A breeding decision reasoning system for intelligent seed industry
By using a blockchain-edge computing collaborative architecture and a lightweight Transformer model, the contradiction between data sharing and security in breeding decision-making systems has been resolved, enabling real-time breeding decisions in remote areas and improving breeding efficiency and variety quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING MAIMAI QUGENG TECH CO LTD
- Filing Date
- 2026-05-21
- Publication Date
- 2026-06-23
AI Technical Summary
Existing breeding decision-making systems suffer from prominent contradictions between data sharing and security, insufficient real-time decision-making capabilities at the edge, poor adaptability, and poor technical collaboration, resulting in low breeding efficiency and difficulty in meeting the breeding needs of remote areas.
Adopting a blockchain-edge computing collaborative architecture, combining zero-knowledge proof and an improved PBFT consensus mechanism, a lightweight Transformer model is deployed at the edge to achieve secure sharing of breeding data and real-time decision-making, supporting multi-device networking and modular collaboration.
It enables secure sharing and real-time decision-making of breeding data, improves breeding efficiency and variety quality, adapts to the breeding needs of remote areas, and reduces the cost of field trials.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart seed industry technology, specifically to a breeding decision-making reasoning system for smart seed industry, applicable to decision support throughout the entire breeding process of various crops, and particularly suitable for application scenarios in breeding bases in remote areas. Background Technology
[0002] Intelligent seed industry is a core support for agricultural modernization. Breeding decision-making systems, as a core component of intelligent seed industry, undertake crucial functions such as data processing, trait prediction, parent selection, and decision generation, directly impacting breeding efficiency and variety quality. Currently, existing breeding decision-making systems still face numerous technical bottlenecks, severely restricting their industrial application and promotion. First, the conflict between data sharing and security is prominent. Breeding data (especially genomic data and core germplasm resource data) is high-value, privacy-sensitive data. Existing systems either fail to achieve cross-institutional and cross-base data sharing, leading to data silos, or lack effective security protection mechanisms during the sharing process, making it easy for data to be leaked or tampered with, damaging the intellectual property rights of breeding institutions. Among existing technologies, some solutions use blockchain for data traceability, but fail to combine it with effective encryption technology to achieve "usable but not visible" sharing, thus failing to balance the needs of data sharing and privacy protection; some solutions use homomorphic encryption technology for data protection, but suffer from high computational complexity and low sharing efficiency.
[0003] Second, the real-time decision-making capabilities at the edge are insufficient. Existing breeding decision-making systems largely rely on cloud computing, uploading all collected field data to the cloud for analysis and processing. This results in high data transmission latency, failing to meet the needs of real-time field decision-making (such as real-time adjustment of greenhouse conditions and timely intervention in pests and diseases). Simultaneously, breeding bases in remote areas often have poor network conditions, making cloud-dependent systems prone to disconnections, lag, and operational failures. Furthermore, existing edge computing solutions are mostly used only for data collection and simple preprocessing, lacking lightweight inference models and thus unable to achieve independent local decision-making.
[0004] Third, there is poor technological synergy and insufficient adaptability. In existing systems, technologies such as blockchain, edge computing, and artificial intelligence are mostly used independently, without forming an effective collaborative architecture, resulting in overall system inefficiency. For example, edge computing and blockchain lack a collaborative mechanism, making it impossible to securely and efficiently upload and share edge data onto the blockchain; artificial intelligence models are mostly general-purpose and not optimized for edge computing power, failing to adapt to the low computing power and low power consumption constraints of edge nodes. Furthermore, the hardware interfaces of existing systems are inconsistent, making it difficult to adapt to different types of sensing devices and resulting in poor scalability.
[0005] Fourth, the feasibility and stability of the models are insufficient. Existing lightweight models mostly use single pruning or quantization methods, which make it difficult to effectively compress model parameters while ensuring inference accuracy, and cannot adapt to the storage and computing power requirements of edge hardware; blockchain consensus mechanisms mostly use the traditional PBFT algorithm, which has the problems of high bandwidth consumption of master nodes and high consensus latency, and cannot meet the needs of high-frequency uploading and real-time sharing of breeding data.
