Environmental status analysis method and system based on smart agriculture

Abstract

The present application provides an environmental status analysis method and system based on smart agriculture, which obtains prior farmland environmental progress data and knowledge information of multiple reference environmental status label decision networks, and then generates a specific focused environmental progress knowledge vector for each environmental status label. Then, by integrating the knowledge information and environmental characteristic pattern information of different reference environmental status label decision networks, special network knowledge information is generated for each environmental status label, and then integrated to configure a target environmental status label decision network for predicting the environmental status of a specific farmland area. Thus, the prior farmland environmental progress data and multi-network information are combined to improve the accuracy and adaptability of environmental status analysis, which helps to manage farmland more accurately, thereby optimizing agricultural production and achieving sustainable development of smart agriculture.

Description

Technical Field

[0001] The present application relates to the field of smart agriculture technology, and specifically, to an environmental status analysis method and system based on smart agriculture. Background Art

[0002] With the continuous advancement of agricultural technology, smart agriculture has become an important direction for the development of modern agriculture. In smart agriculture, accurate analysis of farmland environmental status is crucial, which can help farmers better manage farmland and improve crop yield and quality. However, current environmental status analysis methods still have some challenges and limitations.

[0003] Traditional environmental status analysis methods usually rely on manual observation and simple sensor data, which is not only inefficient but also has limited accuracy. In addition, due to the complexity and variability of farmland environment, a single data source or simple analysis method often fails to fully reflect the true status of farmland. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides an environmental status analysis method based on smart agriculture, the method comprising:

[0005] Acquire a plurality of prior farmland environmental progress data and network knowledge information of a plurality of reference environmental state label decision networks, wherein the reference environmental state label decision network is used to execute a first environmental state label prediction process for a first farmland area, the first environmental state label prediction process corresponds to a plurality of environmental state labels, and the prior farmland environmental progress data represent farmland environmental progress data of a second farmland area in a past monitoring period;

[0006] Determine, based on the prior farmland environmental progress data, focused environmental progress knowledge vectors corresponding to the multiple environmental status labels, respectively, wherein the focused environmental progress knowledge vectors represent environmental feature pattern information corresponding to the prior farmland environmental progress data under the environmental status labels;

[0007] Integrate network knowledge information of multiple reference environment state label decision networks according to the focused environment progress knowledge vector corresponding to the x-th environment state label to generate first network knowledge information corresponding to the x-th environment state label, where x is a positive integer;

[0008] Converting the environmental feature pattern information corresponding to the focused environment progress knowledge vector corresponding to the x-th environmental state label into second network knowledge information corresponding to the x-th environmental state label;

[0009] The first network knowledge information and the second network knowledge information are integrated to generate target network knowledge information corresponding to the x-th environmental state label, and a target environmental state label decision network is configured according to the target network knowledge information corresponding to each environmental state label. The target environmental state label decision network is used to execute a second environmental state label prediction process for the second farmland area. The first environmental state label prediction process and the second environmental state label prediction process correspond to the same multiple environmental state labels.

[0010] In a possible implementation of the first aspect, the priori farmland environmental progress data corresponds to environmental state label annotation data, and the environmental state label annotation data represents the environmental state label corresponding to the priori farmland environmental progress data;

[0011] The step of determining the focused environment progress knowledge vectors corresponding to the plurality of environment state labels respectively according to the prior farmland environment progress data comprises:

[0012] For each type of environmental indicator in the plurality of prior farmland environmental progress data, using a time domain analysis method and a frequency domain analysis method to extract environmental indicator statistical features of each type of environmental indicator as shallow semantic features;

[0013] Extracting deep semantic features from the plurality of prior farmland environment progress data by a time series analysis method, wherein the deep semantic features include trend change features and periodic fluctuation features;

[0014] Based on the shallow semantic features and the deep semantic features, construct a corresponding environment state progress vector for each prior farmland environment, wherein the environment state progress vector is used to describe the dynamic changes of the farmland environment;

[0015] Clustering the environmental state progress vectors of the plurality of prior farmland environmental progress data according to the environmental state label annotation data to generate environmental state progress vector clusters corresponding to the plurality of environmental state labels respectively;

[0016] A focused environment progress knowledge vector corresponding to an x-th environment state label is determined according to an environment state progress vector in an x-th environment state progress vector cluster, wherein the x-th environment state progress vector cluster corresponds to the x-th environment state label.

[0017] In a possible implementation manner of the first aspect, determining the focused environment progress knowledge vector corresponding to the x-th environment state label according to the environment state progress vector in the x-th environment state progress vector cluster includes:

[0018] The environment state progress vector in the x-th environment state progress vector cluster is loaded into the attention network to generate a focused environment progress knowledge vector corresponding to the x-th environment state label.

[0019] In a possible implementation manner of the first aspect, the integrating the first network knowledge information and the second network knowledge information to generate the target network knowledge information corresponding to the x-th environment state tag includes:

[0020] Obtaining network participation coefficients corresponding to the first network knowledge information and the second network knowledge information;

[0021] The first network knowledge information and the second network knowledge information corresponding to the x-th environmental state label are fused according to the network participation coefficient to generate the target network knowledge information corresponding to the x-th environmental state label.

[0022] In a possible implementation manner of the first aspect, the acquiring network participation coefficients corresponding to the first network knowledge information and the second network knowledge information includes:

[0023] Acquire a first training scale corresponding to the training sample data scheduled by the plurality of reference environment state label decision networks during the knowledge learning process, and a second training scale corresponding to the plurality of prior farmland environment progress data;

[0024] The scale numerical relationship between the first training scale and the second training scale is converted into the network participation coefficient according to a predefined conversion strategy.

