Field edge cooperative low-power consumption diagnosis method and system

Abstract

The application provides a field edge cooperative low-power consumption diagnosis method and system. The method comprises the following steps: dividing a target field block into a plurality of field micro-zones, and constructing a field micro-zone health state atlas; generating an abnormal wake-up factor based on the correlation risks in the basic perception data, the node energy consumption state and the field micro-zone health state atlas; determining a hierarchical collection strategy, and collecting crop state data according to the hierarchical collection strategy; performing lightweight diagnosis on the crop state data on the low-power consumption perception node side to generate a candidate diagnosis result and a diagnosis evidence summary; generating an edge cooperative diagnosis confidence, when the edge cooperative diagnosis confidence does not satisfy a confirmation condition, generating a minimum supplementary collection instruction based on diagnosis uncertainty and a new energy consumption cost, and updating the edge cooperative diagnosis confidence according to the supplementary collection result; and outputting a field diagnosis result carrying an evidence chain according to the updated edge cooperative diagnosis confidence. The application can reduce node power consumption, improve weak network diagnosis real-time performance and improve abnormal recognition accuracy.

Description

Technical Field

[0001] This application relates to the field of agricultural intelligent monitoring technology, and in particular to a field edge collaborative low-power diagnostic method and system. Background Technology

[0002] With the development of smart agriculture and field IoT technology, online monitoring of crop growth status, pest and disease risks, and water and fertilizer stress using sensors, image acquisition devices, and edge gateways has become an important means to improve the accuracy of field management. Existing technologies typically collect environmental data or crop images periodically at field nodes and upload the collected data to a cloud platform for centralized identification, or deploy lightweight models on a single edge device to make preliminary judgments on local crop anomalies.

[0003] However, field monitoring equipment largely relies on batteries or solar power. Continuous high-power data collection and frequent communication significantly reduce node endurance. Simultaneously, unstable field network coverage makes centralized cloud-based diagnosis susceptible to communication interruptions and transmission delays, failing to meet real-time early warning requirements. Furthermore, abnormal phenotypes such as crop diseases, pests, water shortages, fertilizer deficiencies, and environmental burns share similarities, making misjudgment prone to occur with single-node or single-modal identification. Existing edge diagnostic methods typically lack the synergistic utilization of field spatial correlations, agricultural activity correlations, and anomaly propagation relationships, making cross-validation between micro-regions difficult. Diagnostic models and data collection strategies are also often fixed configurations, unable to adaptively adjust based on node energy consumption, changes in field risk, and diagnostic feedback, resulting in deficiencies in field diagnostic accuracy, low-power operation capabilities, and result traceability. Summary of the Invention

[0004] In view of this, embodiments of this application provide a field edge collaborative low-power diagnostic method and system to solve the problems of high power consumption acquisition, weak network diagnosis lag, and single-point identification misjudgment in the prior art.

[0005] A first aspect of this application provides a field edge collaborative low-power diagnostic method, comprising: dividing a target field into multiple field micro-regions, and constructing a field micro-region health status map based on spatial association, agricultural association, and historical anomaly association of each field micro-region; acquiring basic sensing data through low-power sensing nodes deployed in the field micro-regions, and generating anomaly wake-up factors based on the basic sensing data, node energy consumption status, and association risks in the field micro-region health status map; determining a hierarchical acquisition strategy for the corresponding low-power sensing nodes based on the anomaly wake-up factors and energy consumption constraints, and acquiring crop status data according to the hierarchical acquisition strategy; and performing processing on the crop status data at the low-power sensing nodes. Lightweight diagnosis generates candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis. A field edge gateway determines the collaborative diagnostic range of the target field micro-region based on the field micro-region health status map, and merges the diagnostic evidence summaries from the target field micro-region and the collaborative diagnostic range to generate a marginal collaborative diagnostic confidence score. When the marginal collaborative diagnostic confidence score does not meet the confirmation criteria, a minimum supplementary sampling instruction is generated based on diagnostic uncertainty and additional energy consumption cost, and the marginal collaborative diagnostic confidence score is updated according to the supplementary sampling results. Based on the updated marginal collaborative diagnostic confidence score, field diagnostic results carrying the evidence chain are output, and the field micro-region health status map and diagnostic control parameters are updated based on diagnostic feedback.

[0006] A second aspect of this application provides a field edge collaborative low-power diagnostic system, comprising: a construction module for dividing a target field into multiple field micro-regions and constructing a field micro-region health status map based on spatial association, agricultural association, and historical anomaly association of each field micro-region; a generation module for acquiring basic sensing data through low-power sensing nodes deployed in the field micro-regions and generating anomaly wake-up factors based on the basic sensing data, node energy consumption status, and association risks in the field micro-region health status map; a collection module for determining a hierarchical collection strategy for the corresponding low-power sensing nodes based on the anomaly wake-up factors and energy consumption constraints, and collecting crop status data according to the hierarchical collection strategy; and a diagnostic module for processing crop status data at the low-power sensing nodes. The system performs lightweight diagnosis, generates candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis; a fusion module is used to determine the collaborative diagnostic range of the target field micro-region based on the field micro-region health status map through the field edge gateway, and merges the diagnostic evidence summaries of the target field micro-region and the collaborative diagnostic range to generate edge collaborative diagnostic confidence; an update module is used to generate a minimum supplementary sampling instruction based on diagnostic uncertainty and additional energy consumption cost when the edge collaborative diagnostic confidence does not meet the confirmation conditions, and update the edge collaborative diagnostic confidence according to the supplementary sampling results; an output module is used to output the field diagnostic results carrying the evidence chain based on the updated edge collaborative diagnostic confidence, and update the field micro-region health status map and diagnostic control parameters based on diagnostic feedback.

[0007] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: The target field is divided into multiple micro-regions, and a health status map of each micro-region is constructed based on spatial correlation, agricultural correlation, and historical anomaly correlation. Basic sensing data is acquired through low-power sensing nodes deployed in the micro-regions. Anomaly wake-up factors are generated based on the basic sensing data, node energy consumption status, and correlation risks in the micro-region health status map. A hierarchical acquisition strategy for the corresponding low-power sensing nodes is determined based on the anomaly wake-up factors and energy consumption constraints, and crop status data is collected according to the hierarchical acquisition strategy. Lightweight diagnostics of the crop status data are performed on the low-power sensing node side, generating candidate diagnostic results and... This invention provides a summary of diagnostic evidence to characterize the diagnostic criteria. It determines the collaborative diagnostic range of a target field micro-region based on the field micro-region health status map via a field edge gateway, and merges the diagnostic evidence summaries from the target field micro-region and the collaborative diagnostic range to generate an edge collaborative diagnostic confidence score. When the edge collaborative diagnostic confidence score does not meet the confirmation criteria, a minimum supplementary sampling instruction is generated based on diagnostic uncertainty and additional energy consumption costs, and the edge collaborative diagnostic confidence score is updated according to the supplementary sampling results. Based on the updated edge collaborative diagnostic confidence score, a field diagnostic result carrying the evidence chain is output, and the field micro-region health status map and diagnostic control parameters are updated based on the diagnostic feedback. This application can reduce node power consumption, improve the real-time performance of weak network diagnosis, and enhance the accuracy of anomaly identification. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the field edge collaborative low-power diagnostic method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the field edge collaborative low-power diagnostic system provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] With the development of smart agriculture and field IoT technology, existing technologies typically involve deploying sensors, image acquisition devices, or edge gateways in the field to collect crop growth environment data, crop image data, or pest monitoring data at fixed intervals. This collected data is then uploaded to a cloud platform for centralized identification, or a lightweight diagnostic model is deployed on a single edge device to make a preliminary judgment on localized crop anomalies. While these methods can achieve field condition monitoring and pest and disease early warning to some extent, their diagnostic process usually relies on single-point data collection, fixed sampling frequencies, and static model configurations.