[0006] In response to the shortcomings of the existing technologies, there is an urgent need for a breeding decision-making reasoning system that can take into account data security and sharing, real-time decision-making, adaptability to remote areas, and technological collaboration. This system would address the pain points of existing technologies, improve the efficiency and security of breeding decisions, and promote the industrialization of smart seed industry. Summary of the Invention
[0007] The purpose of this invention is to overcome the shortcomings of existing breeding decision-making reasoning systems, such as prominent contradictions between data sharing and security, insufficient real-time decision-making capabilities at the edge, poor adaptability, and poor technical synergy. This invention provides a breeding decision-making reasoning system for smart seed industry, which enables secure sharing of breeding data and real-time decision-making, improves the adaptability and stability of the system, reduces breeding costs, and meets the breeding needs of different scenarios.
[0008] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a breeding decision-making reasoning system for smart seed industry, adopting a blockchain-edge computing collaborative architecture, including a perception layer, an edge computing layer, a blockchain layer, a cloud layer, and an application layer. Each layer communicates collaboratively in sequence, with modular collaboration and asynchronous communication, forming a fully closed-loop breeding decision-making system of "data acquisition - edge reasoning - secure storage - cloud iteration - application interaction". The specific technical solution is as follows: The sensing layer of this invention serves as the basic data input unit, used to complete the collection of multi-source heterogeneous data throughout the entire breeding process. The collected data covers four core breeding data categories: genomic data, phenotypic data, environmental data, and management data. Genomic data is acquired by high-throughput sequencing equipment, specifically including SNP data, gene sequence data, and structural variation data. Phenotypic data is collected collaboratively by multispectral imaging equipment, lidar equipment, and a field phenotyping platform, providing key traits such as plant height, canopy structure, yield, and disease resistance. Environmental data is collected by an integrated monitoring station, including temperature, humidity, light, precipitation, soil pH, soil moisture, and GPS location data. Management data is entered through a mobile data collection terminal, covering manual management information such as experimental protocols, hybridization records, and field operations. The sensing layer supports multi-device network operation and synchronous data collection, and also has an offline caching function. The overall data collection error does not exceed 5%, and a daily collection frequency is set for key crop growth stages to ensure the timeliness and accuracy of the data.
[0009] The edge computing layer of this invention is deployed at the breeding base site, serving as the core module for local real-time inference. It can operate independently without cloud networks and specifically includes an edge node controller, a data preprocessing module, and a lightweight inference module. The edge node controller uses the NVIDIA Jetson AGX Orin industrial-grade hardware platform, supporting hot-swappable devices and low-power operation, and can adapt to the complex working environment of high temperature and humidity in the field. The data preprocessing module performs noise filtering, format standardization, and time sequence alignment on the raw collected data, converting unstructured data such as images and sequences into standardized structured data with a data preprocessing latency of no more than 50ms. The lightweight inference module is built on the Transformer architecture and uses a hierarchical threshold pruning algorithm for model optimization. By setting differentiated thresholds for the self-attention module and encoder module to remove redundant tokens, it can output breeding decision results such as parent selection, hybridization prediction, and stress intervention in real time.
[0010] The blockchain layer of this invention serves as the core module for data security and sharing, enabling privacy protection, tamper-proof storage, and full lifecycle traceability of breeding data. Specifically, it includes a consensus node cluster, a data encryption module, and a traceability and evidence storage module. The blockchain layer employs an improved PBFT consensus mechanism based on a beta-distributed three-factor reputation evaluation model. It evaluates node reputation using indicators such as node computing power, data integrity, and interaction frequency, prioritizing high-reputation nodes to participate in the consensus process. This effectively reduces master node bandwidth consumption, with consensus latency not exceeding 500ms. The data encryption module integrates zero-knowledge proofs and NTRU-type homomorphic encryption technology, achieving privacy-preserving sharing of breeding data that is usable but invisible. Data requesters can verify and access data calculation results without obtaining the original data. The traceability and evidence storage module comprehensively records the entire lifecycle of data collection, processing, reasoning, and sharing, ensuring that data is immutable once uploaded to the blockchain.
