determining a preset model normalized resource consumption; multiplying the phenological correlation coefficient with a preset first weight coefficient to obtain a phenological correlation weighted value; multiplying the model normalized resource consumption with a preset second weight coefficient to obtain a resource consumption weighted value; subtracting the resource consumption weighted value from the phenological correlation weighted value to generate a model suitability score for deciding which artificial intelligence model to load.

[0019] The phenological correlation coefficient is a preset value used to characterize the importance of each AI model in different phenological stages; the preset value is pre-defined for each phenological stage based on the knowledge of agricultural experts; the model normalized resource consumption is obtained by performing offline benchmark tests on each AI model on the target edge hardware and normalizing the test results; the first weight coefficient and the second weight coefficient are adjustable parameters preset according to business needs; The essence of the model suitability scoring method lies in establishing a quantitative decision-making mechanism to intelligently schedule the most appropriate AI model for a specific phenological stage. Traditional approaches freeze models once deployed, and this static deployment ignores the dynamic changes in monitoring focus and morphological characteristics of crops throughout their biological cycles. To address the lack of specificity and resource waste associated with static deployment, this scoring method was designed to quantitatively integrate expert knowledge in the field of agronomy with the computing resource constraints of edge devices, thereby achieving intelligent and efficient model scheduling. The quantitative implementation of this score relies on a model suitability score formula: ; It represents the model applicability score, which is a dimensionless real number and is the direct basis for the final decision. The higher the score, the more applicable the model. is a positive integer, which is the model index used to uniquely identify the i-th candidate artificial intelligence model in the local model library; is a positive integer, which is the phenological stage index used to uniquely identify the current growth stage of tea leaves; represents the i-th candidate artificial intelligence model; Indicates the jth current phenological stage, which is determined by the system based on environmental data and phenological knowledge graph queries; Represents the phenological correlation coefficient, which is a dimensionless value between 0 and 1, representing the phenological stage Run the model the importance or necessity of Indicates the normalized resource consumption of the model, which is a dimensionless value between 0 and 1, reflecting the model Computational cost when running on the target edge hardware; and are the first weight coefficient and the second weight coefficient, respectively. They are both positive real numbers and are adjustable parameters preset according to business needs. They represent the importance of phenological correlation and resource consumption in decision-making, respectively, and their sum is usually set to 1, that is, , for easy adjustment; The pre-configuration phase of the system involves rigorous definition of various parameters; the phenological correlation coefficient The determination of the phenological stage depends on the domain knowledge of agricultural experts. The expert team will evaluate the importance of each phenological stage (such as budding stage, leaf expansion stage) and each monitoring task (corresponding artificial intelligence model) in that stage and assign a value between 0 and 1. These corresponding relationships are structured and stored in the phenological knowledge map in the cloud. The model normalizes resource consumption. The determination is completed through rigorous offline benchmark testing. Before model deployment, each candidate AI model is actually run on the target edge hardware, and its key resource consumption indicators, such as floating-point operations per second or average power consumption, are measured using performance profiling tools. After collecting test data for all models, the original consumption values are mapped to the [0, 1] interval using the minimum-maximum normalization method. This normalization process follows the following formula: ; Represents the candidate model Raw resource consumption values (such as average power consumption or floating-point operations per second) obtained from offline benchmarking. Represents the minimum value of the original resource consumption values of all candidate models in the model library; Indicates the maximum value of the original resource consumption values of all candidate models in the model library; This formula ensures that the resource consumption of all models is fairly mapped to the range of 0 to 1 for comparison; The first weight coefficient and the second weight coefficient The system administrator sets it according to the specific business strategy. For example, in the scenario of pursuing the ultimate quality of tea, ; In the scenario where the power of edge devices is limited, you can set ; In a specific application scenario, when the system determines that the current phenological stage is the spring budding period , and there is an aphid recognition model in the local model library and shoot morphology grading model ; According to the preset expert knowledge base, , ; After offline benchmark testing, , If the current business strategy prioritizes tea quality, the weight is set to At this point, the model The score is ;Model The score is ;because , the system will load the decision and run the bud morphology grading model with a higher score The introduction of this decision-making mechanism ensures that the system can dynamically invest limited edge computing resources in the most valuable monitoring tasks at the moment, significantly improving the targeted nature of monitoring and resource utilization efficiency.