[0012] However, field sensing nodes mostly rely on batteries or solar power. Continuous high-frequency data acquisition and high-power data transmission will significantly shorten node battery life. Simultaneously, the unstable field network environment makes centralized cloud-based diagnosis susceptible to communication interruptions, transmission latency, and bandwidth limitations, hindering timely diagnostic results. Furthermore, similarities exist among abnormal phenotypes such as crop diseases, pests, water stress, nutrient abnormalities, and environmental damage, making misjudgments from single nodes or single modalities prone to occur. Therefore, existing technologies suffer from problems such as high-power data acquisition, delayed diagnosis in weak networks, and misjudgments from single-point identification.

[0013] To address the aforementioned issues, this application provides a field edge collaborative low-power diagnostic method. The method divides the target field into multiple field micro-regions and constructs a field micro-region health status map based on spatial, agricultural, and historical anomaly associations among these micro-regions. Basic sensing data is acquired through low-power sensing nodes deployed in the field micro-regions, and anomaly wake-up factors are generated by combining node energy consumption status with associated risks in the field micro-region health status map. A tiered acquisition strategy is then determined based on the anomaly wake-up factors and energy consumption constraints, ensuring that low-power sensing nodes only collect crop status data matching the diagnostic requirements when anomaly triggering conditions are met.

[0014] Furthermore, this application performs lightweight diagnosis on crop status data at the low-power sensing node side, generating candidate diagnostic results and diagnostic evidence summaries. The field edge gateway then determines the collaborative diagnosis range based on the field micro-region health status map, fusing diagnostic evidence summaries from the target field micro-region and the collaborative diagnosis range to generate an edge collaborative diagnosis confidence score. When the edge collaborative diagnosis confidence score does not meet the confirmation criteria, a minimum supplementary sampling instruction is generated based on diagnostic uncertainty and the additional energy consumption cost. The supplementary sampling results are then used to update the edge collaborative diagnosis confidence score, ultimately outputting a field diagnostic result carrying the evidence chain.

[0015] Through the above technical solutions, this application can dynamically control the acquisition intensity based on the field anomaly risk and node energy consumption status, reduce unnecessary high-power acquisition and communication transmission, and lower node power consumption; at the same time, through edge gateways, collaborative diagnosis is performed under the constraints of the field micro-region health status map, reducing reliance on real-time cloud communication and improving the real-time performance of diagnosis in weak network environments; furthermore, through evidence fusion and minimum supplementary acquisition verification between the target field micro-region and related micro-regions, the ability to distinguish similar abnormal phenotypes is improved, thereby enhancing the accuracy of anomaly identification and the traceability of diagnostic results.

[0016] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 This is a schematic flowchart of the field edge collaborative low-power diagnostic method provided in an embodiment of this application. Figure 1 As shown, the method may specifically include: S101, the target field is divided into multiple field micro-regions, and a field micro-region health status map is constructed based on the spatial association, agricultural association and historical abnormal association of each field micro-region; S102 acquires basic sensing data through low-power sensing nodes deployed in field micro-areas, and generates abnormal wake-up factors based on the basic sensing data, node energy consumption status, and associated risks in the field micro-area health status map. S103, determine the hierarchical acquisition strategy for the corresponding low-power sensing node based on the abnormal wake-up factor and energy consumption constraints, and collect crop status data according to the hierarchical acquisition strategy. S104, perform lightweight diagnosis of crop status data on the low-power sensing node side, and generate candidate diagnosis results and a summary of diagnostic evidence to characterize the diagnosis basis; S105, the collaborative diagnosis range of the target field micro-region is determined based on the field micro-region health status map through the field edge gateway, and the diagnostic evidence summary of the target field micro-region and the collaborative diagnosis range is integrated to generate the edge collaborative diagnosis confidence. S106, When the confidence level of edge collaborative diagnosis does not meet the confirmation conditions, generate a minimum supplementary sampling instruction based on the diagnostic uncertainty and the additional energy consumption cost, and update the confidence level of edge collaborative diagnosis according to the supplementary sampling result; S107, output field diagnostic results carrying the chain of evidence based on the updated edge co-diagnostic confidence, and update the field micro-plot health status map and diagnostic control parameters based on diagnostic feedback.

[0018] In some embodiments, the target field is divided into multiple field micro-regions, and a field micro-region health status map is constructed based on the spatial association, agricultural association, and historical anomaly association of each field micro-region, including: Acquire basic management data and historical diagnostic data of the target field, and determine the micro-area division scale according to the field diagnostic accuracy and the coverage capability of low-power sensing nodes; The target field is divided into multiple field micro-regions based on the micro-region division scale, and each field micro-region is configured with a micro-region identifier and a low-power sensing node association identifier. Micro-region association edges are generated based on the spatial proximity between field micro-regions, the consistency of management operations, and the similarity of anomaly propagation, and association weights are assigned to the micro-region association edges; Based on the crop status baseline, environmental status baseline, node credibility, and micro-region association edges of each field micro-region, a field micro-region health status map is constructed.

[0019] Specifically, after the target field is connected to the system, the field edge gateway first reads the basic management data of the field, which includes the field boundary coordinates, planting row direction, crop variety, growth period, irrigation zone, drainage direction, terrain elevation difference, installation location of low power sensing node and nominal coverage radius of node. At the same time, it reads historical diagnostic data, which includes the location of historical pests and diseases, abnormal spread direction, manual verification results and agricultural treatment records.

[0020] For example, taking a tomato planting field with an area of ​​12 hectares as an example, the system generates a field spatial grid based on the field boundary coordinates, and combines the effective coverage radius of the low-power sensing node (25 meters), the early identification accuracy of target lesions, and the width of irrigation branches to determine the micro-area division scale as 30 meters by 30 meters; when there are local topographical undulations or historically high-incidence areas in the field, the system adjusts the micro-area division scale of the corresponding area to 20 meters by 20 meters to improve the resolution of anomaly identification.