[0011] The cloud layer of this invention serves as the core module for global iteration and scheduling, specifically including a model training module, a multi-source data management module, and a collaborative scheduling module. The model training module employs the Optuna automated hyperparameter optimization algorithm, continuously iterating and optimizing the lightweight inference model based on structured data and inference logs uploaded from the edge computing layer. It then incrementally distributes updated model parameters to the edge computing layer, ensuring that the model iteration process does not affect the real-time inference operation of the edge computing layer. The multi-source data management module uses a hybrid storage architecture of time-series databases, spatial databases, and relational databases to achieve classified storage, rapid retrieval, and quality verification of multi-source breeding data. The collaborative scheduling module can adaptively allocate computing tasks between the edge and cloud based on the real-time load status and network communication conditions of the edge nodes. When the computing power of the edge nodes is insufficient, complex analysis tasks are scheduled to be processed in the cloud. In the event of a network interruption, it supports the independent operation of the edge computing layer, ensuring the overall operational stability of the system.
[0012] The application layer of this invention serves as the core unit for human-computer interaction, specifically including a mobile APP and a web-based visual decision-making interface. The mobile APP enables offline data entry for breeding, crop phenotypic image acquisition, and automatic generation of experimental reports, meeting the needs of mobile field operations. The web-based visual decision-making interface is built using WEBGL 3D rendering technology, supporting GGE bipolar plots, heatmaps, and 4D spatiotemporal data visualization. Users do not need a professional statistical background to interpret decision results. It also supports users to manually intervene and adjust the breeding decisions output by the system, improving the flexibility and adaptability of decision-making. Beneficial effects
[0013] Compared with the prior art, the present invention has the following advantages: This invention is the first to apply blockchain and edge computing in deep collaboration to a breeding decision-making reasoning system. By combining zero-knowledge proofs and an improved PBFT consensus mechanism, it solves the pain point of prominent data sharing and security contradictions in existing systems, and realizes the "usable but invisible" sharing and tamper-proof traceability of breeding data. At the same time, it deploys a lightweight Transformer model based on hierarchical threshold pruning at the edge, realizing local real-time decision-making, breaking through the dependence of existing systems on cloud networks, adapting to breeding scenarios in remote areas, and distinguishing itself from the shortcomings of existing technologies where blockchain and edge computing are applied independently and the edge end lacks independent reasoning capabilities. The NVIDIA Jetson AGX Orin edge controller, hierarchical threshold pruning algorithm, and improved PBFT consensus mechanism used in this invention are all existing mature technologies, eliminating the need to develop entirely new technologies and reducing R&D costs and implementation difficulty. The standardized interfaces of each module layer of the system can flexibly adapt to different types of sensing devices and breeding scenarios, support multi-base collaborative breeding, and can be operated by grassroots breeders without professional computer knowledge, making it easy to promote and apply. At the same time, the hardware configuration and software modules of the system can be expanded and optimized according to actual needs to adapt to breeding institutions of different sizes. This invention enables secure management and real-time decision-making for the entire breeding data process, allowing for rapid response to field breeding needs and timely generation of decision-making suggestions such as parent selection, hybrid combination prediction, and environmental stress intervention. During data sharing, zero-knowledge proofs and homomorphic encryption technologies protect data privacy, while blockchain enables traceability, effectively preventing data leakage and tampering and protecting the intellectual property rights of breeding institutions. At the same time, the system can improve the efficiency of breeding data processing and decision-making, reduce field trial costs, and significantly improve breeding efficiency and variety quality. This invention supports various scenarios such as field breeding, greenhouse breeding, and facility breeding, and is suitable for the breeding needs of various crops such as corn, wheat, soybeans, and cotton. The edge node controller can flexibly connect to different types of sensing devices and support multi-device network operation. The system can operate independently without cloud network, which is suitable for scenarios with poor network conditions in remote areas. At the same time, it supports cloud-edge collaborative work to meet the needs of large-scale, multi-base collaborative breeding. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram of the module structure of the edge computing layer of the present invention; Figure 3 This is a flowchart of the system of the present invention; Detailed Implementation
[0015] The present invention will be further described in detail below with reference to specific embodiments, so that those skilled in the art can understand it.
[0016] Example 1
[0017] This embodiment provides a breeding decision-making reasoning system for smart seed industry, applied to soybean breeding bases. For example... Figure 1 As shown, the system adopts a blockchain-edge computing collaborative architecture, which includes a perception layer, an edge computing layer, a blockchain layer, a cloud layer, and an application layer from bottom to top. Each layer communicates with each other in turn to form a closed-loop breeding decision-making system of "collection-processing-sharing-decision-iteration".
