A smart agriculture monitoring method based on brain-like hyperdimensional computing

By adopting brain-like ultra-dimensional computing technology in agricultural intelligent monitoring systems and using super-dimensional vectors to regulate plant growth environment, the problems of high hardware performance and high cost in traditional systems are solved, and efficient monitoring and prediction on microcontrollers are achieved, reducing the number of high-quality monitoring and prediction on microcontrollers. agricultural production costs.

CN116047960BActive Publication Date: 2024-11-15SHANGHAI INST OF MICROSYSTEM & INFORMATION TECH CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211703996.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-28
Publication Date
2024-11-15
Estimated Expiration
2042-12-28

AI Technical Summary

Technical Problem

Traditional agricultural intelligent monitoring systems are based on deep neural networks, with high hardware performance requirements, long training time, and large storage space, so they cannot conduct local learning and prediction on microcontrollers, making it difficult to promote in cost-sensitive agricultural fields.

Method used

Using a smart agricultural monitoring method based on brain-like ultra-dimensional computing, the deployment subsystems on the sensor side, edge side and controller side are used to learn online real-time plant feature data through brain-like intelligent analysis trainers and actuators, and encoded as super Dimension vectors, generate growth environment regulation models, optimize temperature and humidity and other environments.

Benefits of technology

It reduces hardware performance requirements, realizes local learning and prediction on microcontrollers, and is suitable for cost-constrained agricultural IoT intelligent applications, reduces agricultural production costs and increases crop yields.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116047960B_ABST
    Figure CN116047960B_ABST
Patent Text Reader

Abstract

The present invention provides a smart agricultural monitoring method based on brain-like hyperdimensional computing, including: using the device application layer on the sensor side to collect feature data; transmitting data from the device application layer on the sensor side to the device application layer on the edge side; using the device application layer on the edge side to detect whether it is in an analysis training or control execution state, and correspondingly sending data to a brain-like intelligent analysis trainer or a brain-like intelligent executor of the brain-like intelligent layer; the edge side deployment subsystem executes the analysis training process based on the hyperdimensional vector calculation method; the edge side deployment subsystem obtains the device control instruction based on the hyperdimensional vector calculation method and sends it to the device application layer of the control side deployment subsystem, and the device application layer of the control side deployment subsystem executes the device control instruction. The method of the present invention reduces the requirements for hardware performance and reduces costs while monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things intelligent monitoring, and specifically to a smart agriculture monitoring method based on brain-like hyper-dimensional computing. Background Art

[0002] In traditional agriculture, watering, fertilizing and spraying of crops all rely on the farmers' personal experience and feelings, and the amount cannot be accurately controlled. Farmers cannot accurately grasp when fruits and vegetables need watering, fertilizing, spraying and adding nutrition, how to maintain precise temperature, humidity, light, and carbon dioxide concentration in agricultural greenhouses, and how to implement supply on demand. This leads to excessive or insufficient watering, fertilizing and spraying, which ultimately affects the yield and quality of crops.

[0003] In order to reduce the difficulty of growing greenhouse crops such as cucumbers, it is necessary to use an agricultural intelligent monitoring system to monitor the growth status of crops such as cucumbers and automatically adjust the internal environment of the greenhouse. However, the existing various types of agricultural intelligent monitoring systems are generally based on machine learning algorithms of deep neural networks. Due to the large number of parameters, complex floating-point iterative operations and large-scale model structures, they have high requirements for hardware performance, time-consuming and complex training, and the trained algorithm model requires a large storage space. It can only be trained on large cloud servers and inference on local large servers. It is impossible to perform local learning and prediction on single-chip microcomputers. Because the large-scale end-edge-cloud system is highly complex and has high operation and maintenance costs, it is difficult to promote it on a large scale in the cost-sensitive agricultural industry.

[0004] Hyperdimensional computing (HDC) is a computing method inspired by the large number of synapses and neurons in the brain. It is inspired by the abstract representation of the human brain's neuronal circuits and includes three stages: encoding, training, and comparison. Traditional computing uses bits as the smallest unit of computing, while the basic data type of hyperdimensional computing is a supervector (a vector with a dimension of thousands). Hyperdimensional computing uses the high-dimensional distributed representation of data as the identification object, and converts different types of information into hyperdimensional vectors through encoding, that is, vectors with a dimension of thousands or more. Information is randomly and integrally distributed on each element of the hyperdimensional vector and is uniquely referred to and represented by the hyperdimensional vector. Commonly used operations in hyperdimensional computing are bit-by-bit addition, bit-by-bit multiplication, and shift. Among them, the addition operation retains the information of each element, and the result of the operation remains similar to each element, so it is often used to represent a set with multiple elements. In high-dimensional space, a variety of computing behaviors with special mathematical properties can be realized by operating on hyperdimensional vectors, thereby generating learning and memory functions. These operations and operation combinations have good hardware friendliness, extremely low hardware performance requirements, simple calculations, convenient deployment, and fast calculations. At the same time, the hyperdimensional computing model can still maintain stable working performance in special cases such as a small number of available samples and large noise interference. When new training samples are available, the existing model can be directly and incrementally updated through simple multiplication and addition operations, making the model more accurate without the need to retrain the entire model.

[0005] The application field of agricultural intelligent monitoring mainly considers low cost and low power consumption. Traditional machine learning algorithms based on deep neural networks consume a lot of computing power and storage space, and are complex to deploy, which inevitably leads to high power consumption and high cost. It is difficult to popularize on a large scale in the cost-sensitive agricultural field. Brain-like hyperdimensional computing is an intelligent computing method based on spiking neural networks (SNNs). It eliminates the costly multiplication operation and becomes a pure addition operation. It is also driven by pulse events, that is, if no pulse signal is received, no calculation will be performed. Therefore, the power consumption is extremely low, only one thousandth of traditional AI. Because hyperdimensional computing can perform complex calculations with less computing power, it is particularly suitable for IoT edge intelligent application fields with limited computing power, such as smart agricultural edge intelligent monitoring.

[0006] However, brain-like hyperdimensional computing is a brand-new technological field. The brain-like chip hardware that carries this technology is still in the laboratory research stage and has not yet been commercialized on a large scale. The application in various industries is also still in the exploratory stage. Therefore, brain-like hyperdimensional computing has not yet been used in the field of agricultural intelligent monitoring applications. Summary of the invention

[0007] The purpose of the present invention is to provide a smart agricultural monitoring method based on brain-like hyperdimensional computing to reduce the requirements on hardware performance and reduce costs.

