Modern agriculture digital authentication large model

Abstract

The embodiment of the invention provides a modern agriculture digital authentication large model, and belongs to the field of agricultural big data and artificial intelligence, an agricultural data authentication method comprises the steps that data information of a target farm is acquired, and the data information comprises crop information, farm information and production information; according to the crop information, the farm information and the production information, a block chain is constructed, and the block chain comprises a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; block chain evidence storage is carried out on the block chain to obtain an authentication certificate of each crop in the target farm, and the block chain evidence storage comprises production data evidence storage, environmental data evidence storage, management data evidence storage, processing data evidence storage and tactile data evidence storage; and performing data tracking on the crops in the target farm according to the authentication certificate. According to the method, the agricultural yield and value are accurately estimated, and the agricultural planting scheme is optimized.

Description

Technical Field

[0001] The present invention relates to the fields of agricultural big data and artificial intelligence, and in particular to a large model of modern agricultural digital authentication. Background Art

[0002] The currently widely used agricultural product certification method relies primarily on manual operations, such as sampling, record-keeping, and review. This model results in high certification costs and low efficiency, directly impacting the efficiency of agricultural product distribution and the loss rate of fresh produce. Furthermore, insufficient data credibility makes it difficult for financial institutions to provide loans to farmers based on certification data, resulting in low mortgage loan approval rates for farmers. High-quality agricultural products also struggle to command premiums due to the lack of credible certification. Existing IoT data fails to integrate deeply with intelligent analytics. Traditional certification methods rely on manual operations, which are time-consuming, labor-intensive, and prone to errors, increasing production and management costs. Agricultural projects are generally perceived as high-risk, leading to limited support from financial institutions. Consumers lack confidence in the quality of agricultural products, making it difficult to achieve high-quality prices, thus hindering brand building and market expansion. Summary of the Invention

[0003] The purpose of the embodiment of the present invention is to provide a large model of modern agricultural digital authentication, which realizes the accurate estimation of agricultural output and value and optimizes agricultural planting plans.

[0004] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a method for agricultural data authentication, which includes: obtaining data information of a target farm, wherein the data information includes crop information, farm information and production information; constructing a blockchain based on the crop information, farm information and production information, wherein the blockchain includes a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; performing blockchain notarization on the blockchain to obtain authentication certificates for each crop in the target farm, wherein the blockchain notarization includes production data notarization, environmental data notarization, management data notarization, processing data notarization and tactile data notarization; and tracking data of the crops in the target farm based on the authentication certificate.

[0005] Optionally, the main chain includes key data, which includes at least one of the trigger data of field water holding capacity and the characteristic hash value of the pest and disease image; the logistics sub-chain includes the transportation track and the storage environment; the quality inspection sub-chain includes the pesticide detection results and quality ratings; and the production sub-chain includes agricultural operation records and seedling information.

[0006] Optionally, the blockchain notarization of the block chain obtains the authentication certificate of each crop in the target farm, comprising: determining the growth state hash value of the crop according to the main chain and the production subchain; determining the environmental parameter according to the main chain, the logistics subchain and the production subchain, and generating a warning event hash value when the environmental parameter is abnormal; determining the operation trajectory and the correlation hash value of the input according to the logistics subchain, the quality inspection subchain and the production subchain; determining the process chain hash value of each processing batch according to the logistics subchain, the quality inspection subchain and the production subchain; determining the transportation process hash value according to the logistics subchain; and storing the growth state hash value, the warning event hash value, the correlation hash value, the process chain hash value and the transportation process hash value in the chain to obtain the authentication certificate of each crop in the target farm.

[0007] Optionally, the authentication certificate includes crop variety name, grower information, planting time, growth process data, test results, authentication conclusion and risk warning label.

[0008] Optionally, the method further comprises: generating a biometric hash value according to the image information in the data information; constructing a triple according to the biometric hash value, the public key of the authentication certificate and the wallet address; and storing the triple in the blockchain underlying platform for forming a digital identity anchor point of the crop.

[0009] Optionally, the method further comprises preprocessing the data information, comprising: performing cost rule library screening, edge computing and standard formatting processing on the crop information, farm information and production information to obtain qualified data; and performing cleaning and normalization processing on the continuous data in the qualified data to obtain preprocessed data.

[0010] Optionally, the method further comprises: optimizing the crop information, farm information and production information through a deep learning model; the network neck of the deep learning model is provided with an SE attention module for recalibrating channel weights; and the channel shuffling operation of the deep learning model comprises grouping shuffling and point-by-point convolution.

[0011] Optionally, a crop category vector is set for the crop image of the target farm; the convolution kernel weight in the deep learning model is dynamically adjusted through a multi-head self-attention mechanism; the attention degree of the crop in the deep learning model is determined according to the crop category vector and the convolution kernel weight for variety identification and / or pest and disease detection; a path aggregation network is used for adaptive sharpening preprocessing of the crop image for identifying disease spot features; and deep learning model is used to determine soil and meteorological time series data for predicting growth trend and maturity, wherein the meteorological time series data is integrated into the crop growth cycle as a feature node.

[0012] In another aspect, the present application also provides an agricultural data authentication device, comprising: an acquisition module configured to acquire data information of a target farm, the data information comprising crop information, farm information and production information; a first processing module configured to construct a blockchain according to the crop information, the farm information and the production information, the blockchain comprising a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; and a second processing module configured to perform blockchain storage on the blockchain to obtain an authentication certificate of each crop in the target farm, the blockchain storage comprising production data storage, environment data storage, management data storage, processing data storage and tactile data storage, and the data of the crops in the target farm is tracked according to the authentication certificate.

[0013] Optionally, the blockchain storage on the blockchain to obtain the authentication certificate of each crop in the target farm comprises: determining a growth state hash value of the crop according to the main chain and the production sub-chain; determining an environment parameter according to the main chain, the logistics sub-chain and the production sub-chain, and generating an early warning event hash value when the environment parameter is abnormal; determining an operation trajectory and an associated hash value of an input according to the logistics sub-chain, the quality inspection sub-chain and the production sub-chain; determining a process chain hash value of each processing batch according to the logistics sub-chain, the quality inspection sub-chain and the production sub-chain; determining a transportation process hash value according to the logistics sub-chain; and performing on-chain storage of the growth state hash value, the early warning event hash value, the associated hash value, the process chain hash value and the transportation process hash value to obtain the authentication certificate of each crop in the target farm.

[0014] Optionally, the device further comprises a third processing module configured to generate a biometric hash value according to image information in the data information; construct a triple according to the biometric hash value, a public key of the authentication certificate and a wallet address; and perform on-chain storage of the triple on a blockchain underlying platform to form a digital identity anchor point of the crop.

[0015] In another aspect, the present application also provides a machine-readable storage medium having instructions stored thereon, the instructions, when executed by a processor, causing the processor to be configured to perform the above-mentioned agricultural data authentication method.

[0016] The method of agricultural data authentication comprises the following steps: obtaining data information of a target farm, wherein the data information comprises crop information, farm information and production information; constructing a block chain according to the crop information, the farm information and the production information, wherein the block chain comprises a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; performing block chain storage on the block chain to obtain an authentication certificate of each crop in the target farm, wherein the block chain storage comprises production data storage, environment data storage, management data storage, processing data storage and tactile data storage; and performing data tracking on the crops in the target farm according to the authentication certificate. The method realizes real-time collection of full-cycle agricultural data, dynamic value calculation and risk assessment based on a digital agricultural model to generate an authentication certificate, provides a quantifiable credit basis for a financial institution, converts an agricultural production process into a standard digital asset recognizable by a financial institution, provides a right confirmation basis for financial tools such as credit, futures and trust, and realizes accurate estimation of agricultural yield and value.