[0025] In a possible implementation manner of the first aspect, integrating network knowledge information of multiple reference environment state tag decision networks according to the focused environment progress knowledge vector corresponding to the x-th environment state tag to generate first network knowledge information corresponding to the x-th environment state tag includes:

[0026] According to the knowledge connection parameter between the focused environment progress knowledge vector of the x-th environment state label and the network knowledge information, obtaining the knowledge connection degree corresponding to each of the x-th environment state label and the plurality of network knowledge information;

[0027] The plurality of network knowledge information are fused using the knowledge connection degree as a fusion coefficient to generate the first network knowledge information of the x-th environmental state label.

[0028] In a possible implementation of the first aspect, acquiring the knowledge connection degrees respectively corresponding to the x-th environment state label and the plurality of network knowledge information according to the knowledge connection parameter between the focused environment progress knowledge vector of the x-th environment state label and the network knowledge information includes:

[0029] Extracting key features from the plurality of network knowledge information, wherein the key features are used to reflect the core content and structure of the network knowledge information, specifically including key weight features and key bias parameter features used for predicting environmental state labels in the network knowledge information;

[0030] Encode the extracted key features, use discretization or quantization methods to encode continuous features in the key features, and use one-hot encoding or label encoding for categorical features in the key features;

[0031] Mapping the encoded encoding feature to a hash value of a fixed length through a hash function, wherein the hash value serves as a knowledge identification feature of the network knowledge information;

[0032] According to the correlation between the focused environment progress knowledge vector of the x-th environment state label and the y-th knowledge identification feature, the knowledge connection degree between the x-th environment state label and the y-th network knowledge information is generated, where y is a positive integer.

[0033] In a possible implementation of the first aspect, the target network knowledge information is generated by an adaptive learning network, and the adaptive learning network is generated by knowledge learning of a basic adaptive learning network, and the knowledge learning step of the basic adaptive learning network includes:

[0034] Acquire network knowledge information, auxiliary training sample data sequences and verification training sample data sequences of the multiple reference environment state label decision networks, wherein the auxiliary training sample data sequences and the verification training sample data sequences include template farmland environment progress data, and the template farmland environment progress data corresponds to template environment state label data;

[0035] Generate target template network knowledge information corresponding to the plurality of environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence;

[0036] Using the target template network knowledge information as the initial network knowledge information corresponding to the first basic neural network, and using the auxiliary training sample data sequence to perform cyclic knowledge learning on the initial network knowledge information of the first basic neural network to generate a second basic neural network;

[0037] Obtaining environment state label learning data of the verification training sample data sequence on the second basic neural network;

[0038] According to the data difference between the environment state label learning data and the template environment state label data, the basic adaptive learning network is subjected to cyclic knowledge learning to generate the adaptive learning network.

[0039] In a possible implementation manner of the first aspect, performing cyclic knowledge learning on the basic adaptive learning network according to the data difference between the environment state label learning data and the template environment state label data to generate the adaptive learning network includes:

[0040] Determining the training focused environment progress knowledge vectors corresponding to the multiple environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence;

[0041] Determining a first training cost according to a data difference between the training focused environment progress knowledge vector and the template focused environment progress knowledge vector;

[0042] Determining a second training cost according to a data difference between the environment state label learning data and the template environment state label data;

[0043] The basic adaptive learning network is subjected to cyclic knowledge learning according to the first training cost and the second training cost to generate the adaptive learning network.

[0044] On the other hand, an embodiment of the present application also provides an environmental status analysis system based on smart agriculture, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0045] Based on the above aspects, by integrating the network knowledge information of multiple prior farmland environmental progress data and reference environmental status label decision networks, an efficient and accurate analysis of farmland environmental status is achieved. Specifically, by integrating the network knowledge information of multiple reference environmental status label decision networks and combining them with prior farmland environmental progress data, target network knowledge information corresponding to each environmental status label is generated, making full use of the information of multiple reference environmental status label decision networks and data sources, thereby improving the accuracy of environmental status analysis. By integrating knowledge information and environmental feature pattern information from different reference networks, the configured target environmental status label decision network has stronger generalization ability, can better adapt to different farmland environments and conditions, and provide more reliable environmental status label predictions. As a result, farmland managers can monitor the farmland environment more effectively and adjust management measures in a timely manner, thereby improving crop yield and quality and reducing agricultural production risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the execution flow of the environmental status analysis method based on smart agriculture provided in an embodiment of the present application.

[0047] Figure 2 This is a schematic diagram of the hardware architecture of the environment status analysis system based on smart agriculture provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an environmental status analysis method based on smart agriculture provided by an embodiment of the present application. The environmental status analysis method based on smart agriculture is introduced in detail below.

[0049] Step S110, obtaining network knowledge information of multiple prior farmland environmental progress data and multiple reference environmental status label decision networks, wherein the reference environmental status label decision network is used to execute a first environmental status label prediction process for a first farmland area, wherein the first environmental status label prediction process corresponds to multiple environmental status labels, and the prior farmland environmental progress data represents the farmland environmental progress data of the second farmland area during a past monitoring period.

[0050] In detail, the prior farmland environmental progress data refers to environmental data collected in a specific farmland area (second farmland area) over a period of time in the past, covering a variety of farmland environmental indicators, such as soil composition, meteorological conditions, pests and diseases, etc., and used to reflect the dynamic changes in the farmland environment. For example, a farmland monitoring system collects soil moisture, temperature, pH value, nutrient content, and weather conditions (such as rainfall, wind speed, temperature, etc.) of a certain farmland every day in the past year. These data constitute the prior farmland environmental progress data, reflecting the changes in the farmland environment over the past period of time.

[0051] The reference environment state label decision network refers to a trained neural network model that can predict and classify the complex state of the farmland environment based on the input environmental data, that is, give the corresponding environmental state label. These reference environment state label decision networks have been used for similar prediction and classification tasks for farmland environments. For example, a trained neural network model can predict whether the farmland is in a state of "soil nutrient imbalance" based on input soil data, meteorological data, etc. This neural network model is a reference environment state label decision network that can provide a reference framework for understanding the possible state of the current farmland environment.