[0008] In order to achieve the above object, the present invention provides a smart agriculture monitoring method based on brain-like hyperdimensional computing, which is based on a smart agriculture monitoring system based on brain-like hyperdimensional computing with a sensor side deployment subsystem, an edge side deployment subsystem and a controller side deployment subsystem, and includes:

[0009] S1: Execute the uplink monitoring process, including:

[0010] S11: Use the device application layer of the sensor side deployment subsystem to drive various sensors to collect plant feature data and environmental feature data;

[0011] S12: Transmitting plant feature data and environmental feature data from the device application layer of the sensor-side deployment subsystem to the device application layer of the edge-side deployment subsystem

[0012] S13: using the device application layer of the edge-side deployment subsystem to detect whether its brain-like intelligence layer is in an analysis and training state or a control and execution state; if it is detected that the brain-like intelligence layer is in an analysis and training state, the plant feature data and the environmental feature data are sent to the brain-like intelligence analysis and training device of the brain-like intelligence layer; if it is detected that the brain-like intelligence layer is in a control and execution state, the plant feature data and the environmental feature data are sent to the brain-like intelligence executor of the brain-like intelligence layer;

[0013] S2: The edge-side deployment subsystem uses its brain-like intelligent analysis trainer and the brain-like memory storage at the physical layer to perform the analysis and training process based on the hyperdimensional vector computing method;

[0014] S3: After the analysis and training process is completed, the edge-side deployment subsystem uses its brain-like intelligent actuator and the brain-like memory storage of the physical layer to obtain the device control instructions based on the hyperdimensional vector calculation method and sends them to the device application layer of the control-side deployment subsystem. The device application layer of the control-side deployment subsystem executes the device control instructions.

[0015] The step S12 also includes: using the device application layer of the edge side deployment subsystem to perform data preprocessing.

[0016] In step S12, the distribution layer, the base layer and the physical layer of the sensor side deployment subsystem are used to transmit the plant feature data and the environmental feature data from the device application layer of the sensor side deployment subsystem to the device application layer of the edge side deployment subsystem.

[0017] The step S2 specifically includes:

[0018] S21: The brain-like intelligent analysis trainer of the brain-like intelligent layer of the edge-side deployment subsystem encodes the plant feature data and environmental feature data collected in batches into hyperdimensional vectors, and then performs intelligent training on the brain-like memory storage of the physical layer of the edge-side deployment subsystem to obtain training results;

[0019] S22: Storing the training results in the form of hyperdimensional vectors in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, thereby updating the training results stored in the brain-like memory storage.

[0020] The step S21 specifically includes:

[0021] S211: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect multiple batches of plant feature data and environmental feature data as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set;

[0022] S212: Determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S213. Otherwise, directly execute step S213.

[0023] S213: performing feature clustering on the training sample data collected in multiple batches, completing one-dimensionalization and feature binarization of image features;

[0024] S214: feature classification of all vector elements according to the germination stage, seedling stage, initial flowering stage and fruiting stage of the plant and vector assembly respectively to obtain assembled first super-dimensional vectors HV1' to assembled fourth super-dimensional vectors HV4';

[0025] S215: Add the assembled first to fourth super-dimensional vectors HV1'-HV4' to the first to fourth super-dimensional vectors HV1-HV4 stored in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, and obtain the training results as new first to fourth super-dimensional vectors HV1"-HV4";

[0026] In step S22, the stored training results include the stored first to fourth superdimensional vectors HV1~HV4, wherein the first superdimensional vector corresponds to the germination stage of the plant, the second superdimensional vector corresponds to the seedling stage of the plant, the third superdimensional vector corresponds to the early flowering stage of the plant, and the fourth superdimensional vector corresponds to the fruiting stage of the plant.

[0027] The step S3 specifically includes:

[0028] S31: The brain-like intelligent actuator of the brain-like intelligent layer of the edge-side deployment subsystem encodes the currently collected plant feature data and environmental feature data into a current hyperdimensional vector, performs similarity matching between the current hyperdimensional vector and the brain-like model corresponding to the training result stored in the brain-like memory storage device, and generates a corresponding growth environment regulation model according to the similarity result;

[0029] S32: using the brain-like intelligent actuator of the brain-like intelligent layer of the edge-side deployment subsystem, convert the growth environment regulation model into a device control instruction according to the hyperdimensional vector label, and send it to the device application layer of the edge-side deployment subsystem;

[0030] S33: using the device application layer of the edge side deployment subsystem to convert the device control instruction, and then sending it to the device application layer of the control side deployment subsystem;

[0031] S34: Utilize the device application layer of the control-side deployment subsystem to drive the corresponding device according to the device control instruction to adjust the plant growth environment.

[0032] The step S31 specifically includes:

[0033] S311: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect the plant feature data and environmental feature data collected during the current period as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set;

[0034] S312: Determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S313. Otherwise, directly execute step S313.

[0035] S313: performing feature clustering, one-dimensionalization of image features, feature binarization, and vector assembly on the training sample data collected in the current period in sequence to obtain a current hyperdimensional vector HV;

[0036] S314: judging the plant growth stage according to the similarity between the current hyperdimensional vector HV and the training results stored in the brain-like memory storage;

[0037] S315: Generate a growth environment control model according to the judgment result of the plant growth stage and the similarity result and hyperdimensional vector label between the current hyperdimensional vector and the corresponding hyperdimensional vector in the stored training result.

[0038] The stored training results include the first to fourth super-dimensional vectors HV1-HV4, wherein the first super-dimensional vector corresponds to the germination stage of the plant, the second super-dimensional vector corresponds to the seedling stage of the plant, the third super-dimensional vector corresponds to the initial flowering stage of the plant, and the fourth super-dimensional vector corresponds to the fruiting stage of the plant;

[0039] In step S314, the determination result of the plant growth stage includes: determining that the plant is in the germination stage, determining that the plant is in the seedling stage, determining that the plant is in the initial flowering stage, and determining that the plant is in the fruiting stage;

[0040] The step S314 includes: judging whether the plant is in the fruit-bearing stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored fourth hyperdimensional vector HV4; if so, it means that the judgment result is that the plant is in the fruit-bearing stage, and then continue to execute step S315; otherwise, judging whether the plant is in the initial flowering stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored third hyperdimensional vector HV3; if so, it means that the judgment result is that the plant is in the initial flowering stage, and then continue to execute step S315; otherwise, judging whether the plant is in the germination stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored first hyperdimensional vector HV1; if so, it means that the judgment result is that the plant is in the germination stage; if not, it means that the judgment result is that the plant is in the seedling stage, and then continue to execute step S315.

[0041] The similarity is cosine similarity. When the cosine similarity is between 0.5 and 1, the judgment result is yes.

[0042] In step S33, the reverse device control module of the device application layer of the edge side deployment subsystem sends the device control instruction to the different manufacturers' device communication protocol converter of the device application layer of the edge side deployment subsystem, and the different manufacturers' device communication protocol converter converts the device control instruction.

[0043] The present invention's intelligent agricultural monitoring method based on brain-like hyperdimensional computing introduces brain-like intelligent hyperdimensional computing (HDC) technology, and uses a brain-like intelligent analysis trainer to conduct online real-time learning of plant and environmental feature data collected by sensors, and encodes and saves all information such as plant growth stage features and environmental features into a meaningful and consistent hyperdimensional vector, so that the agricultural monitoring system can generate brain-like intelligent analysis and memory capabilities; at the same time, based on the brain-like memory, the most suitable environmental control model for different growth stages of the plant is given, and control instructions are sent to various controller devices through brain-like intelligent actuators, thereby adjusting and optimizing the plant growth environment such as temperature, humidity, and light. This technology has extremely low hardware performance requirements, simple calculations, and convenient deployment. It can be locally learned and predicted on a single-chip microcomputer, and is particularly suitable for promotion in the cost-constrained intelligent application field of agricultural Internet of Things, thereby effectively reducing agricultural production costs and increasing crop yields.