[0017] Other features and advantages of the embodiments of the present application will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following specific embodiments to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings: Figure 1 is a flowchart of a method of agricultural data authentication of the present application; Figure 2 is a flowchart of data collection of the present application; Figure 3 is a specific embodiment schematic diagram of the present application; Figure 4 is another specific embodiment schematic diagram of the present application; Figure 5 is a smart transaction review flowchart of the present application; Figure 6 is a flowchart of embodiment two of the present application; Figure 7 is a schematic diagram of an agricultural data authentication device of the present application.

[0019] Explanation of reference signs 100 - agricultural data authentication device; 200 - acquisition module; 300 - first processing module; 400 - second processing module. DETAILED DESCRIPTION

[0020] The specific implementation of the embodiments of the present application is described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiments of the present application, and is not used to limit the embodiments of the present application.

[0021] It should be noted that the acquisition, transmission, storage, use, processing and the like of data in the technical solutions of the present application comply with the relevant provisions of laws and regulations. In the embodiments of the present application, some industry existing solutions such as software, components, models and the like may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0022] Embodiment one Figure 1 is a flowchart of a method for agricultural data authentication, as Figure 1 shown, the method for agricultural data authentication comprises: step S101, acquiring data information of a target farm, the data information comprising crop information, farm information and production information.

[0023] Specifically, as Figure 2 shown, the present application can construct a full-range data acquisition network through cooperation of various devices such as cameras, soil sensors, weather stations, mobile terminals / PC terminals and data interfaces, remote sensing and unmanned aerial vehicles. The camera captures the real-time growth of crops in the field (including micro-growth characteristics such as plant height, leaf color and fruit shape), records the appearance state information of agricultural facilities such as personnel and equipment activities in the field, greenhouse, irrigation system and the like, and provides visual image basis for agricultural production; the soil sensor adopts a layered deployment method (such as one layer per 10 cm), continuously collects soil temperature and humidity, pH value, electrical conductivity and nutrient content data such as nitrogen, phosphorus and potassium, and real-time masters the dynamic changes of soil fertility and environment. The weather station acquires meteorological parameters such as air temperature and humidity, light intensity, wind speed and direction, precipitation, air pressure at a predetermined sampling frequency, and lays a foundation for agricultural production environment analysis; the mobile terminal / PC terminal and data interface support manual input of sowing time, fertilizer amount, plant protection measures and other agricultural operation information, and agricultural product sales intention and other data, effectively supplementing the automatic collection of uncovered content; remote sensing uses a satellite platform to carry multi-spectral, microwave and other sensors to dynamically collect macro data such as crop growth, soil humidity and vegetation coverage in a large range of farmland at a threshold time period; the unmanned aerial vehicle carries high-resolution imaging, thermal infrared and other sensors through low-altitude flight capability to obtain high-precision, real-time micro data such as crop diseases and insect pests, water and fertilizer conditions, plant height and leaf area in local farmland plots.

[0024] Various devices realize data interaction through a communication network. A camera collects a video stream and performs compression processing through an edge computing node. Soil sensor data is subjected to sliding window polynomial fitting to remove high-frequency noise. Weather station data is subjected to a time series compression algorithm to reduce transmission volume. Mobile terminal data is synchronized to the cloud in real time through an encryption protocol. Key agricultural operations (such as fertilizer amount = soil EC value x 0.8 + basic value) and seedling information (seed variety, batch number) are stored on a blockchain, and an intelligent contract automatically verifies data compliance, standardizes the agricultural production process, and enables enterprises to replicate high-quality production modes based on on-chain data. Farmers can work according to standardized processes, effectively solving the problem of traditional agricultural experience dominance and disorder. The unalterable nature of the blockchain strengthens brand trust and helps the development of agricultural product branding. Consumers can scan a code to view the entire cycle of data records from planting to harvesting, including daily soil temperature and humidity curves, weekly growth images taken by a drone, and manually entered fertilizer and pesticide records, forming a complete data chain from the field to the table. This provides comprehensive and detailed data support for precision decision-making and scientific management in smart agriculture, and provides a reliable underlying data foundation for agricultural digital certification and financial services.

[0025] The method further includes: generating a biometric hash value according to image information in the data information; constructing a triple according to the biometric hash value, a public key of an authentication certificate, and a wallet address, and storing the triple on a blockchain underlying platform for forming a digital identity anchor point of the crop.

[0026] Specifically, at a key stage of crop growth, a high-resolution camera is used to collect images of the crop, ensuring that its biometric features such as leaf texture and fruit morphology are captured clearly. The collected images are first subjected to grayscale processing to convert color images to grayscale images, reducing data volume while retaining key feature information and simplifying subsequent calculations. Next, a Gaussian filter algorithm is used to denoise the images. This algorithm constructs a Gaussian kernel function to perform weighted averaging on each pixel point in the image, effectively removing noise points caused by factors such as light changes and sensor noise, making the image smoother and clearer and providing a high-quality data foundation for subsequent feature extraction.

[0027] The scale-invariant feature transform (SIFT) algorithm is used to extract the feature points of crops. SIFT algorithm constructs a difference of Gaussians (DoG) scale space to detect extreme points at different scales. These extreme points have high invariance to image rotation, scaling, and illumination changes. For crop images, SIFT algorithm can accurately locate feature points such as leaf edge, vein intersection, and fruit outline. For example, when identifying apple varieties, it can accurately extract feature points such as apple skin spot distribution and fruit stem shape. Then, the extracted feature points are described as 128-dimensional feature vectors, which contain gradient information and direction information of the neighborhood pixels around the feature point, fully and uniquely representing the biological characteristics of the feature point.

[0028] The generated feature vectors are input into the secure hash algorithm (SHA-256). SHA-256 algorithm performs multiple rounds of complex operations on the input data, including logical operations and shift operations, and finally outputs a 256-bit hash value. This hash value is a highly condensed and unique numerical representation of the biological characteristics of the crop. Even if there are minor differences in the collection process (slightly different angles), as long as the biological characteristics remain unchanged, the generated hash value will have high consistency. If the biological characteristics change substantially (crop suffers from pests and diseases leading to significant changes in leaf shape), the hash value will be completely different, providing a reliable basis for subsequent identity verification and traceability.

[0029] Taking tomatoes as an example, during the growth of tomatoes to the early mature stage, professional image acquisition equipment is used to collect multi-angle images of each representative tomato plant according to the standard shooting process. After the collected images are processed by the above-mentioned grayscale and Gaussian filtering, the biological feature points are extracted and converted into 128-dimensional feature vectors, and then the biological feature hash value is generated through the SHA-256 algorithm. For example, for a certain tomato, the generated biological feature hash value is a54f321c9876b543 (256 bits).

[0030] This planting base has obtained agricultural product certification certificates from authoritative agricultural certification agencies. The certification certificates contain public key information, and the public key is 0x123abcdef456 (specific public key value). At the same time, the base has its own wallet address on the blockchain underlying platform, and the wallet address is 0xf123456789abcdef (blockchain wallet address format).

[0031] The triple construction and on-chain storage include: combining the generated biometric hash value, authentication certificate public key and wallet address to construct a triple, i.e. a54f321c9876b543, 0x123abcdef456, 0xf123456789abcdef. Through the API interface provided by the underlying platform of the blockchain, the triple data is encapsulated into a transaction request conforming to the data format of the blockchain, and sent to the blockchain network. The nodes in the blockchain network will verify the transaction request, including checking the data format, signature, etc. After verification, the triple data will be packaged into a new blockchain block, and through the consensus mechanism (Proof of Stake PoS, Practical Byzantine Fault Tolerance PBFT, etc.), a consensus is reached in the entire blockchain network, and finally the on-chain storage is completed. Thereafter, the triple data will be permanently stored in the blockchain and cannot be tampered with, becoming the digital identity anchor point of the tomato throughout its life cycle.