[0021] After determining the micro-region division scale, the field edge gateway divides the target field into multiple field micro-regions according to the field spatial grid, and generates a unique micro-region identifier for each micro-region. The micro-region identifier can be generated by combining the field number, row and column coordinates, and growth period batch. For example, the micro-region in the 3rd row and 5th column can be marked as a combination of the field number and the row and column sequence number. The system further associates the node identifier, installation coordinates, node type, power supply status, and communication status of the low-power sensing nodes with the corresponding field micro-regions. When a low-power sensing node covers multiple adjacent field micro-regions, the primary associated micro-region and the secondary associated micro-region are determined according to the coverage area ratio and sampling angle. When a field micro-region is jointly covered by multiple low-power sensing nodes, the node association priority is configured according to the node credibility and spatial distance.

[0022] Subsequently, the system generates micro-region association edges based on the spatial proximity between field micro-regions, consistency of management operations, and similarity of anomaly propagation. Spatial proximity can be determined by the distance between the micro-region center points and the boundary contact relationship; consistency of management operations can be determined by whether they belong to the same irrigation branch, the same fertilization batch, or the same agricultural operation area; and similarity of anomaly propagation can be determined by historical pest and disease spread paths, prevailing wind direction, water flow direction, and the frequency of simultaneous occurrence of anomalies.

[0023] The system converts the above-mentioned correlations into micro-region correlation edges and assigns correlation weights based on the correlation strength. For example, if two adjacent micro-regions are located on the same irrigation branch and have historically exhibited the same type of wilting anomaly multiple times, the system assigns them a higher correlation weight; if two micro-regions are only spatially adjacent but differ in crop variety, growth stage, and irrigation conditions, a lower correlation weight is assigned. The correlation weights can be dynamically adjusted based on seasonal changes, wind direction variations, and agricultural treatment records, ensuring that the correlations between field micro-regions reflect the actual anomaly transmission patterns.

[0024] Furthermore, after configuring the micro-region association edges, the system establishes crop status baselines and environmental status baselines for each field micro-region. The crop status baseline is formed by canopy color, leaf morphology, growth height, and historical health diagnosis results during the normal growth period, while the environmental status baseline is formed by temperature and humidity, soil moisture, light intensity, and leaf wetness duration under normal management conditions in the same micro-region. The system also generates node credibility based on the historical online rate, sampling stability, manual verification consistency rate, and power fluctuation of low-power sensing nodes, and writes the node credibility to the corresponding field micro-region node.

[0025] Ultimately, the field edge gateway uses field micro-regions as graph nodes and micro-region association edges as graph edges, unifying crop status baselines, environmental status baselines, node credibility, and association weights to construct a field micro-region health status graph. This graph provides a structured foundation for subsequent calculation of abnormal awakening factors, determination of collaborative diagnostic range, and fusion of diagnostic evidence, enabling low-power sensing nodes to no longer perform diagnostics in isolation, but to complete edge collaborative judgments under the constraints of field spatial relationships and historical anomaly patterns.

[0026] In some embodiments, basic sensing data is acquired through low-power sensing nodes deployed in field micro-plots. An abnormal wake-up factor is generated based on the basic sensing data, node energy consumption status, and associated risks in the field micro-plot health status map, including: The low-power sensing nodes acquire basic sensing data of the corresponding field micro-regions according to the low duty cycle sampling strategy, and evaluate the temporal stability of the basic sensing data to obtain the micro-region state deviation characterization. Read the node energy consumption status of the low-power sensing node and determine the available energy consumption constraints for the anomaly sensing process based on the node energy consumption status. Based on the field micro-plot health status map, risk transmission information associated with the corresponding field micro-plot is obtained, and associated risk characterization is generated; By integrating micro-area state deviation representation, available energy consumption constraint, and associated risk representation, an abnormal wake-up factor is generated to trigger the hierarchical acquisition strategy.

[0027] Specifically, after the health status map of each field micro-plot is constructed, the low-power sensing nodes deployed in each field micro-plot enter a low duty cycle sampling state. Taking a tomato planting field as an example, each low-power sensing node is woken up every 10 minutes under normal conditions to collect basic sensing data for the corresponding field micro-plot. The basic sensing data includes temperature and humidity, soil moisture, light intensity, leaf surface moisture status, and the node's own operating status. After collection, the low-power sensing node only retains the feature statistics within the current sampling window and immediately returns to a dormant state to reduce continuous working time. For field micro-plots located in areas with historically high anomalies or in meteorological windows prone to disease outbreaks, the field edge gateway can adjust the low duty cycle sampling period to 5 minutes, enabling the basic sensing data to reflect risk changes in a timely manner.

[0028] Low-power sensing nodes locally evaluate the temporal stability of basic sensing data. Specifically, using the most recent six sampling periods as a sliding window, they calculate the magnitude, duration of continuous deviation, and slope of change of environmental characteristics relative to the environmental state baseline, and determine whether the current field micro-plot exhibits an abnormal trend based on the crop state baseline. For example, if leaf moisture is continuously higher than the normal baseline for 40 minutes, air humidity continues to rise, and light intensity decreases, the system identifies this change as a disease-induced environmental deviation; if soil moisture is below the baseline range and temperature rises, the system identifies this change as a water stress deviation. The low-power sensing node integrates the above deviation magnitude, duration, and direction of change to obtain a micro-plot state deviation characterization. The micro-plot state deviation characterization is not a single sampling outlier, but rather describes the degree of temporal deviation of the current field micro-plot relative to the normal state.

[0029] While generating micro-area state deviation representations, low-power sensing nodes read their energy consumption status. This status includes remaining battery power, solar charging status, energy consumption of the most recent communication, sensor wake-up counts, and the availability of high-power modes. The system determines available energy consumption constraints for the anomaly sensing process based on these node energy consumption statuses. For example, when a node has more than 60% remaining battery power and stable communication quality, the available energy consumption constraint allows the node to perform medium-frequency sampling and low-resolution image pre-acquisition; when the node has less than 25% remaining battery power, the constraint restricts the node to only perform basic sensing sampling and postpones high-power image acquisition until the lighting and charging conditions improve. This available energy consumption constraint prevents nodes from frequently activating high-power modes due to a single anomaly detection, ensuring long-term operational stability in the field.

[0030] Furthermore, the field edge gateway obtains risk transmission information associated with the current field micro-region based on the field micro-region health status map. This risk transmission information includes the most recent diagnostic confidence of adjacent field micro-regions, anomaly occurrences in the same irrigation zone, disease risk in upwind field micro-regions, historical abnormal propagation paths, and micro-region association edge weights. Taking the field micro-region in row 3, column 5 as an example, if the upwind adjacent micro-region has already generated a high-confidence candidate diagnosis result for leaf spot disease, and the micro-region association edge weight between the two is high, the field edge gateway will configure the associated risk characterization to a higher level; if the surrounding micro-regions have no anomalies and low historical propagation similarity, even if the current micro-region has a slight humidity deviation, the associated risk characterization will remain at a lower level.