[0044] The present invention adopts a brain-like model for perception in the agricultural monitoring system and a growth environment regulation model to drive the controller. Under special circumstances such as a small amount of available samples and large noise interference, stable working performance can still be maintained.

[0045] The method of the present invention adopts a layered decoupling architecture when deploying an agricultural monitoring system. A system can be deployed on the edge side, sensor side, and controller side at the same time. It can support various industry scenario application ecosystems and be compatible with the management of sensor ecosystems and controller ecosystems from different manufacturers, thereby greatly reducing the cost of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention is an overall framework diagram of a smart agricultural monitoring system based on brain-like hyperdimensional computing according to an embodiment of the present invention.

[0047] Figure 2 The present invention is an interactive flowchart of a smart agriculture monitoring method based on brain-like hyperdimensional computing according to an embodiment of the present invention.

[0048] Figure 3 It is a hyperdimensional computing training and execution flow chart of a monitoring system of a smart agricultural monitoring method based on brain-like hyperdimensional computing according to an embodiment of the present invention.

[0049] Figure 4 It is an analysis and training flowchart of a smart agricultural monitoring method based on brain-like hyperdimensional computing according to an embodiment of the present invention.

[0050] Figure 5A and Figure 5B It is the one-dimensional transformation principle diagram of image features, where Figure 5A shows the data before conversion, Figure 5B The transformed data are shown.

[0051] Fig. 6A and Figure 6B This is a conversion diagram of an example of feature binarization, where Fig. 6A shows the data before conversion, Figure 6B The transformed data are shown.

[0052] Fig. 7A and Figure 7B is another example of the conversion principle diagram of the binary feature, where Fig. 7A shows the data before conversion, Figure 7B The transformed data are shown.

[0053] Figure 8 It is a downlink execution flow chart of a smart agriculture monitoring method based on brain-like hyperdimensional computing according to an embodiment of the present invention.

[0054] Fig. 9 It is a schematic diagram of the hyperdimensional vector calculation method of the smart agricultural monitoring method based on brain-like hyperdimensional computing of the present invention.

[0055] Fig.10It is a schematic diagram of the principle of the smart agricultural monitoring method based on brain-like hyperdimensional computing of the present invention when generating a corresponding growth environment control model.

[0056] Fig.11 It is a schematic diagram of the principle of the smart agricultural monitoring method based on brain-like hyperdimensional computing of the present invention when generating growth environment control instructions. DETAILED DESCRIPTION

[0057] Below in conjunction with specific embodiment, further elaborate the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.In addition, should be understood that after reading the content of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0058] like Figure 1 FIG. 1 is a general framework diagram of a smart agricultural monitoring system based on brain-like hyperdimensional computing according to an embodiment of the present invention. Figure 1 As shown, the smart agricultural monitoring system based on brain-like hyperdimensional computing includes:

[0059] Smart agriculture monitoring system 1; the smart agriculture monitoring system 1 can be deployed on the edge side, sensor side and controller side, that is, the smart agriculture monitoring system 1 includes an edge side deployment subsystem formed by being deployed on the edge side, a sensor side deployment subsystem formed by being deployed on the sensor side, and a controller side deployment subsystem formed by being deployed on the controller side, thereby forming these three different deployment forms.

[0060] The edge side deployment subsystem of the smart agriculture monitoring system 1 is configured to support the edge side 2 upward, that is, to support various scenario-based application ecosystems; the sensor side deployment subsystem of the smart agriculture monitoring system 1 is configured to be backward compatible with the management sensor side 3, that is, to be compatible with sensor ecosystems of different manufacturers; the controller side deployment subsystem of the smart agriculture monitoring system 1 is configured to be backward compatible with the management controller side 4, that is, to be compatible with the controller ecosystem.

[0061] The module types of these three deployment subsystems may be different, and the sub-functions of the modules are different to achieve different corresponding functions. For example, the low-power wireless communication module of the physical layer, the wireless network device driver module of the basic layer, and the distributed collaborative routing module of the distribution layer need to be deployed in all three forms, but the sub-functions within the modules are different; the device application layer is the most different in the three forms, and each deploys a sensor application module, a controller application module, and an edge device management application module; the brain-like intelligence layer is generally only deployed on the edge side.

[0062] The edge side deployment subsystem includes:

[0063] Physical layer 11: provides different hardware forms such as embedded devices, edge gateway devices, local data center server devices, etc. for edge-side deployment subsystems. In addition to providing traditional computing and storage hardware capabilities, it also includes low-power wireless communication capabilities, brain-like chips / brain-like memory storage bodies and other brain-like intelligent computing and storage capabilities; therefore, the physical layer 11 includes hardware-based device bodies, computing and storage nodes, low-power wireless communication modules, and brain-like chips / brain-like memory storage bodies;

[0064] Basic layer 12: Based on different device hardware forms, it provides various types of operating system kernels such as embedded\distributed; the basic layer 12 includes the operating system kernel corresponding to the hardware form of the physical layer 11 and wireless network device driver modules of different hardware;

[0065] Distribution layer 13: mainly includes distributed collaborative routing between sensor devices, controller devices and edge intelligent devices, as well as distributed file systems and data storage modules;

[0066] Device application layer 14: The device application layer 14 of the edge-side deployment subsystem is configured to provide device application management capabilities such as user sensor / controller device access and reverse device control, and support protocol conversion of devices from different manufacturers; therefore, the device application layer 14 of the edge-side deployment subsystem includes a device monitor, a reverse device control module, and a communication protocol converter for devices from different manufacturers.

[0067] Brain-like intelligence layer 15: provides brain-like intelligence application capabilities such as brain-like hyper-dimensional computing and intelligent algorithm library. Therefore, it includes brain-like intelligence executors and brain-like intelligence analysis trainers. It should be noted that the brain-like intelligence layer 15 is the software part, and the brain-like chip / brain-like memory storage of the physical layer 11 is the hardware part.

[0068] Management layer 16: provides management functions such as flow control, task scheduling, and fault isolation for the entire edge-side deployment subsystem;

[0069] Security layer 17: Provides global security management from the application layer to the infrastructure layer, throughout the entire edge-side deployment subsystem.