[0032] The biometric hash value, as a unique digital identifier of the crop, has high uniqueness and stability, and is difficult to be forged or tampered with. The decentralized storage and consensus mechanism of the blockchain ensures that the triple data of the on-chain storage is distributed and stored in the entire network, and any single node failure or malicious tampering cannot affect the integrity of the data. Tampering with data on the blockchain requires control of more than half of the nodes, which is almost impossible in practical applications, thereby ensuring the security and reliability of the digital identity information of the crop.

[0033] When the crop needs to be traced or authenticated, the image information of the crop is collected again, the biometric hash value is generated according to the same process, and the corresponding triple data is queried on the blockchain. If the hash values match and the authentication certificate public key is consistent with the relevant authentication agency information, the source, growth process, authentication status, etc. of the crop can be accurately traced, achieving precise traceability. At the same time, based on the openness and transparency of the blockchain, consumers, regulatory authorities and other relevant parties can query this information through legal channels, enhancing the trust in the quality and safety of agricultural products. For example, in the agricultural product market, consumers can scan the traceability code on the packaging of agricultural products to obtain the digital identity anchor point information stored on the blockchain, including the biometric changes during the growth process (associated with image features at different stages through hash values), authentication status, etc., and purchase with confidence.

[0034] The application of this technology promotes agricultural production enterprises to pay more attention to the standardization and standardization of the production process to ensure that the biometric features of the crops can be accurately collected and identified, meeting the authentication requirements. At the same time, through blockchain storage, regulatory authorities can more efficiently supervise the production and authentication of agricultural products, crack down on fake and inferior agricultural products, maintain market order, and promote the entire agricultural industry to develop in a standardized and high-quality direction.

[0035] The method further comprises preprocessing the data information, including: performing cost rule library screening, edge computing and standard formatting processing on the crop information, farm information and production information to obtain qualified data; and performing cleaning and normalization processing on continuous data in the qualified data to obtain preprocessed data.

[0036] Specifically, the data processing and storage layer performs standardization formatting (such as uniform UTC timestamp, JSON-LD data format) and long-term storage on the data preliminarily processed by the edge computing, prepares for subsequent analysis and on-chain of the blockchain, and retains the agricultural production full-cycle data archives, serving as the data hub for the precise operation of intelligent agriculture in the intelligent agricultural scene. On the one hand, through a three-level data cleaning strategy (such as the physical layer using an abnormal rule library to filter out abnormal values such as vegetables with a price > 50 yuan / kg in real time; the statistical layer using the 3σ principle combined with the Isolation Forest algorithm to remove outliers; and the semantic layer using an agricultural economic ontology mapping tool to convert plan texts into value factors), multi-source data such as soil sensors and weather stations are cleaned and integrated, and after standardization formatting, high-quality data sources are provided for intelligent analysis, supporting disease and pest probability analysis based on the improved YOLOv9 model, crop growth trend prediction based on the Transformer+LSTM hybrid network, and the like, helping to improve the precision of production decisions such as irrigation (generating irrigation instructions within 100 ms when the soil moisture < 60% field water holding capacity) and fertilization (nitrogen fertilizer application amount = soil EC value x 0.8 + basic value), and promoting the transition from experience-based planting to data-driven. On the other hand, the distributed file system stores the agricultural production full-cycle data for a long time, builds a production database of soil spectrum and growth images, and provides a data basis for regional customized irrigation strategy model training based on GNN and new crop recognition model optimization based on meta-learning+GAN, and in combination with the local processing capability of edge computing, realizes continuous data collection and storage in extreme environments, guarantees the stable operation of intelligent agriculture, and improves production efficiency and quality.

[0037] The present application uses the IoT gateway as the core hub for data aggregation and protocol conversion, adopts a multi-channel design (supporting Modbus, MQTT, LoRa and other protocols), realizes real-time aggregation of multi-type sensor and equipment data such as soil sensors, weather stations and 4K cameras, guarantees stable transmission of data to edge computing nodes through a dynamic load balancing algorithm, and the edge computing nodes are locally based on Intel NUC hardware platforms and carry lightweight processing engines to preliminarily process data such as filtering abnormal values and real-time calculation of crop growth environment indicators (such as accumulating the photosynthetically active radiation value through PAR photosynthetically active radiation meter data), effectively reducing the cloud computing power pressure.

[0038] Data transmission and edge computing through the edge-cloud collaborative architecture of NB-IoT / 4G and LoRaWAN hybrid networking, break through the bottleneck of high delay of traditional cloud processing, realize the real-time transmission of key data (such as soil humidity < 60% field water holding capacity trigger data, pest image feature hash value), off-peak transmission of non-sensitive data (such as high-definition video stream) (transmission during the low bandwidth utilization period at night), at the same time, the edge computing device integrates hardware-level TEE encryption isolation technology, and the key data such as irrigation instructions are encrypted by SM4, to prevent data tampering (such as intercepting and warning illegal modification of irrigation time), which adopts IP67 protection design and can continuously operate in extreme environment of-40℃ to 85℃, and is suitable for multiple scenes such as greenhouse and open air, providing stable data support for intelligent agriculture, promoting the transformation of production mode from experience planting to data-driven, for example, a greenhouse realizes early warning of tomato gray mold through the architecture, reducing prevention and control cost.

[0039] Data transmission transmits production subject information and device data stably through SSL / TLS1.3 protocol, and edge computing node preprocesses data through three-level data cleaning (physical layer rule filtering, statistical layer outlier elimination, semantic layer ontology mapping), ensures data authenticity and integrity, lays foundation for production subject digital identity authentication, solves identity verification problem in circulation, its low delay feature helps real-time processing of authentication request in transaction and sharing scene, cooperates with large model and authentication center, shortens identity and ownership authentication time, improves agricultural digital authentication efficiency, and data processed by edge computing promotes its circulation in authentication system, and financial institutions can evaluate credit risk according to environmental stability index on the chain, promote data assetization, and through privacy computing technology of edge computing, help agricultural innovation and drive high-quality development of industry.

[0040] The method further comprises: optimizing the crop information, farm information and production information through a deep learning model; the network neck of the deep learning model is provided with an SE attention module for recalibrating channel weights; and the channel shuffling operation of the deep learning model comprises group shuffling and point-by-point convolution.

[0041] Specifically, MobileNetV3-Large is used as a basic skeleton to reduce the number of parameters and the amount of calculation, and an SE attention module (squeeze and excitation mechanism) is introduced in the network neck to recalibrate the channel weights (enhance the pest feature channel weight to 1.2 times) while maintaining the accuracy of aphid identification on the image. For the ShuffleNetV2 architecture, the channel shuffling operation is optimized to "group shuffling + point-by-point convolution", reducing the calculation loss of cross-group information interaction, improving the model inference speed under the premise of ensuring 91% crop variety recognition rate, and meeting the real-time requirements of edge devices.

[0042] In the ShuffleNetV2 original channel shuffle, channel randomization can easily cause cross-group information redundancy interaction. Group shuffle groups crop features (leaf texture, fruit shape channels into a group) according to crop feature relevance, divides the total channel number C into G feature groups (agricultural scene suggests (G=4-8), adapts to disease, pest feature dimension), and only rearranges channels within the group. The formula is simplified as: ; wherein, is the gth feature group, is the channel shuffle within the group, avoiding irrelevant features (soil background and crop disease spots) across groups, reducing computational redundancy, and GroupShuffle is a custom function name for grouping, shuffling, and splicing agricultural image feature tensors. Concat is a tensor splicing operation along the channel dimension, which merges multiple grouped and shuffled sub-tensors. G is the total number of groups, which can be adjusted according to agricultural task requirements. Q is the index of the group, from 1 to G, representing the qth group (for example, when G=4, q=1, 2, 3, 4 correspond to four groups).