[0020] Example 3: The specific implementation method of the compliance verification includes: calling the operator verification function in the smart contract to determine whether the operator identity is in the preset authorized personnel list; calling the time verification function in the smart contract to determine whether the operation time is within the preset compliance time window; calling the material verification function in the smart contract to determine whether the associated material is in the preset allowed use list; if and only if all verification functions return a positive result, the compliance verification result is determined to be true; otherwise, the compliance verification result is determined to be false.

[0021] The system is further configured to trigger an alarm when the compliance verification result is false and record the farming behavior that caused the verification to be false as an abnormal event; The technical essence of implementing compliance verification lies in converting the production management standard SOP from a static document into an automatically executable, tamper-proof on-chain code, thereby providing in-depth compliance proof for traceability information. Traditional traceability systems only record the behavior itself but cannot determine whether it is compliant, resulting in limited information value. To solve this problem, the present invention solidifies the SOP rules in the smart contract of the alliance blockchain, and by designing a set of logically rigorous verification functions, it realizes automated and real-time auditing of on-site agricultural operations, upgrading traceability information from "behavior records" to "compliance proof."

[0022] From a logical perspective, compliance verification is a Boolean AND operation consisting of multiple sub-verification functions. The result is determined by the logic deployed in the smart contract and can be expressed as the following logical formula: ; Indicates the final SOP compliance verification result, whose value is Boolean true (1) or false (0); is the logical AND operator; These sub-verification functions represent operator verification, time verification, material verification, and geographic location verification. These functions are pre-coded in the smart contract, and their internal judgment rules are derived from the SOP document. Indicates the operator's identity, which can be derived from the output of the face recognition model or the unique ID of the RFID badge; Indicates the current operation timestamp, which is provided by the system clock of the edge device calibrated by the NTP protocol and is in the format of Unix timestamp; Indicates the associated material identifier, which is usually derived from the unique product ID obtained by visual model recognition of the pesticide bottle QR code or packaging features; Indicates the geographic location coordinates, which are read in real time by the device's built-in GPS module; When the system detects a pesticide spraying operation, the edge device immediately captures the operator's identity on site. , Current operation timestamp , related material identification and geographic coordinates ; The system then calls the smart contract function and takes these parameters as input; the logic inside the smart contract begins to execute: The function will query Whether the applicant is included in the preset list of authorized agricultural technicians; The function will determine Is it outside the "15-day safety interval before harvesting" specified in the SOP? The function will check Whether it is included in the list of permitted organic pesticide materials; Function verification Whether it is within the designated tea garden plot; Only when all sub-validation functions return true, The final result is true, the system determines that the pesticide spraying behavior is compliant and records it as a compliance event on the chain; on the contrary, if any of the verifications fails, such as the use of pesticides not on the list, resulting in Returns false, then The result is false; at this time, the system will immediately trigger the alarm mechanism, send an alert to the management background, and record this violation together with the specific failed verification items as an abnormal event on the blockchain; through this verification and recording mechanism, the system not only realizes the rigid constraints of automation on the production process and effectively eliminates illegal operations, but more importantly, it gives irrefutable credibility and deep value to each traceability event on the chain, providing a solid technical endorsement for the brand of high-end tea.