[0070] The sensor side deployment subsystem includes:

[0071] Physical layer 11: provides different hardware forms such as embedded devices, edge gateway devices, local data center server devices, etc. for edge-side deployment subsystems. In addition to providing traditional computing and storage hardware capabilities, it also includes low-power wireless communication capabilities, brain-like chips / brain-like memory storage bodies and other brain-like intelligent computing and storage capabilities; therefore, the physical layer 11 includes hardware-based device bodies, computing and storage nodes, low-power wireless communication modules, and brain-like memory storage devices;

[0072] Basic layer 12: Based on different device hardware forms, it provides various types of operating system kernels such as embedded\distributed; the basic layer 12 includes the operating system kernel corresponding to the hardware form of the physical layer 11 and wireless network device driver modules of different hardware;

[0073] Distribution layer 13: mainly includes distributed collaborative routing modules between sensor devices, controller devices and edge intelligent devices, as well as distributed file systems and data storage modules;

[0074] Device application layer 14: the device application layer 14 of the sensor side deployment subsystem provides sensor device application services such as cameras, plant patches, leaf surface temperature and humidity, and light intensity. That is, the device application layer 14 includes acquisition modules of camera sensors, plant patch sensors, leaf surface temperature and humidity sensors, light intensity sensors, soil comprehensive sensors, and carbon dioxide sensors;

[0075] Management layer 16: provides management functions such as flow control, task scheduling, and fault isolation for the entire edge-side deployment subsystem;

[0076] Security layer 17: Provides global security management from the application layer to the infrastructure layer, throughout the entire edge-side deployment subsystem.

[0077] The controller-side deployment subsystem includes:

[0078] Physical layer 11: provides different hardware forms such as embedded devices, edge gateway devices, local data center server devices, etc. for edge-side deployment subsystems. In addition to providing traditional computing and storage hardware capabilities, it also includes low-power wireless communication capabilities, brain-like chips / brain-like memory storage bodies and other brain-like intelligent computing and storage capabilities; therefore, the physical layer 11 includes hardware-based device bodies, computing and storage nodes, low-power wireless communication modules, and brain-like memory storage devices;

[0079] Basic layer 12: Based on different device hardware forms, it provides various types of operating system kernels such as embedded\distributed; the basic layer 12 includes the operating system kernel corresponding to the hardware form of the physical layer 11 and wireless network device driver modules of different hardware;

[0080] Distribution layer 13: mainly includes distributed collaborative routing modules between sensor devices, controller devices and edge intelligent devices, as well as distributed file systems and data storage modules;

[0081] Device application layer 14: The device application layer 14 of the controller-side deployment subsystem provides controller device application services such as blackout roller blinds, fill light, exhaust fan, etc.; that is, the device application layer 14 includes a blackout roller blind controller, a fill light controller, an exhaust fan controller, a sprinkler / drip irrigation water pump controller, and a driving module for a carbon dioxide generator.

[0082] Management layer 16: provides management functions such as flow control, task scheduling, and fault isolation for the entire edge-side deployment subsystem;

[0083] Security layer 17: Provides global security management from the application layer to the infrastructure layer, throughout the entire edge-side deployment subsystem.

[0084] like Figure 2 and Figure 3 As shown, the present invention provides a smart agriculture monitoring method based on brain-like hyperdimensional computing, which includes:

[0085] Step S1: Execute the uplink monitoring process, including:

[0086] Step S11: using the device application layer of the sensor side deployment subsystem to drive various sensors to collect plant feature data and environmental feature data;

[0087] Step S12: using the distribution layer, the base layer and the physical layer of the sensor side deployment subsystem, the plant feature data and the environmental feature data are transmitted from the device application layer of the sensor side deployment subsystem to the device application layer of the edge side deployment subsystem, and then the device application layer of the edge side deployment subsystem (specifically, the device communication protocol converter of different manufacturers of the device application layer) is used to perform data preprocessing;

[0088] Among them, data preprocessing mainly includes: 1) conversion of communication protocols of equipment from different manufacturers, such as conversion of manufacturer's private protocols into standard protocols such as MQTT; 2) data cleaning and standardization followed by conversion into a unified data format within the system.

[0089] Step S13: using the device application layer of the edge-side deployment subsystem (specifically, using the device monitor of the device application layer) to detect whether its brain-like intelligence layer is in the analysis and training state or the control and execution state; if it is detected that the brain-like intelligence layer is in the analysis and training state, the plant feature data and the environmental feature data are sent to the brain-like intelligence analysis trainer of the brain-like intelligence layer; if it is detected that the brain-like intelligence layer is in the control and execution state, the plant feature data and the environmental feature data are sent to the brain-like intelligence executor of the brain-like intelligence layer;

[0090] It is determined whether the brain-like intelligent layer is in the analysis and training state or the control execution state, which is specifically achieved through the configuration parameters of the edge-side deployment subsystem. In the analysis and training process of the smart agricultural monitoring method based on brain-like hyperdimensional computing of the present invention, the configuration parameters of the edge-side deployment subsystem are configured to the analysis and training state; in the daily operation stage of the analysis and training process of the smart agricultural monitoring method based on brain-like hyperdimensional computing of the present invention, the configuration parameters of the edge-side deployment subsystem are configured to the control execution state.

[0091] Then, if Figure 4 , Figure 8 and Fig. 9 As shown in the figure, the edge-side deployment subsystem executes the analysis training process and the downlink execution process based on the hyperdimensional vector calculation method at its brain-like intelligence layer; wherein the hyperdimensional vector calculation method includes: hyperdimensional vector encoding (Encode), hyperdimensional vector calculation (Compute), hyperdimensional vector matching (Match) and hyperdimensional vector decoding (Decode) 4 parts. Among them, the analysis training process (i.e. model training) requires hyperdimensional vector encoding (Encode) and hyperdimensional vector calculation (Compute) these two parts; the downlink execution process (i.e. reasoning execution) requires hyperdimensional vector encoding (Encode), hyperdimensional vector matching (Match) and hyperdimensional vector decoding (Decode) these three parts.

[0092] Step S2: The edge-side deployment subsystem uses its brain-like intelligent analysis trainer and the brain-like memory storage at the physical layer to perform the analysis and training process based on the hyperdimensional vector calculation method, which specifically includes:

[0093] Step S21: The brain-like intelligent analysis trainer of the brain-like intelligent layer of the edge-side deployment subsystem encodes the plant feature data and environmental feature data collected in batches into hyperdimensional vectors, and then performs intelligent training on the brain-like memory storage of the physical layer of the edge-side deployment subsystem to obtain training results;

[0094] like Figure 4 As shown, the step S21 specifically includes:

[0095] Step S211: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect multiple batches of plant feature data and environmental feature data as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set;

[0096] The purpose of "setting the corresponding hyperdimensional vector label" is for training, so as to match the trained vector model with the actual meaning later.

[0097] Step S212: determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S213. Otherwise, directly execute step S213.

[0098] Among them, if the image feature recognition accuracy is prioritized, convolutional neural network training can be used; but if low power consumption is prioritized, pulse neural network training can still be used.

[0099] Step S213: performing feature clustering (i.e., Encode-1 step) on the training sample data collected in multiple batches, completing the one-dimensionalization of image features (i.e., Encode-2 step) and feature binarization (i.e., Encode-3 step);

[0100] Among them, the collected data features are clustered according to different classifications such as plant characteristics / environmental characteristics / pest and disease characteristics, different data types such as binary images / texts, and data value characteristics such as discrete / continuous types.