[0043] After group shuffling, pointwise convolution is introduced to do linear transformation between channels: ; wherein, is a 1x1 convolution kernel, and in the agricultural scene, the channel compression ratio is set to (0.5-0.7) to balance the amount of calculation and feature preservation. Its role is to compensate for the isolation between groups after group shuffling by fusing leaf disease spots and environmental light and other crop features with only 9 times the low computational cost of a 3x3 convolution. Second, compress 128-channel high-dimensional features to 64-96 channels to adapt to edge device memory. Third, for aphid, downy mildew, and other feature channels, adjust the weights through pre-training to strengthen the disease spot feature response. In the experiment, the aphid channel activation value increased by 35%, is the feature tensor output by pointwise convolution, which is the result of 1x1 convolution transformation after group shuffling, and is used for subsequent model layers. B is the bias term, which is used to fine-tune the output of pointwise convolution and compensate for the offset of linear transformation, so that the feature distribution is more suitable for agricultural task requirements, and is used to optimize agricultural planting schemes.

[0044] For example, using the INT8 symmetric quantization scheme, 32-bit floating-point weights are mapped to the [-128, 127] integer range, reducing quantization error, compressing model size, and reducing edge device memory occupancy. Based on L1 regularization, the convolution layer channels are pruned, and the top 60% of channels are retained (the gradient contribution of the channel to the loss function is calculated through Taylor expansion), which improves the inference speed of the pruned model and reduces the error in the soil EC value prediction task.

[0045] The crop image of the target farm is set with a crop category vector; the convolution kernel weight in the deep learning model is dynamically adjusted through a multi-head self-attention mechanism; the attention degree of the crop in the deep learning model is determined according to the crop category vector and the convolution kernel weight, which is used for variety identification and / or pest and disease detection; an adaptive sharpening preprocessing is performed on the crop image by using a path aggregation network, which is used for identifying disease spot features; soil and meteorological time series data are determined by a deep learning model, which are used for predicting growth trend and maturity, and the meteorological time series data are integrated into the crop growth cycle as feature node encoding.

[0046] Specifically, a crop category embedding vector (such as a solanaceae crop vector [0.8, 0.2, 0.1] and a cruciferous vector [0.2, 0.7, 0.3]) is introduced, and the convolution kernel weight is dynamically adjusted through a multi-head self-attention mechanism, so that the model improves the difference in attention degree to the leaf features of different crops. In the image, the frost spot recognition accuracy across crops reaches 95.3%.

[0047] The multi-head self-attention mechanism includes: the model sets 4-8 parallel attention heads (adapted to the crop feature dimension), and each head focuses on capturing different types of agricultural features. For example, some heads focus on leaf edge texture (distinguish solanaceae compound leaves from cruciferous single leaves), some heads pay attention to disease spot color and shape (light brown spots of downy mildew and black aggregation features of aphids), and some heads are specially associated with crop growth stage features (difference in leaf size between seedling stage and mature stage). Each attention head independently calculates attention weight, and then fuses multi-dimensional feature association through splicing.

[0048] The weight matrix (dimension matching the number of convolution kernels) output by the attention mechanism is element-wise weighted with the original convolution kernel. For the feature channel with high attention degree (disease spot channel), the weight of the corresponding convolution kernel is scaled to 1.2-1.5 times (to strengthen the feature extraction capability); for the background channel with low attention degree (soil, weeds), the convolution kernel weight is compressed to 0.5-0.8 times (to suppress interference). For example, in the detection of downy mildew, the model identifies the light brown spot on the back of the leaf as a key feature, and the weight of the corresponding convolution kernel is dynamically increased, so that the response value of the disease spot area is enhanced by more than 40%.

[0049] For example, the application adopts a PANet feature pyramid, adds horizontal connections and top-down / bottom-up paths between C3, C4 and C5 feature layers output by the backbone, enhances the feature extraction capability of small targets, and increases the small target detection recall rate to 89%. Adaptive sharpening preprocessing is performed on the leaf image to highlight the edge features of the disease spot, and the recognition accuracy of early gray mold (disease spot area <5% of the leaf) is increased by 12 percentage points. The Transformer module optimizes the growth trend prediction and maturity assessment for soil and meteorological time series data. Through adaptive position encoding in the agricultural cycle, the position encoding formula is optimized as follows: where SEASON(t) is the solstice encoding (spring equinox = 0.2, summer solstice = 0.6), which integrates the characteristics of the crop growth cycle, reduces the tomato maturity prediction error from 3.2 days to 1.8 days; At the same time, a gated recurrent unit (GRU) is added to the Transformer decoder, which uses a long-term attention gated mechanism to assign high attention weights (0.7) to the meteorological data (temperature, light) in the past 7 days and low weights (0.3) to the data 14 days ago, effectively solving the information dilution problem in long-term sequence dependence, greatly improving the R 2 value of yield prediction, PE(t) is the position encoding, t is the natural time step, recording the number of days accumulated from the start of crop growth or the observation starting point, representing the continuous time sequence. i is the feature dimension index, which generates position encoding components of different frequencies. D is the feature dimension of the model, which determines the complexity of position encoding and feature representation.

[0050] Step S102 is to construct a blockchain according to the crop information, farm information and production information, wherein the blockchain includes a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain.

[0051] According to a specific embodiment, the main chain includes key data, the key data includes at least one of trigger data of field water holding capacity and feature hash value of pest image; the logistics sub-chain includes transportation track and storage environment; the quality inspection sub-chain includes pesticide detection result and quality rating; and the production sub-chain includes farming operation record and seedling raising information.

[0052] According to a specific embodiment, as shown in Figure 3 In the agricultural digital authentication scene, the data processing and storage layer is the cornerstone of the construction of agricultural digital authentication credibility. Through standardized formatting processing of agricultural production subject information (such as farmer ID, enterprise qualification) and farming operation data (such as sowing time, fertilizer amount), the SHA-256 hash algorithm is used to ensure data authenticity and integrity, providing reliable data source for production subject digital identity authentication, and solving the problem of identity verification in circulation.

[0053] The present application can process data by using a DeepSeek large model, as shown in Figure 4 The preprocessing branch prepares for accurate analysis by cleaning data and normalization processing (such as using a three-level data cleaning strategy, establishing an agricultural economic data anomaly rule library including rules such as considering vegetables with a price of more than 50 yuan / kg as abnormal and considering logistics costs more than 3 times the regional average as abnormal at the physical layer, using a rule engine for real-time filtering to achieve an abnormal data identification rate); the statistical layer first removes data outside the mean ± 3 times the standard deviation for continuous data using the 3σ principle, and then uses the isolated forest algorithm to consider samples with a tree depth of more than 20 as outliers to improve the accuracy of outlier identification; the semantic layer uses an agricultural economic ontology mapping tool to convert plan text into value factors such as the premium of products corresponding to green certification rewards, performs semantic consistency verification through knowledge graph reasoning to reduce the dimensionality of data features and improve model training efficiency, and normalizes soil nutrient data and other data of different dimensions to convert them into comparable indicators.