[0023] Example 4: The multimodal data collected by the edge sensing computing unit also includes spectral data; The system further includes an online tea quality assessment module for extracting a plurality of original quality characteristic values based on the loaded optimal artificial intelligence model and the collected spectral data; the original quality characteristic values include: the bud-to-leaf ratio representing the morphology of tea buds, the color uniformity representing the color of leaves, and the spectral absorption peak representing the contents of tea leaves; each of the original quality characteristic values is normalized and multiplied by the corresponding quality weight coefficient, and all the products are summed to generate a tea quality assessment score; the structured traceability event data is also generated based on the tea quality assessment score; The quality weight coefficient is obtained by a machine learning regression modeling method, which includes the following specific steps: collecting multimodal feature data of a large number of tea samples and the actual sensory evaluation scores given by tea review experts, and training a multivariate linear regression model with the multimodal feature data as independent variables and the actual sensory evaluation scores as dependent variables, and using the normalized regression coefficient of the multivariate linear regression model as the quality weight coefficient; To incorporate product quality, a core value dimension, into the traceability system, the system integrates an online tea quality assessment module. Traditional traceability systems focus more on the safety of the production process and less on the quality of the final product. The value of tea is highly correlated with quality indicators such as its morphology, color, and inclusions. However, traditional evaluation methods for these indicators rely on expert sensory evaluation, which is highly subjective, unscalable, and difficult to trace. This paper proposes an online evaluation method that integrates multimodal data. Through artificial intelligence models and spectral analysis, an objective and quantitative quality assessment model is established, which converts high-dimensional sensory qualities into recordable and verifiable numerical scores. The quantitative evaluation of tea quality is completed through a quality score fusion formula: ; Indicates the comprehensive quality score of tea, which is a final calculated real number and can be mapped to a range of 0-100 points; is the feature index, a positive integer from 1 to N; It is the total number of original quality characteristic values used for evaluation, and its specific number and types are determined by the domain knowledge of tea review experts; represents the kth original quality eigenvalue, which comes from the direct output of various sensors and artificial intelligence models, for example, The bud-to-leaf ratio may come from the output of the image analysis model, The absorption peak of the spectral sensor in a specific band is used to characterize the content of specific ingredients in tea leaves; Is a standard normalization function, such as minimum-maximum normalization, used to convert the original eigenvalues of different dimensions Map to a uniform interval of [0, 1] to eliminate the dimension effect; Represents the quality weight coefficient of the kth feature, which represents the contribution of this feature to the final quality score. All weight coefficients are positive real numbers and their sum is ; The specific calculation formula is: ; is the measured value of the kth original quality eigenvalue; and are the minimum and maximum values of the reasonable range of values of the characteristic value determined based on a large amount of historical sample data or expert knowledge; During tea picking, when the edge perception computing unit collects images and spectral data of a specific batch of tea, the online tea quality assessment module is activated. Assume that the system is configured with two quality features (N=2): the integrity of the young buds obtained through AI image analysis. (Original value), obtained by spectral analysis of the spectral absorption peak of a certain inclusion (original value); after normalization, we get and ; The quality weight coefficient obtained by machine learning regression modeling is and ; The quality score of this batch of tea is The system multiplies this by 100, yielding a final score of 82. This quality assessment score, along with compliance verification results for that batch of harvested tea and operator information, forms structured traceability event data and is recorded on the blockchain. This design allows consumers to verify the compliance of the production process when tracing products, while also obtaining the objectively assessed quality score of that batch of tea at the time of harvest. This creates a digital twin that reflects the true state of the tea garden, providing a solid data foundation for precise tea management, yield forecasting, and quality optimization. The quality weight coefficient The determination of the weight coefficient is not a subjective setting, but is obtained through a data-driven machine learning regression modeling method; this method ensures the objectivity and scientificity of the weight coefficient. Its core steps are as follows: collect a large number of representative tea samples; for each sample, use the edge perception computing unit of this system to collect its multimodal feature data and extract multiple original quality feature values On the other hand, a number of senior tea evaluation experts were organized to conduct standard sensory evaluation on these samples and give an authoritative real score; then, the collected multimodal feature data were used as the independent variable of the model (feature matrix), and the real sensory evaluation score given by the experts was used as the dependent variable (target vector) to train a multivariate linear regression model; after the model training was completed, the regression coefficients of each independent variable learned by it objectively reflected the degree of influence of each feature on the final sensory score; finally, these regression coefficients were normalized, for example, by dividing by the sum of the absolute values of all coefficients so that the sum is 1, and the result is the quality weight coefficient. ; This normalization process is to convert the original regression coefficient (may be positive or negative, with different dimensions) is converted into a dimensionless weight that represents its contribution to the final quality score; the specific calculation formula is as follows: ; in, is the kth quality feature obtained after training the multiple linear regression model The original regression coefficient of is its absolute value, is the total number of quality features. Using absolute values ensures that the weight are all positive numbers, and the sum of all weights is strictly equal to 1, which makes It can be clearly interpreted as the importance of the k-th feature in the comprehensive quality evaluation.