[0018] The perception layer serves as the core of data acquisition, used to collect multi-source heterogeneous data throughout the entire soybean breeding process. The genome sequencing equipment uses an Illumina NovaSeq 6000 sequencer, capable of generating 10TB of SNP data daily, enabling high-throughput acquisition of soybean genome data. The phenotypic acquisition equipment utilizes the Top Cloud Agriculture unmanned vehicle-based high-throughput phenotypic acquisition and analysis platform, equipped with multi-dimensional imaging units including visible light, depth imaging, hyperspectral imaging, thermal infrared, and lidar, to achieve three-dimensional acquisition of phenotypic data such as soybean plant height, canopy structure, and yield. The environmental monitoring equipment deploys 10 integrated environmental monitoring stations, spaced 50 meters apart, collecting parameters such as air temperature and humidity, photosynthetically active radiation, precipitation, soil temperature, humidity, salinity, and pH. The monitoring stations are equipped with GPS positioning, support network operation, and the phenotypic data acquisition error is controlled within 4.5%. Data acquisition during key growth periods (flowering and pod-setting stages) is performed once daily. The mobile acquisition terminal uses a smartphone equipped with the BioE-Collection APP, supporting functions such as single-plant trait entry, image acquisition, and voice recording, adapting to refined breeding scenarios such as soybean backcrossing and conversion.
[0019] like Figure 2 As shown, the edge computing layer is deployed at the soybean breeding base and includes a data preprocessing module and a lightweight inference module. The edge node controller uses the NVIDIA Jetson AGX Orin industrial version, configured with 64GB of storage, and its power can be adjusted between 15W and 60W. It has power management and heat dissipation optimization functions, and is suitable for the high temperature and high humidity environment in the field. The data preprocessing module is developed using Python and integrates noise filtering (based on Gaussian filtering algorithm), format conversion (converting image data to JSON format), and time series alignment (based on dynamic time warping algorithm). The preprocessing latency is controlled within 40ms. The lightweight inference module is developed based on the Transformer model. The initial model parameter scale is 1 billion. Using a hierarchical threshold pruning algorithm, the self-attention submodule and encoder submodule of the model are set with thresholds of 0.3 and 0.5, respectively. After removing redundant tokens, the model parameters are compressed to 280 million (compressed to 28% of the original scale). The inference accuracy loss is controlled within 4%, and the inference latency is ≤80ms. It can output suggestions for soybean parent selection, hybrid combination prediction, and pest and disease intervention suggestions.
[0020] The blockchain layer is used to achieve secure sharing, tamper-proof traceability, and intellectual property protection of soybean breeding data. This embodiment deploys 8 consensus nodes, including 3 master nodes and 5 backup nodes. An improved PBFT consensus mechanism is used, and a three-factor (node computing power, data integrity, and interaction frequency) reputation evaluation method based on beta distribution is used to calculate node reputation values, with consensus latency controlled within 450ms. The data encryption module uses zero-knowledge proofs (zk-SNARKs algorithm) and NTRU-type homomorphic encryption technology to encrypt soybean genome data and core phenotypic data, achieving "usable but invisible" sharing of breeding data. Data transmission uses the AES-256 encryption algorithm to prevent data leakage and tampering. The traceability and evidence storage module uses the Hyperledger Fabric blockchain framework to record information throughout the entire lifecycle of data collection, preprocessing, inference, and sharing, forming an immutable traceability chain with 100% accuracy.
[0021] The cloud layer is deployed on Alibaba Cloud servers, serving as the system's model iteration and global collaborative scheduling center. The model training module employs the Optuna automated hyperparameter optimization algorithm, combined with 1 million soybean breeding data points, to iteratively optimize the lightweight inference model. A model update is completed every 7 days, and the optimized model parameters are distributed to the edge computing layer via an encrypted channel, enabling dynamic model updates without affecting the edge computing layer's real-time decision-making capabilities. The data management module uses a hybrid storage architecture combining InfluxDB time-series database (for storing environmental time-series data), PostgreSQL relational database (for storing management data), and MongoDB database (for storing unstructured data), achieving categorized storage and rapid querying of multi-source data. The collaborative scheduling module uses Kubernetes container orchestration technology, adaptively allocating data analysis tasks based on the CPU, memory utilization, and network bandwidth of each edge node. When the edge node's CPU utilization exceeds 80%, some complex analysis tasks are scheduled to the cloud layer for processing. When the network is interrupted, the edge layer runs independently, ensuring system stability.