[0054] Specifically, the present application determines crop growth trends, environmental stress, and pest occurrence probability based on the crop information, farm information, and production information, and uses a three-layer bidirectional LSTM network with 256 memory units per layer and a Layer Normalization layer to solve the gradient disappearance problem for agronomic value time series modeling. The input includes historical yield, accumulated temperature, precipitation, sunshine, and other meteorological factors and crop phenology data at a 12-month weekly granularity. The gating mechanism is used to capture key time series characteristics during the growth period, such as during the rice filling period (July-August), the model uses the formula Wherein, γ is the sensitive coefficient of the growth period, and f(t) is the phenology indicator function. The temperature factor weight is automatically promoted, and when the historical market price fluctuation is detected to be over the threshold value, the cross-year pattern matching algorithm is triggered, the DTW dynamic time warping is used to calculate the similarity between the current and historical sequences (if the similarity is > 80%, the corresponding year weight coefficient is reused, and the response delay is controlled within 100 ms), in addition, the image and environmental data are combined to warn the occurrence probability of pests and diseases, such as identifying aphids, downy mildew spots, etc. in 224x224 pixel images through the improved YOLOv9 model, early warning, combining GNN graph network modeling of crops, environment and pathogenic bacteria, simulating the pathogenic bacteria diffusion path through the message passing neural network, and based on the Transformer+LSTM hybrid network, the agricultural cycle characteristics such as solstice and circadian rhythm are fused through adaptive position encoding, the prediction error of key nodes such as maize jointing stage and rice tillering stage is controlled within the predetermined time, the soil moisture and light data are combined to build a density, ventilation and disease correlation model, and the best planting spacing is automatically recommended (such as the incidence of powdery mildew is reduced by 28% when the tomato row spacing is optimized from 1.2 meters to 1.5 meters), and through GNN, a variety of environmental and management knowledge graphs are built, the correlation coefficient of nitrogen application amount and yield is quantified to 0.78, and the organic fertilizer ≥500 kg / acre and other agricultural rule constraints are embedded, the multi-task learning framework is used to combine soil spectrum data and meteorological factors, the yield estimation error is reduced, and the environmental adaptability and other multi-dimensional feature vectors are generated through the capsule network to dynamically analyze the potential impact of extreme weather on yield.

[0055] The branch output specific operation instruction is executed to connect intelligent decision-making with actual production, such as when the soil humidity < 60% of the field water holding capacity, the instruction of opening the electromagnetic valve for 20 minutes of irrigation is generated within 100 ms, which is sent to the intelligent valve through the LoRa wireless module, and the edge node confirms the instruction reception through the relay state feedback before the valve executes, and the instruction is automatically re-sent if the response time is exceeded. After irrigation is completed, the decision effectiveness is verified based on the soil humidity recovery curve (such as the humidity rising to more than 70% within 2 hours), and manual intervention is triggered in case of abnormality. For example, when the AI identifies that the number of pests > 50 per trap, the pest control lamp and unmanned aerial vehicle pesticide spraying linkage are automatically triggered, the response time < 10 minutes, and the analysis results can be used to generate fertilizer formulation adjustment suggestions such as nitrogen application amount = soil EC value x 0.8 + base value.

[0056] The intelligent decision system integrates the analysis results to generate production schemes suitable for local farmland, realizing intelligent decision-making in agricultural production. In the intelligent agricultural scene, as the core engine for realizing precision and intelligent production, it combines deep analysis of multi-source data with deep learning model algorithms to accurately predict crop growth trends, disease and pest occurrence probabilities, and environmental stress impacts, such as predicting the impact of high temperature and drought on yield and generating response strategies. Based on the analysis results, it automatically issues instructions to precisely control irrigation, fertilization, plant protection, and other equipment, achieving efficient use of resources. For example, it can improve fertilizer utilization rate and enhance crop yield and quality. In addition, through continuous learning and optimization, it continuously improves production strategies, promoting the transition of agricultural production from experience-driven to data-driven. For example, it can provide early warning of tomato gray mold, reducing yield prediction errors and assisting in the development of agricultural modernization.

[0057] Step S103 is to store the blockchain in the blockchain to obtain the authentication certificate of each crop in the target farm. The blockchain storage includes production data storage, environmental data storage, management data storage, processing data storage, and tactile data storage.

[0058] According to a specific embodiment, the blockchain storage of the blockchain to obtain the authentication certificate of each crop in the target farm includes: determining the growth state hash value of the crop according to the main chain and the production sub-chain; determining the environmental parameter according to the main chain, the logistics sub-chain, and the production sub-chain, and generating a warning event hash value when the environmental parameter is abnormal; determining the operation trajectory and the associated hash value of the input according to the logistics sub-chain, the quality inspection sub-chain, and the production sub-chain; determining the process chain hash value of each processing batch according to the logistics sub-chain, the quality inspection sub-chain, and the production sub-chain; determining the transportation process hash value according to the logistics sub-chain; and storing the growth state hash value, the warning event hash value, the associated hash value, the process chain hash value, and the transportation process hash value in the chain to obtain the authentication certificate of each crop in the target farm. Specifically, the authentication certificate includes crop variety name, grower information, planting time, growth process data, test results, authentication conclusion, and risk warning label.

[0059] By integrating long-term storage of processed data through an object storage system and blockchain technology (using the FISCO BCOS consortium blockchain with a single-chain throughput of ≥10,000 TPS), key agricultural production data, such as land ownership and historical yield and quality benchmarks, stored on the main chain, along with agricultural product transportation data (origin, transit points, destination) and storage environment data (temperature, humidity, and storage duration) recorded on the logistics subchain, quality control data such as pesticide residue test results and quality ratings (such as sweetness grading) stored on the quality inspection subchain, and agricultural operation records (such as fertilization formula and irrigation frequency) and seedling information (variety and time) recorded on the production subchain, are stored on the chain, forming an immutable agricultural data asset archive. Financial institutions can use this on-chain data to assess farmers' creditworthiness and shorten loan approval cycles. Regulatory authorities can verify the entire process data of agricultural products to achieve quality supervision and traceability; e-commerce platforms can carry out tiered sales based on on-chain quality rating data, support the application of agricultural digital certification in finance, e-commerce and other scenarios, ensure the security of data circulation through encryption algorithms, and promote the assetization of agricultural data and industrial collaboration. The blockchain module uses distributed storage and encryption technology to generate tamper-proof digital identity cards for agricultural production entities, ensure clear data ownership and traceable flow in data sharing and transaction links, support farmers, banks, regulators and other parties to participate in maintaining data ledgers, provide a decentralized, transparent and trustworthy operating environment for the agricultural digital certification system, and promote the digital transformation of the agricultural industry. For example, a certified tomato can achieve full-process traceability from sowing (seed RFID tags are associated with blockchain traceability codes) to processing through this system, and consumers scan the code to obtain the certification report of the data node.

[0060] Step S104 is to track data of the crops in the target farm according to the certification certificate.

[0061] According to a specific embodiment, Figure 5 As shown, through multi-layer verification mechanisms such as biometrics (such as fingerprints and faces), authentication certificates and blockchain address binding, a unique and trusted digital identity is established for farmers, agricultural enterprises and other entities, ensuring the trustworthiness of agricultural data on the chain (such as production data, logistics data, etc.) and financial transactions (such as loans and insurance). At the same time, the financial service interface is based on the production data (including historical output, quality ratings, etc.), logistics data (transportation trajectory, storage temperature and humidity), and quality inspection data (pesticide residue detection results, quality grading) stored in the blockchain, and uses the random forest algorithm to provide a basis for credit assessment for agricultural entities, supporting financial innovations such as order pledge loans and weather index insurance (such as automatic claims when the cumulative duration of high temperature exceeds the threshold).

[0062] The device permission management (such as sensor data collection permission classification) and data authenticity verification (SHA-256 hash chain comparison) ensure the credibility of the whole process of production data, for example, to ensure that the EC value collected by the soil sensor has not been tampered with, to provide reliable input for the intelligent decision system, and to innovate weather index insurance products based on authenticated real data (such as when the greenhouse temperature > 35℃ for 24 hours, the smart contract automatically triggers the claim), while based on the order pledge loan mode, reducing the approval cycle, using the tamper-proof characteristics of blockchain to monitor credit risks in real time to promote the intelligent and industrialization of agricultural production, and to improve the loan pass rate of farmers.