[0052] The environmental status label is a concise description of the current state of the farmland environment, such as "early warning of soil nutrient imbalance", "risk of pest and disease outbreak", etc. These environmental status labels are predicted by the reference environmental status label decision network based on the input data.

[0053] That is, in this embodiment, the server obtains a large amount of prior farmland environmental progress data from the farmland monitoring system, which covers multiple dimensions such as soil chemical composition, microbial activity, meteorological data, and the occurrence of pests and diseases. At the same time, the server accesses the network knowledge information of multiple trained reference environmental state label decision networks, which are originally used to predict and classify complex states similar to farmland environments.

[0054] Step S120, determining the focused environment progress knowledge vectors corresponding to the multiple environmental status labels respectively based on the prior farmland environmental progress data, wherein the focused environment progress knowledge vectors represent the environmental feature pattern information corresponding to the prior farmland environmental progress data under the environmental status label.

[0055] In detail, the focused environmental progress knowledge vector refers to a set of features closely related to a specific environmental state label extracted from prior farmland environmental progress data, capturing the main characteristics and changing trends of the farmland environment under a specific environmental state.

[0056] The environmental characteristic pattern information refers to the unique characteristics and patterns exhibited by the farmland environment under specific environmental conditions, which are obtained by analyzing prior farmland environmental progress data and are used to construct a focused environmental progress knowledge vector.

[0057] In this embodiment, the server processes the prior farmland environmental progress data. For each environmental status label, such as "early warning of soil nutrient imbalance", the server can mine a feature set closely related to the environmental status label from the prior farmland environmental progress data to form a focused environmental progress knowledge vector, which may include abnormal fluctuations in specific nutrients in the soil, changing patterns of microbial activities, etc.

[0058] Step S130, integrating network knowledge information of multiple reference environment state tag decision networks according to the focused environment progress knowledge vector corresponding to the x-th environment state tag, and generating first network knowledge information corresponding to the x-th environment state tag, where x is a positive integer.

[0059] The network knowledge information refers to the parameter information of the decision network by referring to the environment state label. In the network, this network knowledge information can exist in the form of weights, biases or other parameters, representing the knowledge learned by the decision network by referring to the environment state label.

[0060] The first network knowledge information is information generated by integrating network knowledge information of multiple reference environment state tag decision networks and based on a focused environment progress knowledge vector of a specific environment state tag (such as the xth one).

[0061] In this embodiment, for each complex environmental state label, the server can use its corresponding focused environmental progress knowledge vector to filter and integrate network knowledge information of multiple reference environmental state label decision networks. For example, for the label "early warning of soil nutrient imbalance", the server can find out the network knowledge information that can effectively identify the precursors of nutrient imbalance in the reference environmental state label decision network, and merge these network knowledge information to form the first network knowledge information that is highly relevant to the environmental state label.

[0062] Step S140: converting the environment feature pattern information corresponding to the focused environment progress knowledge vector corresponding to the x-th environment state tag into second network knowledge information corresponding to the x-th environment state tag.

[0063] In this embodiment, the server further converts the environmental feature pattern information in the focused environment progress knowledge vector into second network knowledge information suitable for neural network processing, and can convert complex feature patterns into high-dimensional vectors or embedded representations so that the neural network can understand and utilize this information.

[0064] Step S150, integrate the first network knowledge information and the second network knowledge information to generate the target network knowledge information corresponding to the x-th environmental state label, and configure the target environmental state label decision network according to the target network knowledge information corresponding to each environmental state label, the target environmental state label decision network is used to execute the second environmental state label prediction process for the second farmland area, the first environmental state label prediction process and the second environmental state label prediction process correspond to the same multiple environmental state labels.

[0065] Finally, the server combines the first network knowledge information with the second network knowledge information to generate target network knowledge information for each complex environmental state label. Using this target network knowledge information, the server can configure a more powerful and accurate target environmental state label decision network, which will be able to comprehensively consider data from multiple dimensions and accurately predict and classify complex states of farmland environments, such as "early warning of soil nutrient imbalance" and "risk of pest and disease outbreaks". Such prediction capabilities are of great significance for taking timely agricultural management measures and improving crop yields and quality.

[0066] Based on the above steps, by integrating the network knowledge information of multiple prior farmland environmental progress data and reference environmental status label decision networks, an efficient and accurate analysis of farmland environmental status is achieved. Specifically, by integrating the network knowledge information of multiple reference environmental status label decision networks and combining the prior farmland environmental progress data, the target network knowledge information corresponding to each environmental status label is generated, and the information of multiple reference environmental status label decision networks and data sources is fully utilized, thereby improving the accuracy of environmental status analysis. By integrating the knowledge information and environmental feature pattern information from different reference networks, the configured target environmental status label decision network has stronger generalization ability, can better adapt to different farmland environments and conditions, and provide more reliable environmental status label predictions. As a result, farmland managers can monitor the farmland environment more effectively and adjust management measures in a timely manner, thereby improving crop yield and quality and reducing agricultural production risks.

[0067] In a possible implementation, the priori farmland environmental progress data corresponds to environmental status label annotation data, and the environmental status label annotation data represents the environmental status label corresponding to the priori farmland environmental progress data.

[0068] Step S120 includes:

[0069] Step S121, for each type of environmental indicators in the plurality of prior farmland environmental progress data, extracting environmental indicator statistical features of each type of environmental indicators as shallow semantic features using a time domain analysis method and a frequency domain analysis method.

[0070] In this embodiment, the server has acquired a large amount of prior farmland environmental progress data, which cover multiple dimensions such as soil chemical composition, microbial activity, meteorological data, etc., and these prior farmland environmental progress data correspond to environmental status label annotation data. These environmental status label annotation data clearly point out the environmental status corresponding to each piece of environmental progress data (such as "early warning of soil nutrient imbalance", etc.).