[0031] Subsequently, the system integrates micro-area state deviation representations, available energy consumption constraints, and associated risk representations to generate an anomaly wake-up factor. This integration process can be achieved by combining a marginal rule model with a lightweight scoring model. The micro-area state deviation representation characterizes the intensity of local anomalies, the associated risk representation characterizes the likelihood of external propagation of anomalies, and the available energy consumption constraint corrects the wake-up intensity and acquisition upper limit. For example, when the local deviation is moderate but the neighborhood risk is high and the node has sufficient power, the anomaly wake-up factor is raised to the tiered acquisition trigger range; when the local deviation is high but the node has insufficient power, the anomaly wake-up factor is still generated, but the subsequent tiered acquisition strategy will prioritize low-power supplementary sampling. Thus, the anomaly wake-up factor can simultaneously reflect field anomaly risks, node operational capabilities, and map propagation risks, providing a basis for subsequent dynamic triggering of tiered acquisition strategies, thereby reducing invalid acquisitions and improving the timeliness of anomaly capture.

[0032] In some embodiments, a hierarchical acquisition strategy for the corresponding low-power sensing node is determined based on an abnormal wake-up factor and energy consumption constraints, and crop status data is acquired according to the hierarchical acquisition strategy, including: The abnormal awakening factor is matched with the preset awakening level rule to determine the abnormal triggering level of the corresponding field micro-region; A tiered data acquisition strategy is generated based on the anomaly trigger level, the available energy consumption constraints of low-power sensing nodes, and the diagnostic requirements of crop status data. The low-power sensing nodes are controlled to switch between different acquisition intensities according to the hierarchical acquisition strategy, and the corresponding acquisition time window and acquisition accuracy are determined. Crop status data matching the current anomaly trigger level is obtained based on the acquisition time window and acquisition accuracy.

[0033] Specifically, after the low-power sensing node generates an abnormal wake-up factor, the field edge gateway calls the preset wake-up level rules to match the abnormal wake-up factor to a specific level. The preset wake-up level rules can be pre-configured based on crop variety, growth stage, historical abnormal frequency, and node power supply capacity, or can be dynamically corrected based on diagnostic feedback.

[0034] Taking a tomato planting field as an example, when the abnormal wake-up factor is below 0.30, the system classifies the corresponding field micro-area as normal, and the low-power sensing node maintains a low duty cycle environment for sampling; when the abnormal wake-up factor reaches 0.30 to 0.55, it is classified as a level of concern, and the node increases the sampling frequency of basic sensing data; when the abnormal wake-up factor reaches 0.55 to 0.75, it is classified as a suspected abnormal level, and the node starts low-resolution crop image acquisition and local feature extraction; when the abnormal wake-up factor is above 0.75, it is classified as a high-risk level, and the node starts high-precision crop status acquisition within the allowable energy consumption range and reports the acquisition task status to the field edge gateway.

[0035] After determining the anomaly trigger level, the system generates a tiered acquisition strategy by combining the available energy consumption constraints of low-power sensing nodes and the diagnostic needs of crop status data. Available energy consumption constraints are jointly determined by the node's remaining power, solar power replenishment trends, communication link quality, and the availability of high-power modes. The diagnostic needs for crop status data are determined by the preliminary judgment of the anomaly type. For example, if the current anomaly is mainly manifested as continuously increasing leaf surface moisture and enhanced risk of neighboring diseases, the diagnostic needs will focus on leaf images and changes in canopy moisture; if the current anomaly is mainly manifested as decreasing soil moisture and enhanced risk of plant wilting, the diagnostic needs will focus on continuous changes in soil moisture and changes in canopy morphology. The system selects the corresponding acquisition mode, sampling frequency, and data precision according to different diagnostic needs, avoiding initiating the same high-power acquisition process in all anomaly scenarios.

[0036] During the generation of the tiered data acquisition strategy, the field edge gateway configures a data acquisition intensity switching command for each low-power sensing node. Data acquisition intensity can include sustain sampling, enhanced sampling, low-precision phenotypic acquisition, and high-precision diagnostic acquisition. Sustain sampling is used for the normal level, retaining only basic environmental features; enhanced sampling is used for the level of interest, shortening the sampling cycle based on the original low duty cycle sampling; low-precision phenotypic acquisition is used for the suspected anomaly level, acquiring low-resolution leaf or canopy images and extracting color shift, spot candidate regions, and morphological change features at the node level; high-precision diagnostic acquisition is used for the high-risk level, acquiring crop status data that meets the input requirements of the edge lightweight diagnostic model, and simultaneously recording the acquisition angle, lighting conditions, and time stamp.

[0037] The data acquisition window and accuracy are jointly determined by the anomaly trigger level and field environmental conditions. For example, in a scenario with suspected leaf spot disease risk, the system prioritizes a 30-minute acquisition window in the early morning when the leaves are still moist and have not yet evaporated, and configures the image resolution to a medium level suitable for extracting candidate lesion areas. In a scenario with suspected water stress, the system sets the acquisition window to the midday high transpiration period and the evening recovery period, and focuses on acquiring soil moisture change curves and canopy drooping. When the node power is below 25%, even if the anomaly trigger level is high, the system will reduce the image acquisition frequency and prioritize acquiring basic sensing data with greater diagnostic gain, and perform supplementary acquisition after the solar energy replenishment status improves.

[0038] After low-power sensing nodes perform data acquisition according to a tiered acquisition strategy, they write the acquired crop status data into their local cache and mark the acquisition strategy number, anomaly trigger level, and energy consumption record. If excessive light, blurry images, or communication failures occur during acquisition, the node marks the corresponding acquisition results as low-quality data and determines whether to re-acquire within the same acquisition window based on the tiered acquisition strategy. In this way, the system can dynamically control the acquisition intensity based on the degree of anomaly risk and energy consumption conditions, ensuring that crop status data matches current diagnostic needs, reducing invalid high-power acquisition, and improving the endurance of low-power sensing nodes and the efficiency of field anomaly detection.

[0039] In some embodiments, lightweight diagnostics are performed on crop state data at a low-power sensing node to generate candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis, including: Node-side quality verification and feature compression are performed on crop state data to obtain crop state features suitable for low-power inference. The lightweight diagnostic model deployed on the low-power sensing node is invoked to identify the anomaly type and calculate the confidence level of crop status characteristics, and generate candidate diagnostic results. Based on the candidate diagnostic results, extract the key features, data collection conditions, and model output information that contribute to the diagnostic judgment to form a diagnostic basis representation; The diagnostic evidence is associated and encapsulated with node identifiers, micro-region identifiers, and time identifiers to generate a summary of diagnostic evidence.

[0040] Specifically, after the low-power sensing node acquires crop status data according to a hierarchical acquisition strategy, the node-side processor first performs quality verification on the crop status data. Taking a suspected leaf spot disease diagnosis scenario in a tomato field as an example, the crop status data includes low-resolution leaf images, canopy color features, leaf surface moisture duration, air temperature and humidity changes, and light intensity at the time of acquisition. The low-power sensing node performs brightness equalization, blur detection, and target area integrity judgment on the leaf images. When an image is overexposed, occluded, or has motion blur, the node marks the corresponding image as low-quality data and prioritizes using backup images within the same acquisition window. When consecutive images do not meet the quality conditions, the node retains environmental perception features and low-quality labels to avoid incorrect diagnosis due to invalid images. For environmental crop status data, the node verifies the data through differences between adjacent sampled values, sensor drift range, and historical stable intervals, and removes or downweights sudden spikes and sampled values ​​that significantly exceed the sensor's operating range.