[0063] In the agricultural digital certification scenario, digital certification and financial services constitute the key pillars of the trust ecosystem. Digital certification uses hardware security module (HSM) to store the subject's private key, and combines zero-knowledge proof technology to realize anonymous verification of identity information, establishing a unique trusted digital identity for agricultural products and transaction behavior. Consumers can trace the whole process data from RFID tags (associated with variety registration information) to processing and sterilization temperature by scanning the code, solving the problem of subject identity verification. Financial services rely on the certification system, based on the soil fertility index and device utilization rate data assets stored in the blockchain, to build a risk control model (such as risk index = environmental stability x 40% + management compliance x 35% + quality rating x 25%), to realize accurate credit assessment. Agricultural enterprises can obtain special loans with a lower interest rate based on their continuous twelve-month certification compliance records, and package future income rights of certified farmland into ABS products, promoting the deep integration of agricultural data assetization and financial services, and driving the digital transformation of the agricultural industry. For example, a certified strawberry realizes terminal market premium through this system and obtains automatic supply chain finance loans.

[0064] Aggregating data from various sub-chains of the blockchain, the platform supports consumers and regulatory authorities to query the whole process information of agricultural products from "farmland to table" (such as production environment, production process, quality inspection results), helping to improve quality supervision and brand building. In the context of smart agriculture, the traceability platform is an important tool for optimizing production management and improving industry competitiveness. It integrates soil environment, weather conditions, and agricultural operations data to help farmers and enterprises trace the production process and analyze historical data to optimize planting and breeding strategies, such as improving plant protection programs by reviewing pest control records. At the same time, platform data can support the iteration of intelligent decision systems, providing historical reference for precise irrigation and scientific fertilization, improving resource utilization efficiency. In addition, the platform can also show consumers information about the growth environment of agricultural products and pesticide use, enhancing product market trustworthiness, helping to create high-quality agricultural brands, and improving the added value and market competitiveness of agricultural products.

[0065] In the agricultural digital certification scenario, the traceability query platform is the core hub of building a trusted agricultural ecosystem. It visualizes the certification data stored in the blockchain, such as the identity of the production subject, the quality inspection report of agricultural products, and the logistics track, enabling regulatory authorities to quickly verify product compliance and promptly identify data falsification risks, thereby ensuring the authority of the certification system. For consumers, they can verify the digital identity of agricultural products by scanning the code and obtain information from the farm to the table, enhancing their trust in the certification results. In the financial sector, financial institutions can trace the production and operation data of enterprises through the platform, assess credit risks, and provide more accurate financial services to agricultural entities, promoting the digitalization and trustworthiness of the agricultural industry.

[0066] Embodiment Two As shown in Figure 6 , the present application constructs a complete ecological system from physical perception to digital application through multi-dimensional technology integration and full-chain data value mining. In terms of technical implementation, the perception layer devices form a three-dimensional monitoring network: the ground monitoring system deploys a high-definition camera array, which realizes machine vision analysis of growth characteristics such as plant height, leaf color, and fruit shape through OpenCV and AI models, while also serving as a security monitoring function; soil sensors are layered and buried at threshold intervals, integrating temperature, humidity (pH value, EC value, and NPK nutrient content), and other sensing units, which convert soil physical and chemical properties into electrical signals through physical and chemical principles and are regulated by a local controller; the weather monitoring system collects microclimate data such as air temperature and humidity, light intensity, wind speed and direction, precipitation, and atmospheric pressure at a fixed frequency; the low-altitude remote sensing system uses unmanned aerial vehicles equipped with multispectral / thermal infrared cameras and laser radars to identify pests and diseases, monitor crop lodging, and measure plant height; the macro monitoring system relies on multispectral / high spectral / microwave sensors of satellites or large unmanned aerial vehicles to obtain macro data such as NDVI vegetation index and soil moisture in a large range of farmland at a preset period; the human-computer interaction system supports farmers to input structured management data such as sowing time and fertilizer amount through mobile App and PC platform, supplementing the information gap of automated devices.

[0067] The edge intelligent computing layer deploys a lightweight engine to implement three levels of data cleaning: physical layer filters out out-of-range data, statistical layer removes outliers through IQR method, and semantic layer checks logical rationality based on agricultural knowledge graph; at the same time, it calculates key parameters such as DLI (photosynthetically active radiation integral) and soil water potential threshold in real time, and extracts pest and disease image features using lightweight models. The security mechanism integrates trusted execution environment and federated learning framework to ensure the security of data processing and cross-domain collaboration, with response delay controlled within a set time, and non-sensitive data using staggered transmission strategy.

[0068] The blockchain storage link selects FISCOBCOS alliance chain, and the key business data such as fertilization amount calculation formula and sowing variety batch are stored and verified on the chain. The compliance of data such as fertilization amount is automatically verified by smart contract, so as to realize standardization of agricultural process and credibility of operation. In the data value application layer, the precise agricultural decision integrates soil humidity, ET evaporation transpiration and other data to realize precise irrigation, combines soil NPK and crop spectrum data to dynamically adjust the fertilization scheme (such as the on-chain formula of fertilization amount = soil EC value x 0.8 + basic value), uses unmanned aerial vehicle image and AI model to early warn tomato gray mold and other diseases and pests, and optimizes the environment regulation and control of greenhouse through data. The agricultural product traceability integrates the whole cycle data of sowing, growth and picking, consumers can query the unforgeable whole-chain information by scanning the code, and a strawberry brand realizes premium by this means, and provides electronic proof for organic certification and the like. In the field of agricultural finance, the data such as environmental stability index and agricultural compliance rate improve the loan passing rate of farmers, reduce the non-performing rate, and provide objective basis for insurance pricing and loss determination. In addition, the production data of successful farms on the chain promote the replication of high-quality modes, and the cleaned data is assetized and circulated for farm rating, realizing the whole-chain empowerment from technical perception to industrial value.

[0069] The data transmission and edge computing module is a key hub connecting physical agriculture and digital agriculture, and its technical architecture integrates heterogeneous data aggregation, edge intelligent processing, secure transmission and digital authentication and other multi-level technologies, forming a complete closed loop from data collection to value release. In the aspect of heterogeneous data aggregation, the IoT gateway adopts modular protocol stack design, integrates multiple communication protocols, and realizes unified data modeling through protocol abstraction layer: the 224 band spectrum data of soil sensor is converted to JSON format after LoRaWAN transmission, the temperature and humidity data sampled by weather station at 1 Hz is parsed through Modbus-RTU and time stamp is added, and the 4K H, 265 video stream at 25 fps is accessed and metadata is extracted through RTSP protocol. The protocol analysis engine of Eclipse Kura framework is built in the gateway, which manages high-concurrency data flow through ring buffer, and a single gateway can support more than 500 node access; at the same time, the priority scheduling algorithm based on QoS is adopted, and the key data such as soil humidity alarm is set as the highest priority to realize real-time direct connection to edge node, and 4K video stream realizes night peak shifting transmission through bandwidth prediction model, combined with forward error correction coding to ensure that the packet loss rate is less than 0.1%, and automatically switches to NB-IoT / 4G backup link when the network is congested.