[0022] The application layer presents decision-making results to breeders in an intuitive and user-friendly manner. The mobile app, developed for Android and iOS systems, supports eight field positioning methods and features single-plant trait entry, image capture, voice recording, and visualization report generation (GGE bipolar plots, heatmaps), adapting to soybean backcrossing and breeding scenarios. The visual decision-making interface utilizes an HTML5 WebGL 3D display module, combined with historical time dimensions, to achieve four-dimensional visualization of soybean breeding data. It supports data query, decision intervention, and report export functions. Grassroots breeders can access the app directly through a browser, interpreting decision-making results without requiring a statistical background.
[0023] like Figure 3 As shown, the system workflow of this embodiment includes the following steps: S1. Data Acquisition: The perception layer collects multi-source data on soybean breeding through the above-mentioned devices, including genome sequencing data, phenotypic data, environmental data and management data, and transmits them to the edge computing layer via 5G network or WiFi.
[0024] S2. Edge Preprocessing and Lightweight Inference: The data preprocessing module of the edge computing layer performs noise filtering, format conversion, and time-series alignment on the collected raw data to remove invalid data and noise, and convert unstructured data into structured data. The lightweight inference module performs real-time analysis on the preprocessed structured data and outputs suggestions for soybean parent selection, hybrid combination prediction, and pest and disease intervention. At the same time, the structured data and inference logs are uploaded to the blockchain layer and cloud layer through an encrypted channel.
[0025] S3. Data Encryption and Traceability: The consensus nodes of the blockchain layer verify the uploaded data, the data encryption module encrypts and stores the data, and the traceability and evidence storage module records the data flow information, generating an immutable traceability chain that supports secure data sharing across institutions.
[0026] S4. Cloud-based model iteration and collaborative scheduling: The cloud-based model training module performs in-depth analysis of the uploaded data, optimizes the hyperparameters of the lightweight inference model, and sends the optimized model parameters to the edge computing layer to achieve dynamic model iteration; the collaborative scheduling module adaptively allocates tasks according to the load status of edge nodes.
[0027] S5. Decision Presentation and User Intervention: The mobile APP and visual decision interface of the application layer receive and display the decision suggestions output by the edge computing layer. Users can view the decision results, enter field management records, and manually intervene in the decision suggestions, forming field operation data feedback to the perception layer, completing the closed loop of the entire soybean breeding decision-making process.
[0028] After being applied in a soybean breeding base, the system in this embodiment achieved the following implementation effects: In terms of data sharing and security, through zero-knowledge proof and blockchain technology, secure sharing of soybean genome data and core phenotypic data is achieved, with a data leakage rate of 0% and a traceability accuracy of 100%, effectively protecting breeding intellectual property rights; In terms of real-time decision-making capabilities, edge inference latency is ≤80ms and data preprocessing latency is ≤40ms, enabling real-time response to field breeding needs and timely generation of pest and disease intervention suggestions to reduce crop losses; In terms of adaptability to remote areas, it can operate independently without cloud networks, adapting to scenarios with poor network conditions in remote areas, with system stability exceeding 99.5%; In terms of improved breeding efficiency, soybean breeding data processing and decision-making efficiency is improved by 55%, hybrid combination screening efficiency is improved by 60%, field trial costs are reduced by 45%, and the breeding cycle is shortened by 30%.
[0029] Example 2
[0030] This embodiment provides an application of a breeding decision-making reasoning system for smart seed industry in maize breeding. For example... Figure 1 As shown, the system adopts the same blockchain-edge computing collaborative architecture as in Example 1.
[0031] In terms of the sensory layer configuration: the genome sequencing equipment uses the Illumina NovaSeq 6000 sequencer, which can generate 10TB of SNP data per day; the phenotypic acquisition equipment uses the Top Cloud field fixed plant phenotypic monitoring system, which integrates IoT, multispectral and infrared thermal imaging, artificial intelligence and other technologies, and can automatically collect maize growth images and multi-dimensional phenotypic data around the clock; the environmental monitoring equipment deploys 20 integrated environmental monitoring stations, each 100 meters apart, to collect parameters such as air temperature and humidity, photosynthetically active radiation, precipitation, soil temperature, humidity and salinity, and pH value; the mobile acquisition terminal uses a smartphone equipped with the BioE-Collection APP, which supports maize field trait input, image acquisition and other functions.