[0071] Therefore, for each type of environmental indicator (such as soil moisture, temperature, pH value, etc.), the server uses time domain analysis methods and frequency domain analysis methods to extract the statistical features of these indicators. In time domain analysis, the server can calculate the average value, maximum value, minimum value, standard deviation, etc. of the indicator; in frequency domain analysis, the server can analyze the periodic changes of these indicators or the components of specific frequencies through methods such as Fourier transform. These extracted statistical features are regarded as shallow semantic features because they directly reflect the basic properties and change laws of environmental indicators.

[0072] For example, for the environmental indicator of soil moisture, the server can analyze its time domain characteristics such as the average value and fluctuation range over a period of time, as well as frequency domain characteristics such as whether there are obvious periodic changes (such as daily changes, seasonal changes), etc.

[0073] Step S122, extracting deep semantic features from the plurality of prior farmland environment progress data by a time series analysis method, wherein the deep semantic features include trend change features and periodic fluctuation features.

[0074] In this embodiment, in addition to shallow semantic features, the server also extracts deep semantic features through time series analysis methods, which mainly reflect the long-term change trend and periodic fluctuation pattern of environmental indicators. For example, the server can use methods such as ARIMA model and LSTM neural network to predict and analyze the future change trend of environmental indicators, or identify the seasonal fluctuation pattern in environmental indicators.

[0075] For example, taking soil nutrient content as an example, the server can discover an obvious trend: as crops grow, certain nutrients in the soil (such as nitrogen and phosphorus) will gradually decrease; at the same time, the server may also identify periodic changes in nutrient content due to agricultural activities such as fertilization and irrigation.

[0076] Step S123: constructing a corresponding environment state progression vector for each prior farmland environment based on the shallow semantic features and the deep semantic features, wherein the environment state progression vector is used to describe the dynamic changes of the farmland environment.

[0077] Specifically, the server can concatenate or fuse the shallow semantic features and deep semantic features of various environmental indicators to form a long vector, which not only contains the current status information of the environmental indicators (such as mean value, standard deviation, etc.), but also includes information such as their changing trends and periodic fluctuation patterns.

[0078] Step S124 , clustering the environmental state progress vectors of the plurality of prior farmland environmental progress data according to the environmental state label annotation data, to generate environmental state progress vector clusters corresponding to the plurality of environmental state labels respectively.

[0079] Next, the server can cluster these environmental state progress vectors using the environmental state label annotation data. The purpose of clustering is to classify environmental state progress vectors with similar environmental states into one category, thereby forming multiple environmental state progress vector clusters. Each cluster corresponds to a specific environmental state label (such as "early warning of soil nutrient imbalance", etc.).

[0080] In the clustering process, the server can use K-means, hierarchical clustering and other algorithms. By continuously adjusting the clustering parameters and thresholds, the server can ensure that the environmental progress vectors within each cluster have a high degree of similarity, while the environmental progress vectors between different clusters have obvious differences.

[0081] Step S125, determining a focused environment progress knowledge vector corresponding to the xth environment state tag according to the environment state progress vector in the xth environment state progress vector cluster, wherein the xth environment state progress vector cluster corresponds to the xth environment state tag.

[0082] Finally, for each environmental state progress vector cluster, the server can determine a focused environmental progress knowledge vector, which is a representative representation of all environmental progress vectors within the environmental state progress vector cluster, capturing the main characteristics and changing trends of the farmland environment under the environmental state.

[0083] In order to obtain the focused environment progress knowledge vector, the server can average or weighted average all environment progress vectors in the environment state progress vector cluster. In addition, the server can also use some more complex machine learning algorithms (such as principal component analysis PCA or autoencoder Autoencoder) to extract the main features of the environment progress vector in the cluster, thereby forming a more compact and effective focused environment progress knowledge vector.

[0084] As a result, the server can extract valuable information and features from a large amount of prior farmland environmental progress data, providing strong support for subsequent environmental status prediction and classification tasks.

[0085] In a possible implementation, step S125 includes: loading the environment state progress vector in the x-th environment state progress vector cluster into the attention network, and generating a focused environment progress knowledge vector corresponding to the x-th environment state label.

[0086] In this embodiment, the server starts a pre-trained attention network model, which has the ability to process environmental state progress vectors. The server passes the environmental state progress vector in the x-th environmental state progress vector cluster loaded into the memory as input data to the attention network. The attention network begins to process these input data and identifies which parts of these environmental state progress vectors are the most important, that is, the parts that need to be "focused" by calculating the attention weights.

[0087] After the attention network processes the input environmental state progress vector, it will output a focused environmental progress knowledge vector, which highlights the most important and relevant information in the original data, while weakening the unimportant parts. After receiving this focused environmental progress knowledge vector, the server will store it in the database for subsequent analysis, prediction or decision-making. This focused environmental progress knowledge vector is a refinement and extraction of the original environmental state data, which can more effectively support the work in application fields such as environmental protection, weather forecasting, and urban planning.

[0088] Therefore, the server can generate a focused environment progress knowledge vector corresponding to the x-th environment state label based on the environment state progress vector in the x-th environment state progress vector cluster through the processing of the attention network, and can extract the most critical and valuable information from a large amount of environment state data, providing strong data support for various environment-related applications.

[0089] In a possible implementation, step S150 may include:

[0090] Step S151: Obtain network participation coefficients corresponding to the first network knowledge information and the second network knowledge information.

[0091] In this embodiment, before the server starts to integrate the first network knowledge information and the second network knowledge information, it first needs to obtain the network participation coefficients corresponding to the two network knowledge information. These network participation coefficients are calculated by a specific algorithm or model, reflecting the importance and contribution of each type of network knowledge information in the integration process. The server retrieves these coefficients from the relevant data storage or computing module to ensure that they are up-to-date and optimized for the current environment state tag (i.e., the xth environment state tag).