[0041] After quality verification, the low-power sensing node performs feature compression on the crop state data to obtain crop state features suitable for low-power inference. For image data, the node does not retain the complete image, but extracts color shift features, lesion candidate area proportions, edge texture intensity, and local morphological anomaly features within the target area of ​​the leaf, and quantizes these features into low-bit feature vectors. For environmental data, the node generates temperature and humidity deviation amplitudes, leaf surface wetness durations, soil moisture decline slopes, and light anomaly durations based on multiple recent sampling windows. For canopy state data, the node extracts canopy color index, drooping degree, and change trends. Through the above feature compression, image and time-series data that originally required large storage and transmission resources are converted into lightweight feature sets, reducing the burden on node-side inference and subsequent communication.

[0042] Subsequently, the low-power sensing node invokes a locally deployed lightweight diagnostic model to identify anomaly types. The lightweight diagnostic model can be a multimodal edge model that has undergone pruning, quantization, and distillation, or a node-side diagnostic model formed by combining a lightweight neural network with a rule-based discrimination model. The model input is crop state features, and the output is candidate anomaly types and their corresponding confidence scores. For example, when the proportion of candidate lesion areas increases, the duration of leaf surface moisture exceeds the disease induction threshold, and air humidity remains consistently high, the model outputs candidate diagnostic results for leaf spot diseases and calculates a corresponding confidence score of 0.68; when the degree of leaf drooping increases, the slope of soil moisture decreases significantly, and lesion characteristics are not significant, the model outputs candidate diagnostic results for water stress. If the confidence scores of different anomaly types are close, the node retains multiple candidate diagnostic results together and marks them as pending collaborative verification.

[0043] After generating candidate diagnostic results, the low-power sensing node further extracts key features, acquisition conditions, and model output information that contribute to the diagnostic judgment, forming a diagnostic basis representation. Key features may include the proportion of candidate lesion areas, the magnitude of color anomalies, the duration of leaf surface wetting, the degree of soil moisture deviation, and the canopy change trend; acquisition conditions may include the acquisition window, light conditions, acquisition accuracy, data quality level, and the node's current energy consumption status; model output information may include candidate anomaly type, confidence level, model version, and uncertainty indicator. For the aforementioned candidate results for leaf spot diseases, the node records the key features that contribute significantly to the diagnosis as the duration of leaf surface wetting and the proportion of candidate lesion areas, and records the image quality level as acceptable, the acquisition time as early morning, and the model version as the current version of the edge lightweight model.

[0044] Finally, the low-power sensing node associates and encapsulates the diagnostic evidence representation with node identifiers, micro-region identifiers, and time identifiers to generate a diagnostic evidence summary. This summary does not contain the complete original image or sensor sequence; instead, it includes compressed features, quality identifiers, confidence information, and source identifiers to support candidate diagnostic results. Upon receiving the diagnostic evidence summary, the field edge gateway can locate the corresponding field micro-region based on the micro-region identifier, determine the data source credibility based on the node identifier, perform time-series alignment based on the time identifier, and conduct collaborative diagnosis by combining the field micro-region health status map. In this way, the low-power sensing node can complete lightweight initial screening and evidence summary generation locally, reducing the need for original data transmission and providing traceable and fusionable diagnostic evidence for subsequent edge collaborative inference.

[0045] In some embodiments, a field edge gateway determines the collaborative diagnostic range of a target field micro-region based on a field micro-region health status map, and fuses diagnostic evidence summaries from the target field micro-region and the collaborative diagnostic range to generate an edge collaborative diagnostic confidence score, including: Starting from the map node of the target field micro-region in the field micro-region health status map, the associated micro-regions that meet the synergy conditions are determined according to the micro-region association edges and association weights, thus forming the synergistic diagnosis range; Obtain diagnostic evidence summaries from the target field micro-regions and the scope of collaborative diagnosis, and screen the diagnostic evidence summaries for validity based on node credibility and evidence timeliness; Based on the field micro-region health status map, a correlation propagation consistency analysis was performed on the screened diagnostic evidence summary to generate a collaborative evidence characterization. By fusing collaborative evidence representation with candidate diagnostic results of the target field micro-region, a marginal collaborative diagnostic confidence score is generated.

[0046] Specifically, after the low-power sensing node uploads the diagnostic evidence summary, the field edge gateway first locates the corresponding map node of the target field micro-region in the field micro-region health status map based on the micro-region identifier in the diagnostic evidence summary. Taking the candidate diagnostic result of leaf spot disease in the tomato micro-region in the 3rd row and 5th column as an example, the field edge gateway uses this map node as the starting point, reads the micro-region association edges connected to this map node, and determines which associated micro-regions can participate in collaborative diagnosis based on the association weights. Collaboration conditions may include the association weight reaching a preset threshold, matching the abnormal propagation direction, the sampling time being within a valid time window, and the low-power sensing node being in a trusted state.

[0047] In some examples, if the micro-area in row 3, column 4 and the micro-area in row 3, column 6 are spatially adjacent to the target field micro-area and belong to the same irrigation branch, and the micro-area in row 2, column 5 is located upwind and has a history of similar disease transmission, then the above micro-areas are included in the scope of collaborative diagnosis. If a micro-area is close in distance but has significant differences in crop variety, growth stage, or management operation, then even if there is a spatial proximity relationship, it may not be included in the scope of collaborative diagnosis to avoid irrelevant evidence interfering with the fusion reasoning.

[0048] Furthermore, after establishing the collaborative diagnostic scope, the field edge gateway acquires diagnostic evidence summaries for the target field micro-region and the collaborative diagnostic scope. These summaries can be uploaded in real-time by low-power sensing nodes or retrieved from the edge cache within the most recent valid time window. The field edge gateway then filters the diagnostic evidence summaries based on factors including node credibility, evidence timeliness, data quality level, and the completeness of candidate diagnostic results.

[0049] In some examples, node credibility can be determined by the node's historical online rate, sampling stability, consistency rate of manual review, and current energy consumption status; the timeliness of evidence can be determined by the interval between the generation time of the diagnostic evidence summary and the current diagnosis time. For example, in the scenario of diagnosing leaf spot diseases, the system only retains diagnostic evidence summaries generated within the last 2 hours that meet the data quality level requirements; for nodes with low power levels leading to incomplete sampling, the system reduces the fusion weight of the corresponding diagnostic evidence summary, or only retains environmental features for auxiliary judgment.

[0050] Furthermore, after completing the validity screening, the field edge gateway performs a correlation propagation consistency analysis on the screened diagnostic evidence summaries based on the field micro-region health status map. Specifically, it performs map constraint matching between the candidate anomaly types, key features, and confidence levels of the target field micro-region and the diagnostic evidence summaries of each associated micro-region within the collaborative diagnosis range to determine whether the anomaly features show consistent changes along spatial proximity, irrigation, or wind direction propagation relationships.