[0032] like Figure 2 As shown, in the edge computing layer, the edge node controller uses the NVIDIA Jetson AGX Orin industrial version with 128GB of storage; the data preprocessing module is developed in C++ and integrates noise filtering, format conversion, and time-series alignment functions, with preprocessing latency controlled within 30ms; the lightweight inference module is developed based on the Transformer model, with an initial model parameter scale of 1.5 billion. Using a hierarchical threshold pruning algorithm, the model parameters are compressed to 450 million (compressed to 30% of the original scale), the inference accuracy loss is controlled within 3%, and the inference latency is ≤60ms. It can output decision suggestions such as maize parent selection recommendations, hybrid combination prediction, and environmental stress intervention.
[0033] The blockchain layer deploys 10 consensus nodes, including 4 master nodes and 6 backup nodes. It adopts an improved PBFT consensus mechanism and calculates node reputation values based on a three-factor reputation evaluation method with beta distribution. The consensus latency is controlled within 400ms. The data encryption module uses zero-knowledge proof and NTRU-type homomorphic encryption technology to encrypt maize genome data and core phenotypic data. The traceability and evidence storage module adopts the Hyperledger Fabric blockchain framework to record the entire lifecycle flow information of data.
[0034] The cloud layer is deployed on Tencent Cloud servers and uses the Optuna automated hyperparameter optimization algorithm. Combined with 500,000 pieces of big data on maize breeding, iteratively optimizes the lightweight inference model and completes a model update every 5 days. The data management module adopts a hybrid architecture of time-series database, spatial database and relational database. The collaborative scheduling module adaptively allocates data analysis tasks based on the real-time load status of each edge node.
[0035] like Figure 3 As shown, the system workflow of this embodiment includes five steps from S1 to S5: S1, the perception layer collects multi-source data on maize breeding and transmits it to the edge computing layer via a 5G network; S2, the edge computing layer preprocesses and performs lightweight inference on the data, outputs suggestions for maize parent selection, hybrid combination prediction, and pest and disease intervention, and uploads the data to the blockchain layer and the cloud layer; S3, the blockchain layer encrypts and stores the data, verifies consensus, and uploads it to the chain, generating an immutable traceability record; S4, the cloud layer performs in-depth analysis on the data, optimizes the lightweight inference model, and sends the optimized model parameters to the edge computing layer; S5, the application layer receives decision suggestions, supports user interaction and decision intervention, and completes the entire maize breeding decision-making process.
[0036] After the system of this embodiment was applied in the maize breeding base, the following implementation effects were achieved: edge inference latency ≤60ms, data preprocessing latency ≤30ms, and real-time response to field breeding needs; through zero-knowledge proof and blockchain technology, secure sharing of maize genome data was achieved, with a data leakage rate of 0 and a traceability accuracy of 100%; maize breeding data processing and decision-making efficiency was improved by 60%, hybrid combination screening efficiency was improved by 65%, field trial costs were reduced by 50%, and the breeding cycle was shortened by 35%.
[0037] Example 3
[0038] This embodiment provides an application of a breeding decision-making reasoning system for smart seed industry in rice breeding. For example... Figure 1 As shown, the system architecture is consistent with the aforementioned embodiments.
[0039] In terms of the sensory layer configuration: the genome sequencing equipment uses the Illumina NovaSeq 6000 sequencer; the phenotypic acquisition equipment uses multispectral imaging equipment, lidar, and field phenotypic monitoring system to collect core phenotypic characteristics such as plant height, canopy structure, and yield; the environmental monitoring equipment deploys 15 integrated environmental monitoring stations to collect parameters such as air temperature and humidity, photosynthetically active radiation, precipitation, soil temperature, humidity, salinity, and pH value; the mobile acquisition terminal uses a smartphone equipped with the BioE-Collection APP, which supports functions such as rice field trait entry and image acquisition.
[0040] like Figure 2 As shown, in the edge computing layer, the edge node controller uses the NVIDIA Jetson AGX Orin industrial version with 64GB of storage; the data preprocessing module is developed using Python, with preprocessing latency controlled within 45ms; the lightweight inference module is developed based on the Transformer model, with model parameters compressed to 25% of the original size, inference accuracy loss controlled within 5%, and inference latency ≤90ms.