[0092] Step S152: fusing the first network knowledge information and the second network knowledge information corresponding to the x-th environmental state tag according to the network participation coefficient to generate the target network knowledge information corresponding to the x-th environmental state tag.

[0093] In this embodiment, once the server obtains the network participation coefficient, it can start to fuse the knowledge information. In this process, the server can perform weighted averaging or other forms of fusion on the first network knowledge information and the second network knowledge information according to the network participation coefficient. Specifically, if the network participation coefficient of a certain network knowledge information is higher, then its influence in the fusion process will be greater. The server ensures that the two network knowledge information can be reasonably integrated according to their importance through precise mathematical operations.

[0094] After the above fusion process, the server finally generates the target network knowledge information corresponding to the x-th environmental state tag. The target network knowledge information is obtained by fusing the first network knowledge information and the second network knowledge information. It is more comprehensive and accurate and can better reflect the characteristics and trends of the current environmental state. The server can store this target network knowledge information in a database and may send it to other systems or users for further analysis and application.

[0095] As a result, the server can effectively integrate the first network knowledge information and the second network knowledge information to generate target network knowledge information corresponding to a specific environmental status label (the xth environmental status label). This process not only improves the comprehensiveness and accuracy of the information, but also provides strong data support for subsequent environmental monitoring, prediction and decision-making.

[0096] In a possible implementation, step S151 includes:

[0097] Step S1511, obtaining a first training scale corresponding to the training sample data scheduled by the multiple reference environment state label decision networks during the knowledge learning process, and a second training scale corresponding to the multiple prior farmland environment progress data.

[0098] In this embodiment, a large amount of training sample data is stored in the data storage of the server, and these training sample data are used to train the reference environment state label decision network. The server will first access these training sample data and count the number of training sample data used by the reference environment state label decision network in the knowledge learning process, that is, the first training scale, which reflects the degree of dependence and usage of the decision network on data during the training process.

[0099] In addition to the above training sample data, the server also stores a large amount of prior farmland environment progress data, which is based on actual observation and recording of the farmland environment and is of great significance for understanding and predicting changes in the farmland environment. The server can count the number of these prior farmland environment progress data, i.e., the second training scale, which reflects the richness and availability of the actual environment data.

[0100] Step S1512: converting the scale numerical relationship between the first training scale and the second training scale into the network participation coefficient according to a predefined conversion strategy.

[0101] After obtaining the first training scale and the second training scale, the server will convert the numerical relationship between the two scales into a network participation coefficient according to a predefined conversion strategy. The conversion strategy can take into account multiple factors such as the quality, quantity, and distribution of the training data to ensure that the converted network participation coefficient can reasonably reflect the importance and contribution of the two types of knowledge information in the integration process. After the conversion is completed, the server obtains the network participation coefficient for subsequent knowledge information integration.

[0102] As a result, the server can obtain the network participation coefficients corresponding to the first network knowledge information and the second network knowledge information based on the scale of the training data and the predefined conversion strategy. These coefficients will serve as an important basis for subsequent knowledge information integration to ensure the accuracy and reliability of the integration results.

[0103] In a possible implementation, step S130 may include:

[0104] Step S131 , obtaining the knowledge connection degrees respectively corresponding to the x-th environment state tag and the plurality of network knowledge information according to the knowledge connection parameters between the focused environment progress knowledge vector of the x-th environment state tag and the network knowledge information.

[0105] In this embodiment, the server first obtains the focused environment progress knowledge vector corresponding to the x-th environment state label, which is obtained through a series of processing steps and contains key information about the current environment state. Next, the server can traverse the network knowledge information of multiple reference environment state label decision networks stored in the network. For each piece of network knowledge information, the server calculates the knowledge connection parameters between the network knowledge information and the focused environment progress knowledge vector. After the calculation is completed, the server obtains the knowledge connection degree corresponding to the x-th environment state label and multiple network knowledge information.

[0106] Step S132: fuse the plurality of network knowledge information by taking the knowledge connection degree as a fusion coefficient to generate the first network knowledge information of the x-th environmental state label.

[0107] After obtaining the degree of knowledge connection, the server can use these knowledge connection degrees as fusion coefficients to fuse multiple network knowledge information. Specifically, the server can perform weighted averaging or other forms of fusion operations on each network knowledge information according to the degree of knowledge connection corresponding to each network knowledge information. In this way, the network knowledge information with a higher correlation with the x-th environmental state label will occupy a greater weight in the fusion process, thereby ensuring that the generated first network knowledge information is more accurate and reliable. After the fusion is completed, the server obtains the first network knowledge information corresponding to the x-th environmental state label. This first network knowledge information integrates the wisdom of multiple reference environmental state label decision networks, providing strong data support for subsequent environmental monitoring, prediction and decision-making.

[0108] Therefore, the server can integrate the network knowledge information of multiple reference environment state label decision networks based on the focused environment progress knowledge vector corresponding to the x-th environment state label, and generate the first network knowledge information corresponding to the x-th environment state label. This process fully considers the correlation between different network knowledge information and the current environment state, ensuring the accuracy and reliability of the generated information.

[0109] In a possible implementation, step S131 may include:

[0110] Step S1311, extracting key features from the multiple network knowledge information, wherein the key features are used to reflect the core content and structure of the network knowledge information, specifically including key weight features and key bias parameter features in the network knowledge information for predicting environmental state labels.

[0111] In this embodiment, the server will first extract the key features of each network knowledge information from the multiple stored network knowledge information. These key features mainly include key weight features and key bias parameter features used for environmental state label prediction. These features can reflect the core content and structure of the network knowledge information and are the basis for subsequent processing.

[0112] For example, the server can use specific layer outputs in a deep learning model or model parameters as key features, which capture important information and judgment basis of the model when processing environmental states.

[0113] Step S1312, encoding the extracted key features, using discretization or quantization methods to encode continuous features in the key features, and using one-hot encoding or label encoding to encode categorical features in the key features.