[0101] For example, for the germination stage of plants, the training sample data is clustered into the following features:

[0102] 1) Plant seed image (vector element 1)

[0103] 2) Plant bud and leaf image (vector element 2)

[0104] 3) Plant true leaf image (vector element 3)

[0105] 4) Plant stage characteristics (vector elements 4 to 6)

[0106] 5) Stage growth environment (vector elements 7 to 10)

[0107] 6) Pest and disease images (vector elements 11, 12)

[0108] 7) Pest and disease control (vector element 13)

[0109] …

[0110] The one-dimensionalization of image features refers to the one-dimensionalization of multi-dimensional binary image data corresponding to image features, such as the one-dimensionalization of a two-dimensional plant seed feature image (3×3 pixels as an example, which can actually reach 4K pixels or even higher). For example, Figure 5A The data shown is converted to Figure 5B Data shown.

[0111] Feature binarization refers to converting the data values ​​of various discrete / continuous features of training sample data into binary values ​​of +1 / -1.

[0112] For example, the binary value in the one-dimensional plant seed feature image is converted into a binary value of +1 / -1, that is, Fig. 6A The data shown is converted to Figure 6B Data shown.

[0113] For example, the characteristic data of the plant growth environment: temperature 15°, soil humidity 30%, air humidity 80%, soil pH 6.5, are binarized into +1 / -1 binary values ​​according to whether they are within the corresponding range, that is, Fig. 7A The data shown is converted to Figure 7B The specific corresponding relationship of the data shown is shown in Table 1 below.

[0114] Table 1 Binarization results corresponding to each vector element

[0115]

[0116] Step S214: feature classification of all vector elements according to the germination stage, seedling stage, initial flowering stage and fruiting stage of the plant and vector assembly respectively to obtain assembled first hyperdimensional vector HV1' to assembled fourth hyperdimensional vector HV4' (i.e., Encode-4 step);

[0117] The assembly includes assembling the binarization results of various training sample data (ie, each feature vector obtained after binarization) into the form of a super-dimensional vector.

[0118] For example, plant seed feature image vectors, plant bud and leaf feature image vectors, plant true leaf feature image vectors, plant stage feature vectors, plant growth environment feature vectors, pest and disease image vectors, etc. are assembled into a plant germination stage hyperdimensional vector HV1.

[0119] The above-mentioned Encode-1 step (feature clustering), Encode-2 step (one-dimensionalization of image features), Encode-3 step (feature binarization) and Encode-4 step (vector assembly) together constitute the super-dimensional vector encoding (Encode).

[0120] Step S215: perform super-dimensional vector addition on the assembled first to fourth super-dimensional vectors HV1'-HV4' and the first to fourth super-dimensional vectors HV1-HV4 stored in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, and obtain the training results as new first to fourth super-dimensional vectors HV1"-HV4" (i.e., super-dimensional vector calculation step Compute);

[0121] Among them, the first superdimensional vector (including the assembled first superdimensional vector HVl', the stored first superdimensional vector HVl and the new first superdimensional vector HV1") corresponds to the germination stage of the plant, the second superdimensional vector (including the assembled second superdimensional vector HV2', the stored second superdimensional vector HV2 and the new second superdimensional vector HV2") corresponds to the seedling stage of the plant, the third superdimensional vector (including the assembled third superdimensional vector HV3', the stored third superdimensional vector HV3 and the new third superdimensional vector HV3") corresponds to the early flowering stage of the plant, and the fourth superdimensional vector (including the assembled fourth superdimensional vector HV4', the stored fourth superdimensional vector HV4 and the new fourth superdimensional vector HV4") corresponds to the fruiting stage of the plant.

[0122] In other embodiments, other hyper-dimensional calculations may be performed in addition to hyper-dimensional vector addition.

[0123] For example, Fig. 9 The figure shows the calculation result of performing super-dimensional vector addition on a plurality of assembled first super-dimensional vectors HV1 ′ as new samples and the stored first super-dimensional vector HV1 .

[0124] When performing super-dimensional vector addition on the binarized results of +1 / -1 in the first to fourth super-dimensional vectors, the elements whose intermediate quantities are greater than +1 are reset to +1 in the new first to fourth super-dimensional vectors HV1"~HV4", and the elements whose intermediate quantities are less than -1 are reset to -1;

[0125] In addition, in order to avoid the occurrence of elements with intermediate quantities equal to 0, it is necessary to ensure that the first to fourth super-dimensional vectors HV1'~HV4' assembled as an even number of new samples and the first to fourth super-dimensional vectors HV1~HV4 stored are super-dimensionally added. Specifically, when an odd number of new samples appear, it is possible to wait until an even number of new samples appear before sending them in for super-dimensional calculation.

[0126] Step S22: Store the training results in the form of super-dimensional vectors in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, thereby updating the training results stored in the brain-like memory storage; that is, store the new first to fourth super-dimensional vectors HV1"~HV4" in the brain-like memory storage, thereby updating the first to fourth super-dimensional vectors HV1~HV4 stored in the brain-like memory storage.

[0127] In this embodiment, the hyperdimensional vector labels and feature binarization results corresponding to each vector element are as follows:

[0128] (1) The labels corresponding to the vector elements in the first hyperdimensional vector HV1 of the germination stage are: (+1, +1, +1, -1, -1, +1, -1, ...), including:

[0129] (I) Plant characteristic elements at this germination stage

[0130] Vector element 1: seed image feature: binary image code, the binary image code is obtained by extracting image class features;

[0131] That is, the vector label of vector element 1 is the seed image feature, and the data value of the image class feature of vector element 1 is the binary image code.

[0132] Vector element 2: bud and leaf image features: binary image encoding;

[0133] Vector element 3: True leaf image features: binary image encoding;

[0134] Vector element 4: whether there are buds and leaves, the corresponding binary value is: +1 / -1, yes is +1, no is -1;

[0135] That is, the vector label of vector element 4 is whether there are buds and leaves, and the binarization result of vector element 1 is a binary value of +1 / -1.

[0136] Vector element 5: whether the number of true leaves is less than 5, the corresponding binary value is: +1 / -1, +1 if yes, -1 if no;

[0137] Vector element 6: Whether volatile organic compounds (VOCs) can detect hexanal substances, the corresponding binary value is: +1 / -1.