[0051] For example, taking leaf spot diseases as an example, if the proportion of candidate lesion areas in the target field micro-area increases, the duration of leaf surface moisture is prolonged, and similar lesion characteristics have appeared in the upwind micro-area in an earlier time window, while the downwind micro-area shows a slight color shift, then the system determines that the consistency of the associated transmission is high. If the target field micro-area has yellowing leaves, but only the micro-area of ​​the same irrigation branch within the collaborative diagnosis range shows abnormal soil moisture, and the lesion characteristics are discontinuous, then the system points the consistency of transmission to water stress or management abnormalities, rather than disease spread.

[0052] The field edge gateway further transforms the consistency analysis results of the correlation propagation into a collaborative evidence representation. This collaborative evidence representation includes the evidence strength of the target field micro-region, the support level of the associated micro-regions, the matching degree of the propagation path, the strength of conflicting evidence, and the correction amount for evidence credibility. Subsequently, the system inputs the collaborative evidence representation and the candidate diagnostic results of the target field micro-region into the edge fusion inference model. The edge fusion inference model can be implemented using a lightweight graph attention model, a Bayesian evidence fusion model, or a rule-constrained multimodal fusion model.

[0053] During the fusion inference process, the node-side confidence of the target field micro-region provides the initial diagnostic basis, the evidence summary within the collaborative diagnosis range provides spatial and propagation relationship verification, and conflicting evidence is used to reduce the confidence of unstable diagnostic results. If the node-side confidence of leaf spot disease in the target field micro-region is 0.68, the neighborhood propagation consistency is high and the upwind evidence is clearly supportive, the confidence of the edge collaborative diagnosis after fusion can be increased to 0.83; if the neighborhood evidence conflicts with the candidate results of the target micro-region, the confidence after fusion is reduced, and the subsequent minimum supplementary sampling process is triggered. In this way, the field edge gateway can use the health status map to perform structured collaborative verification of the diagnostic evidence of multiple low-power nodes, improving the ability to distinguish similar abnormal phenotypes and the reliability of edge diagnosis in weak network environments.

[0054] In some embodiments, when the confidence level of edge collaborative diagnosis does not meet the confirmation criteria, a minimum supplementary sampling instruction is generated based on the diagnostic uncertainty and the additional energy consumption cost, and the confidence level of edge collaborative diagnosis is updated according to the supplementary sampling result, including: Based on the difference between the confidence level of edge co-diagnosis and the confirmation criteria, the sources of diagnostic uncertainty in the target field micro-regions were determined; By combining the sources of diagnostic uncertainty, the available energy consumption constraints of low-power sensing nodes, and the evidence gaps within the scope of collaborative diagnosis, the targets and constraints for supplementary sampling are determined. The minimum replenishment command is generated based on the replenishment target and replenishment constraints, and then sent to the corresponding low-power sensing node. Receive the supplementary sampling results generated by the low-power sensing node according to the minimum supplementary sampling instruction, and convert the supplementary sampling results into a supplementary diagnostic evidence summary; Based on the supplementary diagnostic evidence summary, fusion inference is re-executed to update the confidence of edge co-diagnostic findings.

[0055] Specifically, after the field edge gateway completes the initial fusion inference, the system compares the edge co-diagnosis confidence level with the confirmation criteria. Confirmation criteria can be pre-configured based on crop type, anomaly type, and risk level. For example, the confirmation threshold for tomato leaf spot diseases is 0.80, and the confirmation threshold for water stress anomalies is 0.75. When the edge co-diagnosis confidence level of the target field micro-plot is 0.68, which does not reach the leaf spot disease confirmation threshold, the field edge gateway does not directly output the confirmation diagnosis result. Instead, it analyzes the reasons for the insufficient confidence level and determines the source of diagnostic uncertainty.

[0056] Sources of diagnostic uncertainty may include insufficient quality of image evidence for the target micro-area, inconsistent direction of support from neighboring evidence, lack of environmental triggering conditions, similar confidence levels for different candidate anomaly types, or lack of effective evidence for key propagation paths. For example, if a lesion candidate area exists in the micro-area in row 3, column 5, but the image of the upwind micro-area is blurry, and soil moisture anomalies are also present in the same irrigation branch micro-area, the system will determine the source of uncertainty as a combination of insufficient evidence of disease spread and interference from water stress.

[0057] After identifying the sources of diagnostic uncertainty, the field edge gateway, combining the available energy consumption constraints of the low-power sensing nodes and the evidence gaps within the collaborative diagnostic scope, determines the supplementary sampling targets and constraints. Supplementary sampling targets define the field micro-regions, sampling objects, and diagnostic features requiring further verification. Supplementary sampling constraints define the supplementary sampling window, sampling accuracy, allowable energy consumption, and communication methods. For example, if the confidence level of candidate lesion regions in the leaf images of the target micro-region is insufficient, but the duration of leaf wetting meets the disease induction conditions, the system determines the supplementary sampling targets as local leaf images of the target micro-region and early lesion features in the upwind micro-region. If the main evidence gap stems from water stress and the inability to distinguish between disease phenotypes, the system determines the supplementary sampling targets as soil moisture changes and canopy sagging characteristics in the micro-region along the same irrigation branch. When the remaining power of the target low-power sensing node is above 50%, the supplementary sampling constraints allow for medium-precision image supplementary sampling; when the remaining power is below 25%, the supplementary sampling constraints prioritize environmental feature supplementary sampling or delay execution until a period with better solar energy replenishment.

[0058] Furthermore, after determining the supplementary data collection target and constraints, the field edge gateway generates a minimum supplementary data collection instruction. This instruction includes the target micro-area identifier, target node identifier, supplementary data collection mode, supplementary data collection window, collection accuracy, maximum number of re-collections, maximum energy consumption, and result return format. Taking an uncertain scenario involving leaf spot diseases as an example, the minimum supplementary data collection instruction can require the target micro-area node to collect two medium-resolution partial leaf images within 30 minutes of dawn, and require the upwind micro-area node to extract only the proportion of candidate lesion areas and the magnitude of color anomalies, without uploading the complete image. For water stress disturbance scenarios, the minimum supplementary data collection instruction can require nodes along the same irrigation branch to collect soil moisture changes and canopy morphology characteristics once at midday and once in the evening. In this way, the system only supplements data with the greatest diagnostic gain for the current uncertainty, avoiding the full wake-up of all low-power sensing nodes.

[0059] After receiving the minimum data acquisition command, the low-power sensing node first verifies its remaining battery power, sensor availability, and current acquisition conditions. If the conditions are met, it performs data acquisition according to the acquisition window and accuracy. If the conditions are not met, the node returns the reason for failure and alternative acquisition capabilities. After data acquisition, the node performs quality verification and feature compression on the acquired data to generate acquisition results. The acquisition results may include acquisition features, quality level, energy consumption, acquisition time, and execution status. After receiving the acquisition results, the field edge gateway converts them into a supplementary diagnostic evidence summary and performs time alignment and source association with the original diagnostic evidence summary.