[0041] The blockchain layer deploys 6 consensus nodes, including 2 master nodes and 4 backup nodes, and adopts an improved PBFT consensus mechanism, with consensus latency controlled within 500ms; the data encryption module adopts zero-knowledge proof and NTRU-type homomorphic encryption technology; the traceability and evidence storage module records the entire lifecycle flow information of the data.
[0042] The cloud layer is deployed on Huawei Cloud servers and uses the Optuna automated hyperparameter optimization algorithm. Combined with 800,000 rice breeding big data points, the lightweight inference model is iteratively optimized, and the model is updated every 6 days.
[0043] like Figure 3 As shown, the system workflow is similar to that of Example 1, including five steps: S1 data acquisition, S2 edge preprocessing and lightweight inference, S3 data encryption and on-chain traceability, S4 cloud model iteration and collaborative scheduling, and S5 decision presentation and user intervention.
[0044] After the system of this embodiment was applied in the rice breeding base, the following implementation effects were achieved: edge inference latency ≤90ms, data preprocessing latency ≤45ms; rice breeding data processing and decision-making efficiency improved by 50%, hybrid combination screening efficiency improved by 55%, field trial cost reduced by 40%, and breeding cycle shortened by 25%.
[0045] Example 4
[0046] This embodiment provides an application of a breeding decision-making reasoning system for smart seed industry in a breeding base in a remote area. For example... Figure 1As shown, the system architecture is the same as the previous embodiment, but the equipment selection and configuration have been adapted and optimized for the special conditions of remote areas.
[0047] In terms of the sensor layer configuration: the genome sequencing equipment uses a portable gene sequencer, which is convenient for field operation; the phenotypic collection equipment uses a drone multispectral imaging system, equipped with a 5-band (RGB + red edge + near infrared) imaging device with a resolution of 0.5cm / pixel; the environmental monitoring equipment deploys 5 integrated environmental monitoring stations, which support solar power supply and are suitable for scenarios with insufficient power supply in remote areas; the mobile collection terminal uses a smartphone, which supports offline data entry and automatically synchronizes after the network is restored.
[0048] like Figure 2 As shown, in the edge computing layer, the edge node controller uses the NVIDIA Jetson AGX Orin industrial version, configured with 32GB of storage, and supports solar power; the data preprocessing module adopts an optimized algorithm to reduce computational complexity and adapt to the resource-constrained conditions at the edge; the lightweight inference module uses a simplified Transformer model, with model parameters compressed to 20% of the original size, inference accuracy loss controlled within 5%, and inference latency ≤100ms.
[0049] The blockchain layer deploys four consensus nodes and adopts an improved PBFT consensus mechanism, keeping the consensus latency within 500ms. The data encryption module uses a lightweight encryption algorithm to reduce computational complexity and adapt to edge computing power conditions.
[0050] The cloud layer is deployed on a local server and uses the Optuna automated hyperparameter optimization algorithm, combined with local breeding big data, to iteratively optimize the lightweight inference model.
[0051] like Figure 3 As shown, due to poor network conditions in remote areas, the system workflow of this embodiment has the following characteristics: S1, the perception layer collects multi-source breeding data and transmits it to the edge computing layer via short-range communication methods such as WiFi or Bluetooth; S2, the edge computing layer operates independently, preprocesses the data and performs lightweight inference, and outputs breeding decision suggestions without relying on the cloud network; when network conditions permit, steps S3 and S4 are executed to upload the data to the blockchain layer and the cloud layer for encrypted storage and model iteration; S5, the application layer receives decision suggestions from the edge and supports offline viewing and manual intervention.
[0052] After the system in this embodiment was applied in a breeding base in a remote area, it achieved independent operation without cloud network conditions, effectively solving the technical problems of poor network conditions and high cloud latency in remote areas, and ensuring the real-time nature of breeding decisions and the continuity of system operation.
[0053] It should be noted that the above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any obvious modifications, substitutions, or alterations made without departing from the technical concept of the present invention should fall within the protection scope of the present invention.