[0138] (II) Characteristic elements of the growth environment suitable for this germination stage

[0139] Vector element 7: whether the temperature is between 25 and 30°C, binary value: +1 / -1;

[0140] Vector element 8: whether the soil moisture is 40% to 50% of the seed weight, binary value: +1 / -1;

[0141] Vector element 9: whether the air humidity is between 70% and 90%, binary value: +1 / -1;

[0142] Vector element 10: whether the soil pH value is within the range of 5.5 to 7.2 (preferably close to 6.5), binary value: +1 / -1;

[0143] (III) Characteristic elements of pests and diseases at this germination stage

[0144] Vector element 11: cucumber damping-off disease image feature: binary image code, the binary image code is obtained by extracting image class features;

[0145] Vector element 12: Cucumber damping-off image features: binary image encoding;

[0146] Vector element 13: control measures, no need to use fungicides, binary value: +1 / -1;

[0147] …

[0148] (2) The labels corresponding to the vector elements in the second hyperdimensional vector HV2 of the seedling stage are: (+1, -1, -1, -1, +1, -1, ...), including:

[0149] (I) Plant characteristic elements at the seedling stage

[0150] Vector element 1: true leaf image features: binary image encoding;

[0151] Vector element 2: stem image features: binary image encoding;

[0152] Vector element 3: whether the number of true leaves is greater than or equal to 5, binary value: +1 / -1;

[0153] Vector element 4: whether there is a flower, binary value: +1 / -1;

[0154] Vector element 5: whether the volatile organic compounds (VOCs) are hexanal substances, binary value: +1 / -1;

[0155] (II) Characteristic elements of the suitable growth environment at the seedling stage

[0156] Vector element 6: whether the temperature is between 22 and 25°C, binary value: +1 / -1;

[0157] Vector element 7: soil moisture is between 60% and 70%, binary value: +1 / -1;

[0158] Vector element 8: air humidity is between 70% and 90%, binary value: +1 / -1;

[0159] Vector element 9: whether the soil pH value is within the range of 5.5 to 7.2 (preferably close to 6.5), binary value: +1 / -1;

[0160] (III) Characteristic elements of pests and diseases at the seedling stage

[0161] Vector element 10: Cucumber late blight image features: binary image encoding;

[0162] Vector element 11: Cucumber scab image features: binary image encoding;

[0163] …

[0164] (3) The labels corresponding to the vector elements in the third hyperdimensional vector HV3 of the initial blooming stage are (-1, +1, +1, +1, -1, -1, -1, ...), including:

[0165] (I) Plant characteristic elements at the early flowering stage

[0166] Vector element 1: Male flower image features: binary image encoding;

[0167] Vector element 2: Female flower image features: binary image encoding;

[0168] Vector element 3: whether there is a female flower, binary value: +1 / -1;

[0169] Vector element 4: whether the first female flower is seated, binary value: +1 / -1;

[0170] Vector element 5: whether there are aromatic aldehyde-alcohol compounds in volatile organic compounds (VOCs), binary value: +1 / -1;

[0171] (II) Characteristic elements of a suitable growth environment during the initial flowering stage

[0172] Vector element 6: whether the daytime temperature is between 20 and 30°C, binary value: +1 / -1;

[0173] Vector element 7: whether the night temperature is between 10 and 17°C, binary value: +1 / -1;

[0174] Vector element 8: soil moisture is between 80% and 90%, binary value: +1 / -1;

[0175] Vector element 9: whether the air humidity is between 70% and 90%, binary value: +1 / -1;

[0176] Vector element 10: whether the soil pH value is within the range of 5.5 to 7.2 (preferably close to 6.5), binary value: +1 / -1;

[0177] (III) Characteristic elements of pests and diseases in the early flowering stage

[0178] Vector element 11: Cucumber gray mold image features: binary image encoding;

[0179] …

[0180] (4) The labels corresponding to the vector elements in the fourth hyperdimensional vector HV4 of the result stage are: (+1, -1, -1, +1, +1, +1, -1, +1+1…), including:

[0181] (I) Plant characteristic elements at this fruiting stage

[0182] Vector element 1: Fruit image features: binary image encoding

[0183] Vector element 2: whether there is fruit, binary value: +1 / -1

[0184] Vector element 3: Whether the first fruit is seated, binary value: +1 / -1

[0185] Vector element 4: Whether there are nonanal compounds in volatile organic compounds (VOCs), binary value: +1 / -1

[0186] Characteristic elements of a suitable growth environment for this fruiting stage

[0187] Vector element 5: whether the temperature is between 25 and 30 degrees Celsius, binary value: +1 / -1

[0188] Vector element 6: Soil moisture is between 80% and 90%, binary value: +1 / -1

[0189] Vector element 7: Air humidity is between 70% and 90%, binary value: +1 / -1

[0190] Vector element 8: Whether the soil pH value is within the range of 5.5 to 7.2 (preferably close to 6.5), binary value: +1 / -1

[0191] Vector element 9: Is the soil potassium concentration > 550 mg / kg (secondary fertility), binary value: +1 / -1

[0192] Vector element 10: Is soil nitrogen concentration > 1.62 g / kg? Binary value: +1 / -1

[0193] Vector element 11: Is the soil calcium concentration > 700 mg / kg? Binary value: +1 / -1

[0194] Vector element 12: Is the soil magnesium concentration > 200 mg / kg? Binary value: +1 / -1

[0195] Vector element 13: Is the soil phosphorus concentration > 20 mg / kg? Binary value: +1 / -1

[0196] Vector element 14: Is soil organic matter concentration > 30 g / kg? Binary value: +1 / -1

[0197] Characteristic elements of pests and diseases at this fruiting stage

[0198] Vector element 15: Cucumber brown spot image features: binary image encoding

[0199] Vector element 16: Cucumber wilt image features: binary image encoding

[0200] …

[0201] Step S3: After the analysis and training process is completed, the edge-side deployment subsystem executes the downlink execution process based on the hyperdimensional vector calculation method, that is, the edge-side deployment subsystem uses its brain-like intelligent actuator and the brain-like memory storage of the physical layer to obtain the device control instruction based on the hyperdimensional vector calculation method and sends it to the device application layer of the control-side deployment subsystem, and the device application layer of the control-side deployment subsystem executes the device control instruction;

[0202] like Figure 8 As shown, the step S3 specifically includes:

[0203] S31: The brain-like intelligent executor of the brain-like intelligent layer of the edge-side deployment subsystem encodes the currently collected plant feature data and environmental feature data into a current hyper-dimensional vector, performs similarity matching between the current hyper-dimensional vector and the brain-like model corresponding to the training result stored in the brain-like memory storage device, and generates a corresponding growth environment regulation model MV based on the similarity result.

[0204] Among them, the sample data of "multi-batch collection" in the above step S2 is used for training the brain-like intelligent analysis trainer; the sample data of "current collection" here is used for daily reasoning and execution processing of the brain-like intelligent executor.

[0205] The step S31 specifically includes:

[0206] Step S311: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect the plant feature data and environmental feature data collected in the current period as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set;

[0207] Step S312: determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S313. Otherwise, directly execute step S313.

[0208] Step S313: Perform hyperdimensional vector encoding (Encode) on the training sample data collected in the current period, that is, perform feature clustering (i.e., Encode-1 step), one-dimensionalization of image features (i.e., Encode-2 step), feature binarization (i.e., Encode-3 step) and vector assembly (i.e., Encode-4 step) in sequence to obtain the current hyperdimensional vector HV;

[0209] Step S314: performing hyperdimensional vector matching (Match), that is, judging the plant growth stage according to the similarity (such as cosine similarity) between the current hyperdimensional vector HV and the first to fourth hyperdimensional vectors HV1-HV4 stored in the brain-like memory;

[0210] Cosine Similarity is the cosine of the angle between two n-dimensional vectors in n-dimensional space. It is equal to the dot product (vector product) of the two vectors divided by the product of the lengths (or magnitudes) of the two vectors.