[0070] The data processing link implements three-stage cleaning: for soil temperature and other data streams, the IQR quartile range method is applied according to the set time window, Q1 and Q3 are calculated, and abnormal values (such as ±10℃ jumps) outside the range of [Q1-1.5×IQR, Q3+1.5×IQR] are removed; PAR photosynthetically active radiation is collected by a spectrometer calibrated by NIST standard light source, and the linear compensation algorithm is used to control the DLI daily cumulative light integral error to be <3%. In terms of transmission optimization, the meteorological station uses a rotating door algorithm to compress the data, only the inflection points and deviation points are retained, and the compressed point data is realized. Key data are transmitted through the edge node of the special network, and non-sensitive data are transmitted through LoRaWAN in the low bandwidth valley period at night, and the edge node is deployed with a dual-stack communication agent to realize automatic data routing. In terms of security protection, the key data such as irrigation instructions are encrypted by SM4-CTR mode, the key is stored in the security area, and the instruction hash value is checked in real time to prevent tampering. The data-driven agricultural digital authentication system constructs a trusted data chain through three-stage cleaning: the physical layer filters hardware failure data outside the pH value range of 0-14, the statistical layer eliminates random noise by combining IQR and sliding window smoothing, and the semantic layer verifies the logical rationality based on the crop growth rule library. Through the deep coupling of protocol fusion, edge computing, security encryption and blockchain storage, the application not only solves the low power consumption, low delay and anti-interference requirements of agricultural data transmission, but also provides core support for precise farming, financial risk control, brand value-added and other scenarios through data cleaning and trust processing, forming a technical closed loop of sensing, transmission, calculation and application, and promoting the paradigm shift of agricultural production from experience-driven to data-driven.

[0071] Large files can be stored using the Ethereum public chain combined with IPFS instead of a consortium chain. This solution has the characteristics of low deployment cost and high flexibility, and is suitable for small and medium-sized agricultural cooperatives or individual farmers. On the public chain, data authentication and management are realized through smart contracts, and large files are stored using the distributed storage characteristics of IPFS to reduce data storage costs.

[0072] For edge devices with limited computing resources (such as field gateway devices), the agricultural digital authentication large model can be improved to be lightweight, using lightweight neural network architectures such as MobileNet, ShuffleNet, etc., to reduce the number of model parameters and computational complexity. Through model compression and quantization techniques, the model parameters are reduced by more than 60%, allowing it to run on edge devices, enabling local real-time analysis, reducing dependence on cloud servers, and improving the real-time performance and reliability of authentication.

[0073] The application can also use an ensemble learning method instead of a single large model, such as combining multiple different types of machine learning models (such as random forests, support vector machines, etc.) for authentication analysis. Each model is responsible for processing a specific type of data or task, and the results of each model are integrated through a voting mechanism or weighted average, etc. to improve the accuracy and robustness of authentication. In the field crop planting scene, unmanned aerial vehicle inspection can be used instead of fixed cameras for data collection. The unmanned aerial vehicle carries a multispectral camera or high-resolution camera and regularly takes aerial photographs of the field crops to obtain crop canopy images and multispectral data (such as NDVI values, red edge index, etc.) for analyzing crop growth conditions, pest and disease occurrence, and yield prediction. Unmanned aerial vehicle inspection has the characteristics of wide coverage and high collection efficiency, and is suitable for monitoring and authentication of large-area farmland. For livestock and poultry breeding scenes, intelligent wearable devices (such as smart collars, ear tags, etc.) can be used instead of some sensors for data collection. The intelligent wearable device can monitor the body temperature, heart rate, activity level, and other physiological indicators of livestock and poultry in real time, and transmit the data to the cloud server through Bluetooth or wireless communication technology. This method can more accurately obtain real-time data of individual livestock and poultry, providing more detailed basis for the health management and authentication of livestock and poultry.

[0074] The data processing and storage layer technology architecture of smart agriculture is centered around data lifecycle management, and through the integration of multiple dimensions such as standardized processing, intelligent analysis, trusted notarization, and security protection, an efficient, accurate, and secure data value transformation system is constructed. In terms of data standardization and formatting processing, UTC timestamp synchronization with ±1ms level accuracy is achieved through NTP protocol, eliminating time zone differences in different scenarios, and JSON-LD format of W3C standard is used for semantic encapsulation of multi-source data, giving data structured semantic association, such as soil sensor data containing device ID, geographic location, measurement indicators, and timestamp information after encapsulation. Lightweight format conversion middleware pre-resolves raw binary data at the edge node, and the cloud uses Apache NiFi pipeline to complete the final formatting, ensuring data format uniformity. At the same time, a three-level data cleaning process is implemented: the physical layer filters abnormal values in real time based on a rule base, such as identifying market abnormal data with vegetable unit price exceeding 50 yuan / kg; the statistical layer combines the 3σ principle and isolation forest algorithm to remove abnormal values in soil nutrient data that deviate from the mean by 3 times the standard deviation; and the semantic layer maps agricultural economic ontology to quantify plan text into cost coefficients.

[0075] In the field of pest and disease identification, an improved attention mechanism is used to accurately identify small targets such as aphids and gray mold. Once the edge computing identifies the characteristics of the pests, the plant protection unmanned aerial vehicle can plan the operation path within the predetermined time. The crop growth prediction uses a fusion model of Transformer and LSTM, the former encodes the historical sequence of weather, and the latter decodes the growth period state.

[0076] The agricultural digital authentication credible system realizes data storage and value application through a multi-chain collaborative blockchain architecture. The main chain stores land ownership and yield baseline; the logistics sub-chain records temperature and humidity trajectory, and ensures data security through channel isolation; the quality inspection sub-chain verifies pesticide residue compliance using zero-knowledge proof; and the production sub-chain automatically checks the compliance of agricultural operation formula with the help of smart contract. The storage process covers data hashing, joint signature (farmer and supervisor), and cross-chain anchoring (main chain aggregation sub-chain). Data assetization application is significant.

[0077] The present application constructs a digital identity verification system of three-in-one of biometric features, national secret certificates and on-chain addresses; writes continuous variables such as high-temperature cumulative duration into smart contracts to innovate the on-chain insurance actuarial model; and converts farmland income rights into tradable ABS through the environmental stability index. This module not only solves the financial pain points of insufficient agricultural data credibility, but also expands agricultural financing channels through data assetization, promoting the deep integration of agricultural production and financial services.

[0078] The heterogeneous data normalization processing adopts a classification conversion strategy: sensor data is converted into JSON-LD format, such as soil sensor data encapsulated as structured data containing timestamps and geographic locations; 1 frame of key frame is extracted per second from video data and compressed to 1080P, and then stored after timestamping; and text data is mapped to feature vectors through ontology, such as “500 kg / acre of organic fertilizer” mapped to the crop fertilizer requirement node in the knowledge graph.

[0079] The present application realizes millisecond-level penetration query of soil EC value at a specific time by constructing a spatio-temporal data cube, integrating geographic information, time series and blockchain business data; and realizes double value release of data assets, which optimizes the plant protection model through historical pest data on the internal side. Through technology integration, the platform not only solves the trust problem of agricultural product traceability, but also provides support for production optimization and financial services through data mining, promoting the digitalization and value transformation of the agricultural industry chain.

[0080] Embodiment three The present application also proposes an agricultural data authentication device, i.e. a modern agricultural digital authentication large model, such as Figure 7As shown, the agricultural data authentication device 100 comprises: an acquisition module 200, configured to acquire data information of a target farm, the data information comprising crop information, farm information and production information; a first processing module 300, configured to construct a blockchain according to the crop information, the farm information and the production information, the blockchain comprising a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; and a second processing module 400, configured to perform blockchain storage on the blockchain to obtain an authentication certificate of each crop in the target farm, the blockchain storage comprising production data storage, environment data storage, management data storage, processing data storage and tactile data storage, and the crops in the target farm are tracked according to the authentication certificate. The blockchain storage on the blockchain to obtain the authentication certificate of each crop in the target farm comprises: determining a growth state hash value of the crop according to the main chain and the production sub-chain; determining an environment parameter according to the main chain, the logistics sub-chain and the production sub-chain, and generating an early warning event hash value when the environment parameter is abnormal; determining an operation track and an associated hash value of an input product according to the logistics sub-chain, the quality inspection sub-chain and the production sub-chain; determining a process chain hash value of each processing batch according to the logistics sub-chain, the quality inspection sub-chain and the production sub-chain; determining a transportation process hash value according to the logistics sub-chain; and performing on-chain storage on the growth state hash value, the early warning event hash value, the associated hash value, the process chain hash value and the transportation process hash value to obtain the authentication certificate of each crop in the target farm.