[0024] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A crop life cycle traceability system based on big data, characterized by: include: An edge sensing computing unit, configured to collect multimodal data of target crops; the multimodal data includes current environmental parameters, historical meteorological data, and tea tree image data; A cloud service platform for storing preset crop phenology knowledge graphs and preset standard operating procedure rule bases; A consortium blockchain network for solidifying operating rules derived from the standard operating procedure rule base in the form of smart contracts; The edge sensing computing unit is also used to: Based on the collected current environmental parameters and the historical meteorological data, and by querying the crop phenology knowledge graph, determining the current phenological stage of the target crop; Based on the current phenological stage, the optimal AI model is selected and loaded from the local model library using the model suitability scoring method; Using the loaded optimal artificial intelligence model, the collected tea tree image data is analyzed online to identify farming behaviors; When the agricultural behavior is identified, a compliance check of the agricultural behavior is triggered; The compliance verification is to call the smart contract in the alliance blockchain network to verify the operator identity, operation time, related materials and geographical location of the agricultural behavior in real time to generate a compliance verification result; It is also used to generate structured traceability event data based on the compliance verification results; The structured traceability event data is associated with the hash value of the multimodal data and recorded together in the alliance blockchain network.

2. The crop life cycle traceability system based on big data according to claim 1, characterized in that: The specific implementation of the model suitability scoring method includes: Determine the preset phenological correlation coefficient; Determine the preset model normalization resource consumption; Multiplying the phenological correlation coefficient by a preset first weight coefficient to obtain a phenological correlation weighted value; Multiplying the model normalized resource consumption by a preset second weight coefficient to obtain a resource consumption weighted value; The resource consumption weighted value is subtracted from the phenological relevance weighted value to generate a model suitability score for deciding which artificial intelligence model to load.

3. The crop life cycle traceability system based on big data according to claim 2 is characterized in that: The phenological correlation coefficient is a preset value used to characterize the importance of each artificial intelligence model in different phenological stages; The preset values are predefined for each phenological stage based on the knowledge of agricultural experts; The model normalized resource consumption is obtained by performing offline benchmark tests on each artificial intelligence model on the target edge hardware and normalizing the test results; The first weight coefficient and the second weight coefficient are adjustable parameters preset according to business requirements.

4. The crop life cycle traceability system based on big data according to claim 1, characterized in that: The specific implementation of the compliance verification includes: Calling the operator verification function in the smart contract to determine whether the operator's identity is in the preset authorized personnel list; Calling the time verification function in the smart contract to determine whether the operation time is within the preset compliance time window; Calling the material verification function in the smart contract to determine whether the associated material is in the preset allowed use list; If and only if all verification functions return positive results, the compliance verification result is determined to be true; otherwise, the compliance verification result is determined to be false.

5. The crop life cycle traceability system based on big data according to claim 1 is characterized in that: The multimodal data collected by the edge sensing computing unit also includes spectral data.

6. The crop life cycle traceability system based on big data according to claim 5 is characterized in that: It also includes a tea quality online evaluation module for extracting a plurality of original quality characteristic values based on the loaded optimal artificial intelligence model and the collected spectral data; The original quality characteristic values include: the bud-to-leaf ratio representing the morphology of the tea buds, the color uniformity representing the color of the leaves, and the spectral absorption peak representing the contents of the tea leaves; Normalizing each of the original quality characteristic values, multiplying them by their corresponding quality weight coefficients, and then summing all the products to generate a tea quality evaluation score; The structured traceability event data is also generated based on the tea quality assessment score.

7. The crop life cycle traceability system based on big data according to claim 6 is characterized in that: The quality weight coefficient is obtained through a machine learning regression modeling method, which includes the following specific steps: collecting multimodal feature data of a large number of tea samples and the true sensory evaluation scores given by tea evaluation experts, and using the multimodal feature data as independent variables and the true sensory evaluation scores as dependent variables to train a multivariate linear regression model, and using the normalized regression coefficient of the multivariate linear regression model as the quality weight coefficient.

8. The crop life cycle traceability system based on big data according to claim 4 is characterized in that: It is also used to trigger an alarm when the compliance verification result is false and record the farming behavior that caused the verification to be false as an abnormal event.

Images