[0211] The formula for cosine similarity is:

[0212]

[0213] Wherein, Similarity(A, B) is the cosine similarity between vector A and vector B, and the value range of Similarity(A, B) is [-1, 1], where -1 means completely dissimilar and 1 means completely similar.

[0214] Among them, when the cosine similarity value is 0.5 to 1, it is considered that the two vectors are very similar.

[0215] In step S314, the judgment result of the plant growth stage includes: judging that the plant is in the germination stage, judging that the plant is in the seedling stage, judging that the plant is in the initial flowering stage, and judging that the plant is in the fruiting stage.

[0216] The step S314 includes: judging whether the plant is in the fruit-bearing stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored fourth hyperdimensional vector HV4; if so, it means that the judgment result is that the plant is in the fruit-bearing stage, and then continue to execute step S315; otherwise, judging whether the plant is in the initial flowering stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored third hyperdimensional vector HV3; if so, it means that the judgment result is that the plant is in the initial flowering stage, and then continue to execute step S315; otherwise, judging whether the plant is in the germination stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored first hyperdimensional vector HV1, and then continuing to execute step S315 according to the judgment result (if so, it means that the judgment result is that the plant is in the germination stage, if not, it means that the judgment result is that the plant is in the seedling stage).

[0217] Specifically, if the cosine similarity value is close to 1 (eg, the cosine similarity is 0.5 to 1), it is determined to be yes.

[0218] An example of the calculation result of cosine similarity is as follows:

[0219] HV and HV1″ vector 1 (plant feature part):

[0220] HV1″(-1,-1,-1,-1,+1,-1,+1,-1,+1,+1,-1,+1,-1,+1,-1,-1,+1,-1,+ 1, -1, +1, +1, +1, -1, -1, +1, -1, +1, +1, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)

[0221] HV(-1,-1,-1,+1,+1,-1,+1,-1,+1,+1,-1,+1,-1,+1,-1,-1,+1,-1,+1, -1,+1,-1,+1,+1,-1,+1,-1,+1,+1,+1,0,0,0,0,,0,0,0,0,0,0,0,0,0)

[0222] Cosine similarity calculation:

[0223] (1+1+1-1+1+1+1+1+1+1+1+1+1+1+1+1+1+1+1+1+1-1+1-1+1+1+1+1+1-1+0+0+0+0+0+0+0+0+0+0+0+0+0) / (Γ30*Γ30)=22 / 30=0.733

[0224] It is close to 1, so the growth characteristics of this plant and cucumber in the germination stage are very similar.

[0225] HV and HVl″ sub-vector 2 (pest and disease characteristics):

[0226] HV1″(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,+1,-1,+1,-1,-1,+1,+1,-1,0)

[0227] HV(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,+1,-1,+1,-1,-1,+1,+1,-1,0)

[0228] Cosine similarity calculation:

[0229] (+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+0+1+1-1+1+1+1+1+1+0) / (Γ8*Γ8)=6 / 8=0.75

[0230] It is close to 1, so the characteristics of this pest and disease are very similar to those of cucumber damping-off and cucumber damping-off in the germination stage.

[0231] Then, the super-dimensional vector decoding is performed, such as Fig. 9 As shown, the super-dimensional vector decoding (Decode) includes two parts: Decode-1 generates a growth environment control model MV and Decode-2 generates a growth environment control instruction.

[0232] Step S315: according to the judgment result of the plant growth stage, the growth environment control model MV is generated according to the similarity result and the hyperdimensional vector label between the current hyperdimensional vector and the corresponding hyperdimensional vector in the stored training result (ie, Decode-1 step);

[0233] For example, Fig.10 As shown in Table 2, the plant is judged to be in the germination stage, and the corresponding first growth environment regulation model MV1 is generated based on the similarity result between the current hyperdimensional vector and the corresponding first hyperdimensional vector HV1 (such as the stage growth environment part and the pest and disease part): (-1, -1, +1, +1, -1).

[0234] In addition, in other embodiments, if the plant is judged to be in the seedling stage, a corresponding second growth environment regulation model MV2 is generated; if the plant is judged to be in the early flowering stage, a corresponding third growth environment regulation model MV3 is generated; if the plant is judged to be in the fruiting stage, a corresponding fourth growth environment regulation model MV4 is generated.

[0235] Table 2 Correspondence between MV1 model and HV vector

[0236]

[0237] Step S32: Generate a growth environment control instruction (i.e., Decode-2 step), that is, use the brain-like intelligent actuator of the brain-like intelligent layer of the edge-side deployment subsystem to convert the growth environment control model MV into a device control instruction according to the hyperdimensional vector label, and send it to the device application layer of the edge-side deployment subsystem (specifically, to the reverse device control module of the device application layer);

[0238] For example, for the germination stage of plants, e.g. Fig.11 As shown, the vector instruction decoding is performed on the first growth environment control model MV1:

[0239] Among them, the correspondence between the elements and control instructions of the first growth environment control model MV1, the second growth environment control model MV2, the third growth environment control model MV3, and the fourth growth environment control model MV4 are shown in Tables 3 to 6 below.

[0240] Table 3 Correspondence between elements and control instructions of the first growth environment control model MV1

[0241]

[0242] Table 4 Correspondence between the second growth environment control model MV2 and control instructions

[0243]

[0244] Table 5 Correspondence between the third growth environment control model MV3 and control instructions

[0245]

[0246]

[0247] Table 6 Correspondence between the fourth growth environment control model MV4 and control instructions

[0248]

[0249] Step S33: using the device application layer of the edge-side deployment subsystem to convert the device control instruction, and then sending it to the device application layer of the control-side deployment subsystem;

[0250] In this embodiment, the reverse device control module of the device application layer of the edge side deployment subsystem sends the device control instruction to the different manufacturers' device communication protocol converter of the device application layer of the edge side deployment subsystem, and the different manufacturers' device communication protocol converter converts the device control instruction.

[0251] Step S34: Utilize the device application layer of the control-side deployment subsystem (i.e., the device controller app of different manufacturers) to drive the corresponding device according to the device control instruction to adjust the plant growth environment.

[0252] The above is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. The above embodiments of the present invention can also be modified in various ways. All simple, equivalent changes and modifications made according to the claims and the description of the present invention fall within the scope of protection of the claims of the present invention. The contents not described in detail in the present invention are all conventional technical contents.