[0060] Subsequently, the field edge gateway re-executes fusion inference based on the supplemented diagnostic evidence summary. If the supplemented leaf images of the target micro-region show an increased proportion of lesion candidate areas, similar color anomalies in the upwind micro-region, and improved propagation path matching, the confidence level of edge co-diagnosis increases from 0.68 to 0.84, meeting the confirmation criteria. If the supplemented soil moisture changes and canopy drooping characteristics are more consistent with water stress, and the lesion characteristics do not expand in the neighborhood, the system reduces the confidence level of leaf spot diseases and increases the confidence level of water stress diagnosis. Thus, the minimum supplementary sampling mechanism can actively verify evidence gaps under limited energy consumption conditions, improving the reliability of edge co-diagnosis results while reducing unnecessary data collection and communication consumption.

[0061] In some embodiments, a field diagnostic result carrying a chain of evidence is output based on the updated edge-co-diagnostic confidence level, and the field micro-plot health status map and diagnostic control parameters are updated based on diagnostic feedback, including: When the updated confidence level of the edge co-diagnosis meets the confirmation criteria, the abnormal diagnosis category and risk level of the target field micro-region are determined based on the fusion inference results. The diagnostic evidence chain is generated by associating the diagnostic evidence summary, supplementary sampling results, collaborative diagnostic scope, and model version involved in the fusion reasoning; Field diagnostic results are generated based on the abnormal diagnostic category, risk level, and diagnostic evidence chain. Obtain manual review feedback or treatment feedback corresponding to the field diagnosis results, update the health status and correlation weight in the field micro-plot health status map based on the consistency of feedback, and simultaneously correct the abnormal awakening factor, hierarchical collection strategy and diagnostic control parameters corresponding to the fusion reasoning process.

[0062] Specifically, after the field edge gateway re-executes fusion inference based on the supplementary diagnostic evidence summary, the system matches the updated edge co-diagnostic confidence with the confirmation criteria. Taking the field micro-plot in the 3rd row and 5th column of a tomato planting field as an example, if the confirmation threshold for leaf spot disease is 0.80, and the edge co-diagnostic confidence obtained after fusion inference is 0.84, the system determines that the confirmation criteria are met, and determines the abnormal diagnosis category as early risk of leaf spot disease based on the fusion inference result, with a risk level of medium to high risk.

[0063] The risk level can be determined comprehensively based on the confidence level of edge co-diagnosis, the abnormal spread trend, the sensitivity of crop growth period, and the affected area of ​​associated micro-regions. For example, when the proportion of candidate lesion areas in the target field micro-region continues to increase, and there is similar abnormal evidence in the upwind micro-region at an earlier time window, the system increases the risk level of disease spread; when there are only localized minor abnormalities in the target field micro-region, and there is no evidence of continuous spread within the scope of co-diagnosis, the system decreases the risk level or generates a diagnostic result to be observed.

[0064] Furthermore, after determining the abnormal diagnosis category and risk level, the field edge gateway generates a diagnostic evidence chain. This chain records key evidence in the diagnostic conclusion formation process and may include: a summary of diagnostic evidence uploaded from the target field micro-region; summaries of diagnostic evidence uploaded from associated micro-regions within the collaborative diagnosis range; supplementary sampling results corresponding to the minimum supplementary sampling instruction; micro-region association edges and association weights involved in fusion inference; the edge fusion inference model version; and the lightweight diagnostic model version.

[0065] For example, taking the early risk of leaf spot diseases as an example, the diagnostic evidence chain records the proportion of candidate lesion areas uploaded by the target micro-region node, the duration of leaf surface wetting, the image quality level, and the node-side confidence level. It also records the color anomaly amplitude, collection time, and propagation path matching degree of the upwind micro-region, as well as the quality verification results of the re-collected images and the increase in confidence level after re-collection. The system can configure a unique evidence chain identifier for the diagnostic evidence chain, enabling subsequent manual review, handling feedback, and model updates to trace back to the corresponding evidence source.

[0066] Subsequently, the system generates field diagnostic results based on the anomaly diagnosis category, risk level, and diagnostic evidence chain. These results may include the target field micro-region identifier, anomaly diagnosis category, risk level, edge collaborative diagnosis confidence level, suggested review location, suggested treatment method, and evidence chain identifier. For example, the system outputs to the farmer's terminal that the tomato micro-region in row 3, column 5 has an early risk of leaf spot disease, suggesting manual review within 24 hours, with a focus on adjacent micro-regions upwind. It also indicates that the evidence sources include leaf image summaries of the target micro-region, lesion summaries of the upwind micro-region, and morning supplementary sampling results. For high-risk results, the field edge gateway can also mark the target micro-region and related micro-regions as key monitoring areas and temporarily increase the anomaly monitoring priority of the corresponding low-power sensing nodes.

[0067] Furthermore, after the field diagnosis results are output, the system obtains corresponding manual review feedback or treatment feedback. Manual review feedback may include whether the review confirms an anomaly, the actual anomaly type, the anomaly area, and a summary of on-site photos; treatment feedback may include actions such as spraying, supplemental irrigation, fertilization, and pruning, as well as any abnormal changes after treatment. The system compares the feedback results with the field diagnosis results for consistency. If the manual review confirms leaf spot disease, consistent with the field diagnosis results, the system updates the health status of the target micro-region to "confirmed abnormal" and increases the association weight of the corresponding anomaly type between the target micro-region and the upwind associated micro-region. If the review results show that the actual cause is water stress, the system reduces the weight of evidence related to leaf spot disease and enhances the diagnostic contribution of water stress characteristics in similar scenarios.

[0068] Furthermore, the field edge gateway synchronously adjusts diagnostic control parameters based on feedback consistency. When a diagnostic result is confirmed and the anomaly propagation rate is higher than expected, the system lowers the anomaly wake-up factor trigger threshold for the corresponding anomaly type and increases the propagation weight of associated micro-regions in the fusion inference process. When a diagnostic result is rejected and the misjudgment is due to excessive lighting or insufficient image quality, the system raises the image quality threshold, adjusts the acquisition window in the hierarchical acquisition strategy, and reduces the contribution of low-quality image summaries to fusion inference. When a node repeatedly shows inconsistencies in verification, the system lowers the credibility of the corresponding node.

[0069] Through the above feedback updates, the field micro-area health status map can continuously reflect real field risk changes, and the abnormal awakening, hierarchical collection and fusion reasoning processes can also be adaptively corrected with diagnostic feedback, thereby improving the accuracy and traceability of subsequent field collaborative diagnosis.

[0070] In some embodiments, after updating the field microplot health status map and diagnostic control parameters based on diagnostic feedback, the method further includes: Based on the discrepancy between diagnostic feedback and field diagnostic results, the adaptation requirements of the lightweight diagnostic model on the low-power sensing node side are determined. Extract summary samples that match the adaptation requirements from the diagnostic evidence chain, and generate incremental adaptation data for model updates based on the summary samples; The field edge gateway determines the timing of incremental adaptation data delivery based on the node energy consumption and communication status of the low-power sensing nodes. When the distribution conditions are met, incremental adaptation data is written to the corresponding low-power sensing node, enabling the lightweight diagnostic model to complete node-side adaptive updates based on diagnostic feedback from field micro-regions.