[0114] After extracting the key features, the server can perform key encoding on these features for subsequent processing. For continuous features, such as specific values ​​of weight values ​​or bias parameters, the server can use discretization or quantization methods to encode and convert these continuous values ​​into a series of discrete labels or levels. For categorical features, such as different types of environmental status labels or classification identifiers of network knowledge information, the server can use One-Hot Encoding or Label Encoding to convert them into numerical forms that can be processed by computers.

[0115] Step S1313, mapping the encoded encoding feature to a hash value of a fixed length through a hash function, and the hash value serves as a knowledge identification feature of the network knowledge information.

[0116] After the encoding is completed, the server can map the encoding features to a fixed-length hash value through a hash function. The hash value is used as the knowledge identification feature of the network knowledge information. It is unique and stable and can represent the core content and structure of the network knowledge information. The design of the hash function ensures that similar network knowledge information will obtain similar hash values, while network knowledge information with large differences will obtain hash values ​​with large differences.

[0117] Step S1314, generating the knowledge connection degree between the xth environmental state label and the yth network knowledge information according to the correlation between the focused environmental progress knowledge vector of the xth environmental state label and the yth knowledge identification feature, where y is a positive integer.

[0118] Finally, the server can calculate the correlation between the focused environment progress knowledge vector of the x-th environment state label and the y-th knowledge identification feature. This correlation can be obtained by calculating the similarity, distance or other statistical indicators between the two vectors. For example, the server can use cosine similarity to measure the angle between the two vectors to determine their similarity. The calculated correlation value represents the degree of knowledge connection between the x-th environment state label and the y-th network knowledge information. The larger the value, the closer the connection between the two; otherwise, it means that the connection between the two is relatively distant.

[0119] As a result, the server can accurately quantify the degree of connection between different environmental status labels and network knowledge information, providing an important basis for subsequent knowledge integration and decision support.

[0120] In a possible implementation, the target network knowledge information is generated by an adaptive learning network, and the adaptive learning network is generated by knowledge learning of a basic adaptive learning network. The knowledge learning steps of the basic adaptive learning network include:

[0121] Step A110, obtaining network knowledge information, auxiliary training sample data sequence and verification training sample data sequence of the multiple reference environment state label decision networks, the auxiliary training sample data sequence and the verification training sample data sequence include template farmland environment progress data, and the template farmland environment progress data corresponds to template environment state label data.

[0122] In this embodiment, the server first obtains network knowledge information from multiple reference environment state label decision networks. These network knowledge information are the knowledge and experience accumulated by each reference environment state label decision network in the previous learning process. At the same time, the server also obtains two sets of data sequences: auxiliary training sample data sequence and verification training sample data sequence. Both sets of data sequences contain template farmland environment progress data, and these template farmland environment progress data correspond to template environment state label data. These template environment state label data will provide important reference and verification basis for subsequent knowledge learning and network generation.

[0123] Step A120: Generate target template network knowledge information corresponding to the multiple environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence.

[0124] After obtaining the necessary network knowledge information and data sequences, the server can use the basic adaptive learning network, combined with the auxiliary training sample data sequence, to start generating target template network knowledge information corresponding to multiple environmental state labels. In this process, the basic adaptive learning network will extract the core knowledge and features related to each environmental state label through learning and analyzing the auxiliary training sample data sequence, and then form the target template network knowledge information. These target template network knowledge information provide an important foundation for the initialization and learning of the subsequent neural network.

[0125] Step A130, using the target template network knowledge information as the initial network knowledge information corresponding to the first basic neural network, and using the auxiliary training sample data sequence to perform cyclic knowledge learning on the initial network knowledge information of the first basic neural network to generate a second basic neural network.

[0126] The server can use the target template network knowledge information as the initial network knowledge information of the first basic neural network. Then, the auxiliary training sample data sequence is used to perform cyclic knowledge learning on the first basic neural network. In this process, the first basic neural network will continuously learn and adjust its own parameters and structure to better adapt to and fit the characteristics and rules in the auxiliary training sample data sequence. After multiple cycles of learning, the first basic neural network will evolve into the second basic neural network, and its network knowledge and performance will be significantly improved.

[0127] Step A140, obtaining environment state label learning data of the verification training sample data sequence on the second basic neural network.

[0128] After generating the second basic neural network, the server can apply it to the verification training sample data sequence to obtain environment state label learning data, which reflects the recognition and prediction ability of the second basic neural network for the environment state labels in the verification training sample data sequence. By analyzing and evaluating these environment state label learning data, the server can understand the performance and accuracy of the second basic neural network.

[0129] Step A150, performing cyclic knowledge learning on the basic adaptive learning network according to the data difference between the environmental state label learning data and the template environmental state label data, to generate the adaptive learning network.

[0130] Finally, the server can further perform cyclic knowledge learning on the basic adaptive learning network based on the data difference between the environment state label learning data and the template environment state label data. In this process, the basic adaptive learning network will continuously adjust and optimize its own network structure and parameter configuration according to the feedback information in the data difference to improve its recognition and prediction accuracy of the environment state label. After multiple cycles of learning and optimization, the basic adaptive learning network will eventually generate an adaptive learning network with stronger adaptability and accuracy, which will be used in the subsequent generation of target network knowledge information and decision support.

[0131] In a possible implementation, step A150 includes:

[0132] Step A151, determining the training focused environment progress knowledge vectors corresponding to the multiple environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence.

[0133] In this embodiment, the server first uses the basic adaptive learning network to process the auxiliary training sample data sequence. For each environmental state label, the server extracts the feature information related to the environmental state label through the network's forward propagation process, and forms the corresponding training focused environmental progress knowledge vector. These training focused environmental progress knowledge vectors are the network's understanding and representation of the current environmental state, which will be used in subsequent training and optimization processes.