Claims

1. A smart agricultural monitoring method based on brain-like hyperdimensional computing, characterized in that: The invention is based on a smart agricultural monitoring system based on brain-like hyperdimensional computing with a sensor side deployment subsystem, an edge side deployment subsystem and a controller side deployment subsystem, and includes: Step S1: Execute the uplink monitoring process, including: Step S11: using the device application layer of the sensor side deployment subsystem to drive various sensors to collect plant feature data and environmental feature data; Step S12: transmitting the plant feature data and the environmental feature data from the device application layer of the sensor side deployment subsystem to the device application layer of the edge side deployment subsystem; Step S13: using the device application layer of the edge-side deployment subsystem to detect whether its brain-like intelligence layer is in an analysis and training state or a control and execution state; if it is detected that the brain-like intelligence layer is in an analysis and training state, the plant feature data and the environmental feature data are sent to the brain-like intelligence analysis trainer of the brain-like intelligence layer; if it is detected that the brain-like intelligence layer is in a control and execution state, the plant feature data and the environmental feature data are sent to the brain-like intelligence executor of the brain-like intelligence layer; Step S2: The edge-side deployment subsystem uses its brain-like intelligent analysis trainer and the brain-like memory storage at the physical layer to execute the analysis training process based on the hyperdimensional vector calculation method; Step S3: After the analysis and training process is completed, the edge-side deployment subsystem uses its brain-like intelligent actuator and the brain-like memory storage of the physical layer to obtain the device control instructions based on the hyperdimensional vector calculation method and sends them to the device application layer of the control-side deployment subsystem. The device application layer of the control-side deployment subsystem executes the device control instructions.

2. The smart agriculture monitoring method based on brain-like hyperdimensional computing according to claim 1 is characterized in that: The step S12 also includes: using the device application layer of the edge side deployment subsystem to perform data preprocessing.

3. The smart agriculture monitoring method based on brain-like hyperdimensional computing according to claim 1 is characterized in that: In step S12, the distribution layer, the base layer and the physical layer of the sensor side deployment subsystem are used to transmit the plant feature data and the environmental feature data from the device application layer of the sensor side deployment subsystem to the device application layer of the edge side deployment subsystem.

4. The smart agriculture monitoring method based on brain-like hyperdimensional computing according to claim 1 is characterized in that: The step S2 specifically includes: Step S21: The brain-like intelligent analysis trainer of the brain-like intelligent layer of the edge-side deployment subsystem encodes the plant feature data and environmental feature data collected in batches into hyperdimensional vectors, and then performs intelligent training on the brain-like memory storage of the physical layer of the edge-side deployment subsystem to obtain training results; Step S22: Storing the training results in the form of hyperdimensional vectors in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, thereby updating the training results stored in the brain-like memory storage.

5. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 4 is characterized in that: The step S21 specifically includes: Step S211: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect multiple batches of plant feature data and environmental feature data as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set; Step S212: determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S213. Otherwise, directly execute step S213. Step S213: performing feature clustering on the training sample data collected in multiple batches, completing one-dimensionalization and feature binarization of image features; Step S214: feature classification of all vector elements according to the germination stage, seedling stage, initial flowering stage and fruiting stage of the plant and vector assembly respectively to obtain assembled first hyperdimensional vectors HV1' to assembled fourth hyperdimensional vectors HV4'; Step S215: perform superdimensional vector addition on the assembled first to fourth superdimensional vectors HV1'-HV4' and the first to fourth superdimensional vectors HV1-HV4 stored in the brain-like memory storage of the physical layer of the edge-side deployment subsystem, and obtain the training results as new first to fourth superdimensional vectors HV1"-HV4"; In step S22, the stored training results include the stored first to fourth superdimensional vectors HV1~HV4, wherein the first superdimensional vector corresponds to the germination stage of the plant, the second superdimensional vector corresponds to the seedling stage of the plant, the third superdimensional vector corresponds to the early flowering stage of the plant, and the fourth superdimensional vector corresponds to the fruiting stage of the plant.

6. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 1 is characterized in that: The step S3 specifically includes: Step S31: The brain-like intelligent actuator of the brain-like intelligent layer of the edge-side deployment subsystem encodes the currently collected plant feature data and environmental feature data into a current hyperdimensional vector, performs similarity matching between the current hyperdimensional vector and the brain-like model corresponding to the training result stored in the brain-like memory storage, and generates a corresponding growth environment regulation model according to the similarity result; Step S32: using the brain-like intelligent actuator of the brain-like intelligent layer of the edge-side deployment subsystem, convert the growth environment regulation model into device control instructions according to the hyperdimensional vector label, and send them to the device application layer of the edge-side deployment subsystem; Step S33: using the device application layer of the edge-side deployment subsystem to convert the device control instruction, and then sending it to the device application layer of the control-side deployment subsystem; Step S34: using the device application layer of the control-side deployment subsystem to drive the corresponding device according to the device control instruction to adjust the plant growth environment.

7. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 6 is characterized in that: The step S31 specifically includes: Step S311: using the brain-like intelligence analysis trainer of the brain-like intelligence layer to collect the plant feature data and environmental feature data collected in the current period as training sample data, each type of training sample data is used as a vector element, and a corresponding hyperdimensional vector label is set; Step S312: determine whether the training sample data is image training data. If so, extract image features through a convolutional neural network or a spiking neural network, and then execute step S313. Otherwise, directly execute step S313. Step S313: performing feature clustering, one-dimensionalization of image features, feature binarization, and vector assembly on the training sample data collected in the current period in sequence to obtain a current hyperdimensional vector HV; Step S314: judging the plant growth stage according to the similarity between the current hyperdimensional vector HV and the training results stored in the brain-like memory; Step S315: Generate a growth environment control model according to the judgment result of the plant growth stage and the similarity result and hyperdimensional vector label between the current hyperdimensional vector and the corresponding hyperdimensional vector in the stored training result.

8. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 7 is characterized in that: The stored training results include the first to fourth super-dimensional vectors HV1-HV4, wherein the first super-dimensional vector corresponds to the germination stage of the plant, the second super-dimensional vector corresponds to the seedling stage of the plant, the third super-dimensional vector corresponds to the initial flowering stage of the plant, and the fourth super-dimensional vector corresponds to the fruiting stage of the plant; In step S314, the determination result of the plant growth stage includes: determining that the plant is in the germination stage, determining that the plant is in the seedling stage, determining that the plant is in the initial flowering stage, and determining that the plant is in the fruiting stage; The step S314 includes: judging whether the plant is in the fruit-bearing stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored fourth hyperdimensional vector HV4; if so, it means that the judgment result is that the plant is in the fruit-bearing stage, and then continue to execute step S315; otherwise, judging whether the plant is in the initial flowering stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored third hyperdimensional vector HV3; if so, it means that the judgment result is that the plant is in the initial flowering stage, and then continue to execute step S315; otherwise, judging whether the plant is in the germination stage according to the similarity between the current hyperdimensional vector HV and the plant characteristic part of the stored first hyperdimensional vector HV1; if so, it means that the judgment result is that the plant is in the germination stage; if not, it means that the judgment result is that the plant is in the seedling stage, and then continue to execute step S315.

9. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 8 is characterized in that: The similarity is cosine similarity. When the cosine similarity is between 0.5 and 1, the judgment result is yes.

10. The method for intelligent agricultural monitoring based on brain-like hyperdimensional computing according to claim 6 is characterized in that: In step S33, the reverse device control module of the device application layer of the edge side deployment subsystem sends the device control instruction to the different manufacturers' device communication protocol converter of the device application layer of the edge side deployment subsystem, and the different manufacturers' device communication protocol converter converts the device control instruction.