[0081] The device further comprises a third processing module configured to generate a biometric hash value according to image information in the data information; construct a triple according to the biometric hash value, a public key of the authentication certificate and a wallet address; and perform on-chain storage of the triple on a blockchain underlying platform to form a digital identity anchor point of the crop.

[0082] The present application forms a closed loop in the aspects of intelligent agricultural production efficiency improvement, blockchain trusted construction, digital authentication security empowerment, financial service innovation support and the like through data penetration and technology fusion, provides a full-chain solution for agricultural modernization transformation, promotes the coordinated development of precision, intelligence, trustworthiness and financialization of agriculture, and helps realize high-quality upgrading and value-added of the agricultural industry.

[0083] The method for agricultural data authentication provided by the application comprises the following steps: obtaining data information of a target farm, wherein the data information comprises crop information, farm information and production information; constructing a block chain according to the crop information, the farm information and the production information, wherein the block chain comprises a main chain, a logistics sub-chain, a quality inspection sub-chain and a production sub-chain; performing block chain storage on the block chain to obtain an authentication certificate of each crop in the target farm, wherein the block chain storage comprises production data storage, environment data storage, management data storage, processing data storage and tactile data storage; and performing data tracking on the crops in the target farm according to the authentication certificate. The method realizes real-time collection of full-cycle agricultural data, dynamic value calculation and risk assessment based on a digital agricultural model to generate an authentication certificate, provides a quantifiable credit basis for a financial institution, converts an agricultural production process into a standard digital asset recognizable by a financial institution, provides a right confirmation basis for financial tools such as credit, futures and trust, and realizes accurate estimation of agricultural yield and value.

[0084] The embodiment of the application provides a storage medium, which stores a program, and the program is executed by a processor to realize the method for agricultural data authentication. The embodiment of the application provides a processor, which is used for running a program, and the program is executed to realize the method for agricultural data authentication. The memory can include a non-permanent memory in a computer readable medium, a random access memory (RAM) and / or a non-volatile memory such as a read-only memory (ROM) or a flash memory (flash RAM). The memory is an example of the computer readable medium.

[0085] The computer readable medium includes permanent and non-permanent, removable and non-removable media, and can be realized by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of the storage medium of the computer include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tape, magnetic tape disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition in this paper, the computer readable medium does not include transitory computer readable medium such as modulated data signals and carriers.

[0086] It should also be noted that the terms "comprising", "comprises" or other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0087] The above embodiments are only used to illustrate the present application, but not to limit it. Instead of the above, various modifications and changes can be made to the application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall fall into the scope of the claims of the application.

Claims

1. A method for agricultural data authentication, characterized in that: The method includes: Acquiring data information of a target farm, wherein the data information includes crop information, farm information, and production information; Constructing a blockchain based on the crop information, farm information, and production information, wherein the blockchain includes a main chain, a logistics sub-chain, a quality inspection sub-chain, and a production sub-chain; Performing blockchain authentication on the blockchain to obtain authentication certificates for each crop in the target farm, wherein the blockchain authentication includes production data authentication, environmental data authentication, management data authentication, processing data authentication, and tactile data authentication; Data tracking of crops in the target farm is performed based on the certification certificate.

2. The method according to claim 1, characterized in that The main chain includes key data, and the key data includes at least one of trigger data of field water holding capacity and feature hash values ​​of pest and disease images; The logistics sub-chain includes transportation tracks and storage environments; The quality inspection sub-chain includes pesticide test results and quality ratings; The production sub-chain includes agricultural operation records and seedling cultivation information.

3. The method according to claim 1, characterized in that The blockchain is stored on the blockchain to obtain the certification certificate of each crop in the target farm, including: Determine the growth status hash value of the crop based on the main chain and the production sub-chain; Determine environmental parameters based on the main chain, logistics sub-chain, and production sub-chain, and generate a warning event hash value when the environmental parameters are abnormal; Determine the operation trajectory and the associated hash value of the input according to the logistics sub-chain, quality inspection sub-chain, and production sub-chain; Determine the process chain hash value of each processing batch based on the logistics sub-chain, quality inspection sub-chain, and production sub-chain; Determine the transport process hash value based on the logistics sub-chain; The growth status hash value, warning event hash value, association hash value, process chain hash value and transportation process hash value are stored on the chain to obtain the certification certificate of each crop in the target farm.

4. The method according to claim 3, characterized in that The certification certificate includes the name of the crop variety, grower information, planting time, growth process data, test results, certification conclusion and risk warning label.

5. The method according to claim 1, wherein The method further includes: Generate a biometric hash value based on the image information in the data information; A triplet is constructed based on the biometric hash value, the public key of the authentication certificate, and the wallet address, and the triplet is stored on the blockchain underlying platform to form a digital identity anchor for the crops.

6. The method according to claim 1, characterized in that The method further includes preprocessing the data information, including: Performing cost rule library screening, edge computing, and standard formatting on the crop information, farm information, and production information to obtain qualified data; The continuous data in the qualified data are cleaned and normalized to obtain preprocessed data.

7. The method according to claim 1, characterized in that The method further includes: Optimizing the crop information, farm information, and production information through a deep learning model; The network neck of the deep learning model is provided with an SE attention module for recalibrating channel weights; The channel shuffling operation of the deep learning model includes group shuffling and point-by-point convolution.

8. The method according to claim 7, characterized in that Setting a crop category vector for the crop image of the target farm; Dynamically adjust the convolution kernel weights in the deep learning model through a multi-head self-attention mechanism; Determining the attention level of crops in the deep learning model based on the crop category vector and the convolution kernel weight for variety identification and / or pest and disease detection; Adopting a path aggregation network to perform adaptive sharpening preprocessing on the crop image to identify disease spot features; Soil and meteorological time series data are determined through deep learning models to predict growth trends and maturity. The meteorological time series data is integrated into the crop growth cycle as characteristic solar term codes.

9. A device for agricultural data authentication, characterized in that: The device includes: An acquisition module is used to acquire data information of a target farm, wherein the data information includes crop information, farm information and production information; A first processing module is configured to construct a blockchain based on the crop information, farm information, and production information, wherein the blockchain includes a main chain, a logistics sub-chain, a quality inspection sub-chain, and a production sub-chain; The second processing module is used to perform blockchain evidence storage on the blockchain to obtain authentication certificates for each crop in the target farm. The blockchain evidence storage includes production data evidence, environmental data evidence, management data evidence, processing data evidence and tactile data evidence, and data tracking of the crops in the target farm is performed based on the authentication certificates.

10. The device according to claim 9, characterized in that The blockchain-based certification process for obtaining certification certificates for each crop in the target farm includes: Determine the growth status hash value of the crop based on the main chain and the production sub-chain; Determine environmental parameters based on the main chain, logistics sub-chain, and production sub-chain, and generate a warning event hash value when the environmental parameters are abnormal; Determine the operation trajectory and the associated hash value of the input according to the logistics sub-chain, quality inspection sub-chain, and production sub-chain; Determine the process chain hash value of each processing batch based on the logistics sub-chain, quality inspection sub-chain, and production sub-chain; Determine the transport process hash value based on the logistics sub-chain; The growth status hash value, warning event hash value, association hash value, process chain hash value and transportation process hash value are stored on the chain to obtain the certification certificate of each crop in the target farm.

11. The device according to claim 9, characterized in that The device also includes: A third processing module, configured to generate a biometric hash value based on the image information in the data information; A triplet is constructed based on the biometric hash value, the public key of the authentication certificate, and the wallet address, and the triplet is stored on the blockchain underlying platform to form a digital identity anchor for the crops.

12. A machine-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the agricultural data authentication method according to any one of claims 1 to 8.

Images