Claims
1. A breeding decision-making reasoning system for intelligent seed industry, characterized in that, include: The layers are: perception layer, edge computing layer, blockchain layer, cloud layer, and application layer. The perception layer, edge computing layer, blockchain layer, and cloud layer are sequentially connected in communication, and the application layer is connected in communication with the edge computing layer and the cloud layer respectively. The sensing layer is used to collect multi-source heterogeneous data throughout the entire breeding process; The edge computing layer is deployed at the breeding site to perform local preprocessing and lightweight inference on the multi-source heterogeneous data, output real-time breeding decisions, and can run independently without cloud networks. The blockchain layer is used for encrypted storage, consensus verification, tamper-proof traceability, and privacy-secure sharing of preprocessed structured data; The cloud layer is used to complete the iterative optimization of the inference model based on data uploaded from the edge, and to send the optimized model parameters to the edge computing layer. The application layer is used to display decision results and provide user interaction and decision intervention interfaces.
2. The system according to claim 1, characterized in that, The sensing layer includes a genome sequencing device, a multispectral imaging device, an integrated environmental monitoring station, and a mobile data acquisition terminal. The multi-source heterogeneous data includes genomic data, phenotypic data, environmental data, and breeding management data; The environmental data includes air temperature and humidity, soil physicochemical parameters, photosynthetically active radiation, and GPS positioning information, and supports synchronous collection by multiple monitoring devices in a network.
3. The system according to claim 1, characterized in that, The edge computing layer includes an edge node controller, a data preprocessing module, and a lightweight inference module; The edge node controller uses the NVIDIA Jetson AGX Orin hardware platform; The data preprocessing module is used to perform noise filtering, format standardization, and time-series alignment on the raw data, converting unstructured data into structured data. The lightweight inference module is a Transformer model optimized by hierarchical threshold pruning. Under the condition that the inference accuracy loss is ≤5%, the model parameters are compressed to less than 30% of the original size.
4. The system according to claim 3, characterized in that, The optimization method for the lightweight inference module is as follows: the initial Transformer model is trained based on cloud-based breeding big data, and differentiated pruning thresholds are set for the model's self-attention submodule and encoder submodule respectively to remove redundant tokens below the threshold and adapt to the computing power and storage constraints of edge nodes.
5. The system according to claim 1, characterized in that, The blockchain layer includes a consensus node cluster, a data encryption module, and a traceability and evidence storage module. The consensus node cluster adopts an improved PBFT consensus mechanism, which selects high-reputation nodes to participate in consensus based on a beta-distributed three-factor reputation evaluation model, thereby reducing the bandwidth consumption of the master node and achieving a consensus latency of ≤500ms. The data encryption module integrates zero-knowledge proof and NTRU-type homomorphic encryption to achieve privacy-preserving sharing of breeding data that is available but not visible. The traceability and evidence storage module is used to record the entire lifecycle information of data, forming an unalterable traceability chain.
6. The system according to claim 5, characterized in that, The execution process of the zero-knowledge proof is as follows: the data provider encrypts the breeding data, generates a verification proof, and uploads it to the blockchain. The data requester does not need to obtain the original data, but can obtain the data calculation results only through the verification proof, thus achieving a balance between privacy protection and data reuse.
7. The system according to claim 1, characterized in that, The cloud layer includes a model training module, a multi-source data management module, and a collaborative scheduling module; The model training module uses the Optuna hyperparameter optimization algorithm to iteratively update the lightweight inference model; The multi-source data management module adopts a hybrid storage architecture of time-series database, spatial database, and relational database; The collaborative scheduling module is used to adaptively allocate edge / cloud computing tasks based on the edge node load and network status.
8. The system according to claim 1, characterized in that, The application layer includes a mobile data acquisition app and a web-based visual decision-making interface; The mobile app supports offline entry of field traits, image acquisition, and synchronization of experimental records. The visualization decision interface supports the visualization of GGE bipolar maps, heat maps, and four-dimensional spatiotemporal data, and allows users to manually intervene in the decision results.
9. The system according to any one of claims 1-8, characterized in that, The system's workflow includes: S1: The perception layer collects multi-source breeding data and transmits it to the edge computing layer; S2: After the edge computing layer preprocesses the data, it completes local real-time inference through the lightweight inference module, outputs breeding decisions, and simultaneously encrypts and uploads the structured data and inference logs to the blockchain layer and the cloud layer. S3: The blockchain layer performs consensus verification, encrypted storage, and on-chain notarization of data to achieve secure data sharing and traceability; S4: The cloud layer optimizes the inference model based on edge data and sends the updated model parameters to the edge computing layer; S5: The application layer receives and displays decision results, supporting user interaction and intervention.