[0134] Step A152: determining a first training cost according to the data difference between the training focused environment progress knowledge vector and the template focused environment progress knowledge vector.

[0135] After obtaining the training focused environment progress knowledge vector, the server will compare it with the predefined template focused environment progress knowledge vector. By calculating the difference or distance between the two, such as using indicators such as Euclidean distance and cosine similarity, the server can quantify the deviation between the training focused environment progress knowledge vector and the template. This deviation is the first training cost, which reflects the accuracy of the network in understanding the environment state. The larger the cost, the greater the difference between the network and the template.

[0136] Step A153, determining a second training cost based on the data difference between the environment state label learning data and the template environment state label data.

[0137] Next, the server can compare the difference between the environment state label learning data and the template environment state label data. This difference can be obtained by calculating the classification accuracy, cross entropy loss, etc. It reflects the performance of the network in predicting the environment state label. The difference is defined as the second training cost. The higher the cost, the worse the prediction performance of the network, and further optimization is needed.

[0138] Step A154: performing cyclic knowledge learning on the basic adaptive learning network according to the first training cost and the second training cost to generate the adaptive learning network.

[0139] After determining the first training cost and the second training cost, the server can use this cost information to perform cyclic knowledge learning on the basic adaptive learning network. Specifically, the server can use the back-propagation algorithm to update the weights and bias parameters of the network according to the gradient information calculated by the cost function to reduce the training cost and improve the performance of the network. This process will be repeated for multiple iterations until the network performance reaches the preset standard or the training cost converges. Ultimately, the basic adaptive learning network that has undergone cyclic knowledge learning will evolve into an adaptive learning network with stronger adaptability and accuracy. The adaptive learning network will be able to better understand and predict environmental state labels, providing strong support for subsequent tasks.

[0140] Figure 2 The hardware structure of the environment state analysis system 100 based on smart agriculture for implementing the above-mentioned environment state analysis method based on smart agriculture provided in the embodiment of the present application is shown. Figure 2 As shown, the environment status analysis system 100 based on smart agriculture may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0141] In a possible design, the environment state analysis system 100 based on smart agriculture can be a single server or a server group. The server group can be centralized or distributed (for example, the environment state analysis system 100 based on smart agriculture can be a distributed system). In some embodiments, the environment state analysis system 100 based on smart agriculture can be local or remote. For example, the environment state analysis system 100 based on smart agriculture can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the environment state analysis system 100 based on smart agriculture can be directly connected to a machine-readable storage medium 120 to access stored information and / or data. In some embodiments, the environment state analysis system 100 based on smart agriculture can be implemented on an environment state analysis system based on smart agriculture. As an example only, the environment state analysis system based on smart agriculture can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0142] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the environment state analysis system 100 based on smart agriculture uses to execute or use to complete the exemplary method described in this application.

[0143] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the environmental status analysis method based on smart agriculture in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0144] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned environmental status analysis system 100 based on smart agriculture. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0145] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned environmental status analysis method based on smart agriculture is implemented.

[0146] It should be noted that in order to simplify the description disclosed in the present application and thereby facilitate the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, various features are sometimes grouped into one embodiment, drawing or description thereof.

Claims

1. A method for analyzing environmental status based on smart agriculture, characterized in that: The method comprises: Acquire multiple prior farmland environmental progress data and network knowledge information of multiple reference environmental status label decision networks, the reference environmental status label decision network is used to execute a first environmental status label prediction process for a first farmland area, the first environmental status label prediction process corresponds to multiple environmental status labels, the prior farmland environmental progress data characterizes the farmland environmental progress data of a second farmland area in a past monitoring period, the network knowledge information refers to parameter information obtained through the reference environmental status label decision network, the network knowledge information exists in the form of weights and biases, and represents the knowledge learned by the reference environmental status label decision network, the prior farmland environmental progress data covers multiple environmental indicators of the second farmland area, the environmental indicators include soil composition, meteorological conditions, and pest and disease conditions, and are used to reflect the dynamic changes of the farmland environment, the reference environmental status label decision network refers to a trained neural network model that can predict and classify environmental status labels of the farmland environment according to input environmental data; Determine, based on the prior farmland environmental progress data, focused environmental progress knowledge vectors corresponding to the multiple environmental status labels, respectively, wherein the focused environmental progress knowledge vector represents environmental characteristic pattern information corresponding to the prior farmland environmental progress data under the environmental status label, wherein the environmental characteristic pattern information refers to the unique characteristics and patterns exhibited by the farmland environment under a specific environmental state, and is obtained by analyzing the prior farmland environmental progress data, and is used to construct the focused environmental progress knowledge vector; Integrate network knowledge information of multiple reference environment state label decision networks according to the focused environment progress knowledge vector corresponding to the x-th environment state label to generate first network knowledge information corresponding to the x-th environment state label, where x is a positive integer; Converting the environmental feature pattern information corresponding to the focused environment progress knowledge vector corresponding to the x-th environmental state label into second network knowledge information corresponding to the x-th environmental state label; The first network knowledge information and the second network knowledge information are integrated to generate target network knowledge information corresponding to the x-th environmental state label, and a target environmental state label decision network is configured according to the target network knowledge information corresponding to each environmental state label. The target environmental state label decision network is used to execute a second environmental state label prediction process for the second farmland area. The first environmental state label prediction process and the second environmental state label prediction process correspond to the same multiple environmental state labels.

2. The environmental status analysis method based on smart agriculture according to claim 1 is characterized in that: The prior farmland environmental progress data corresponds to environmental status label annotation data, and the environmental status label annotation data represents the environmental status label corresponding to the prior farmland environmental progress data; The step of determining the focused environment progress knowledge vectors corresponding to the plurality of environment state labels respectively according to the prior farmland environment progress data comprises: For each type of environmental indicator in the plurality of prior farmland environmental progress data, using a time domain analysis method and a frequency domain analysis method to extract environmental indicator statistical features of each type of environmental indicator as shallow semantic features; Extracting deep semantic features from the plurality of prior farmland environment progress data by a time series analysis method, wherein the deep semantic features include trend change features and periodic fluctuation features; Based on the shallow semantic features and the deep semantic features, construct a corresponding environment state progress vector for each prior farmland environment, wherein the environment state progress vector is used to describe the dynamic changes of the farmland environment; Clustering the environmental state progress vectors of the plurality of prior farmland environmental progress data according to the environmental state label annotation data to generate environmental state progress vector clusters corresponding to the plurality of environmental state labels respectively; A focused environment progress knowledge vector corresponding to an x-th environment state label is determined according to an environment state progress vector in an x-th environment state progress vector cluster, wherein the x-th environment state progress vector cluster corresponds to the x-th environment state label.

3. The environmental status analysis method based on smart agriculture according to claim 2 is characterized in that: The step of determining the focused environment progress knowledge vector corresponding to the x-th environment state label according to the environment state progress vector in the x-th environment state progress vector cluster includes: The environment state progress vector in the x-th environment state progress vector cluster is loaded into the attention network to generate a focused environment progress knowledge vector corresponding to the x-th environment state label.

4. The method for analyzing environmental status based on smart agriculture according to any one of claims 1 to 3, characterized in that: The step of integrating the first network knowledge information and the second network knowledge information to generate target network knowledge information corresponding to the x-th environment status tag includes: Obtaining network participation coefficients corresponding to the first network knowledge information and the second network knowledge information; The first network knowledge information and the second network knowledge information corresponding to the x-th environmental state label are fused according to the network participation coefficient to generate the target network knowledge information corresponding to the x-th environmental state label.

5. The environmental status analysis method based on smart agriculture according to claim 4 is characterized in that: The obtaining of the network participation coefficients corresponding to the first network knowledge information and the second network knowledge information includes: Acquire a first training scale corresponding to the training sample data scheduled by the plurality of reference environment state label decision networks during the knowledge learning process, and a second training scale corresponding to the plurality of prior farmland environment progress data; The scale numerical relationship between the first training scale and the second training scale is converted into the network participation coefficient according to a predefined conversion strategy.

6. The method for analyzing environmental status based on smart agriculture according to any one of claims 1 to 3, characterized in that: The step of integrating network knowledge information of multiple reference environment state tag decision networks according to the focused environment progress knowledge vector corresponding to the x-th environment state tag to generate first network knowledge information corresponding to the x-th environment state tag includes: According to the knowledge connection parameter between the focused environment progress knowledge vector of the x-th environment state label and the network knowledge information, obtaining the knowledge connection degree corresponding to each of the x-th environment state label and the plurality of network knowledge information; The plurality of network knowledge information are fused using the knowledge connection degree as a fusion coefficient to generate the first network knowledge information of the x-th environmental state label.

7. The environmental status analysis method based on smart agriculture according to claim 6 is characterized in that: The step of obtaining the knowledge connection degree corresponding to each of the x-th environment state label and the plurality of network knowledge information according to the knowledge connection parameter between the focused environment progress knowledge vector of the x-th environment state label and the network knowledge information comprises: Extracting key features from the plurality of network knowledge information, wherein the key features are used to reflect the core content and structure of the network knowledge information, specifically including key weight features and key bias parameter features used for predicting environmental state labels in the network knowledge information; Encode the extracted key features, use discretization or quantization methods to encode continuous features in the key features, and use one-hot encoding or label encoding for categorical features in the key features; Mapping the encoded encoding feature to a hash value of a fixed length through a hash function, wherein the hash value serves as a knowledge identification feature of the network knowledge information; According to the correlation between the focused environment progress knowledge vector of the x-th environment state label and the y-th knowledge identification feature, the knowledge connection degree between the x-th environment state label and the y-th network knowledge information is generated, where y is a positive integer.

8. The method for analyzing environmental status based on smart agriculture according to any one of claims 1 to 3, characterized in that: The target network knowledge information is generated by an adaptive learning network, and the adaptive learning network is generated by knowledge learning of a basic adaptive learning network. The knowledge learning steps of the basic adaptive learning network include: Acquire network knowledge information, auxiliary training sample data sequences and verification training sample data sequences of the multiple reference environment state label decision networks, wherein the auxiliary training sample data sequences and the verification training sample data sequences include template farmland environment progress data, and the template farmland environment progress data corresponds to template environment state label data; Generate target template network knowledge information corresponding to the plurality of environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence; Using the target template network knowledge information as the initial network knowledge information corresponding to the first basic neural network, and using the auxiliary training sample data sequence to perform cyclic knowledge learning on the initial network knowledge information of the first basic neural network to generate a second basic neural network; Obtaining environment state label learning data of the verification training sample data sequence on the second basic neural network; According to the data difference between the environment state label learning data and the template environment state label data, the basic adaptive learning network is subjected to cyclic knowledge learning to generate the adaptive learning network.

9. The method for analyzing environmental status based on smart agriculture according to claim 8, characterized in that: The step of performing cyclic knowledge learning on the basic adaptive learning network based on the data difference between the environmental state label learning data and the template environmental state label data to generate the adaptive learning network includes: Determining the training focused environment progress knowledge vectors corresponding to the multiple environment state labels respectively using the basic adaptive learning network according to the auxiliary training sample data sequence; Determining a first training cost according to a data difference between the training focused environment progress knowledge vector and the template focused environment progress knowledge vector; Determining a second training cost according to a data difference between the environment state label learning data and the template environment state label data; The basic adaptive learning network is subjected to cyclic knowledge learning according to the first training cost and the second training cost to generate the adaptive learning network.

10. An environmental status analysis system based on smart agriculture, characterized in that: The environmental status analysis system based on smart agriculture includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the environmental status analysis method based on smart agriculture as described in any one of claims 1 to 9 above.

Images