[0071] Specifically, after the field diagnosis results are manually reviewed or feedback is returned, the field edge gateway first performs a deviation analysis between the diagnostic feedback and the field diagnosis results. Taking a tomato field as an example, if the original output of the field micro-region in the 3rd row and 5th column indicates an early risk of leaf spot disease, and the edge co-diagnosis confidence level is 0.84, but manual review confirms that it is actually water stress, the system marks the deviation relationship as an abnormal category deviation. If the review result confirms leaf spot disease but the actual lesion area is significantly larger than the predicted range, it is marked as a risk level deviation. If the disease in the target micro-region does not continue to expand after treatment, but the system originally predicted a high risk of transmission, it is marked as a transmission trend deviation. The system traces the corresponding diagnostic evidence chain based on the deviation relationship, identifying that the deviation mainly comes from the node-side lightweight diagnostic model, the edge fusion inference process, or the association weights in the field micro-region health status map. For deviations confirmed to be caused by the node-side lightweight diagnostic model, the system generates model adaptation requirements.

[0072] Model adaptation requirements can be further subdivided according to the type of deviation. When the deviation in anomaly category stems from similar image phenotypes, the adaptation requirement points to feature discrimination boundary correction; when the deviation in risk level stems from low local confidence, the adaptation requirement points to confidence calibration; when the same anomaly manifests significantly under different growth stages or light conditions, the adaptation requirement points to micro-area scene adaptation. For example, during the fruit enlargement stage of tomatoes, strong midday light causes localized yellowing of leaves, which is similar in color to early lesions. The lightweight diagnostic model on the low-power sensing node side is prone to misjudging water stress as disease. Based on this, the system determines that it needs to increase the ability to distinguish between yellowing features and lesion features under strong light conditions, and limits the adaptation target to low-power sensing nodes covering this type of micro-area.

[0073] Furthermore, after determining the adaptation requirements, the field edge gateway extracts summary samples that match the adaptation requirements from the diagnostic evidence chain. These summary samples do not use complete original images or complete time-series data; instead, they extract compressed features, acquisition conditions, model output information, and feedback labels that can express diagnostic biases. For example, in cases where water stress is misdiagnosed as leaf spot diseases, the system extracts the proportion of candidate lesion areas, the magnitude of color anomalies, the duration of leaf wetting, the slope of soil moisture decline, image quality level, acquisition lighting conditions, the confidence level of the original candidate diagnoses, and manual verification labels. For risk level biases, the system extracts the matching degree of propagation paths between the target micro-region and related micro-regions, the range of anomaly expansion, changes in confidence after re-collection, and the final treatment results. The system performs deduplication, quality screening, and label consistency verification on the summary samples, eliminating samples with abnormal acquisition conditions, incomplete feedback, or low credibility of evidence sources, retaining the sample set that reflects the bias patterns of local fields.

[0074] After the summary samples are generated, the field edge gateway uses them to generate incremental adaptation data for model updates. This incremental adaptation data can include feature calibration parameters, class discrimination threshold corrections, confidence temperature calibration parameters, lightweight model local layer parameter differences, and sample index summaries. To meet the storage and computational limitations of low-power nodes, the system can generate small-sized incremental packages using edge-side mini-batch adaptation, parameter distillation, or prototype feature updates. For example, to address the misjudgment of yellowing due to strong light, the system generates scene correction parameters for color anomaly features and increases the contribution of soil moisture decline slope in water stress discrimination; to address the issue of low confidence in early lesion identification, the system updates the discrimination threshold of texture features in lesion candidate regions, enabling the model to identify weak lesion features earlier.

[0075] Furthermore, after the incremental adaptation data is generated, the field edge gateway does not immediately distribute it to the low-power sensing nodes. Instead, it determines the distribution timing based on the node's energy consumption and communication status. The node's energy consumption status includes remaining power, solar power replenishment trends, and current data collection task load. The communication status includes link quality, packet loss rate, and edge gateway connection stability. If the node's remaining power is above 50%, solar power replenishment is stable, and it is not currently in a high-risk data collection window, the system determines the distribution timing to be during the current idle period. If the node's power is below 30% or the communication link is unstable, the system caches the incremental adaptation data on the field edge gateway and waits for low-business hours at night or after power replenishment is restored before distributing it. For scenarios where multiple nodes need to be updated, the system determines the batch distribution order according to the deviation impact range, node credibility, and risk micro-region priority to avoid communication congestion and sudden increases in node energy consumption caused by simultaneous updates.

[0076] When the distribution conditions are met, the field edge gateway writes the incremental adaptation data to the corresponding low-power sensing node. Upon receiving the data, the low-power sensing node performs version verification, integrity verification, and compatibility verification. If the verification passes, it writes the incremental adaptation data to the adaptation parameter area of ​​the lightweight diagnostic model and retains the previous version rollback record. After the model update is complete, the node uses a small number of locally cached summary samples for rapid verification. If the consistency between the updated candidate diagnostic results and the feedback labels improves, the new adaptation parameters are enabled; if the verification fails, it rolls back to the previous model version and reports the failure reason to the field edge gateway.

[0077] Through the above implementation methods, the lightweight diagnostic model can complete node-side adaptive updates based on real diagnostic feedback from field micro-regions, which reduces the reliance on centralized cloud training and large-scale data transmission, and improves the local diagnostic adaptability under different field conditions, different growth stages, and different microclimates.

[0078] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0079] Figure 2 This is a schematic diagram of the structure of the field edge collaborative low-power diagnostic system provided in an embodiment of this application. Figure 2 As shown, the system includes: Module 201 is used to divide the target field into multiple field micro-regions and construct a field micro-region health status map based on the spatial association, agricultural association and historical anomaly association of each field micro-region. The generation module 202 is used to acquire basic sensing data through low-power sensing nodes deployed in field micro-regions, and generate abnormal wake-up factors based on the basic sensing data, node energy consumption status and the associated risks in the field micro-region health status map. The acquisition module 203 is used to determine the hierarchical acquisition strategy of the corresponding low-power sensing node based on the abnormal wake-up factor and energy consumption constraints, and to acquire crop status data according to the hierarchical acquisition strategy. The diagnostic module 204 is used to perform lightweight diagnostics on crop status data at the low-power sensing node side, and generate candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis. The fusion module 205 is used to determine the collaborative diagnosis range of the target field micro-region based on the field micro-region health status map through the field edge gateway, and to fuse the diagnostic evidence summary of the target field micro-region and the collaborative diagnosis range to generate edge collaborative diagnosis confidence. The update module 206 is used to generate a minimum supplementary sampling instruction based on diagnostic uncertainty and additional energy consumption cost when the confidence level of edge collaborative diagnosis does not meet the confirmation conditions, and update the confidence level of edge collaborative diagnosis according to the supplementary sampling result. The output module 207 is used to output field diagnostic results carrying the chain of evidence based on the updated edge co-diagnostic confidence, and to update the field micro-plot health status map and diagnostic control parameters based on the diagnostic feedback.

[0080] In some embodiments, Figure 2 The construction module 201 acquires the basic management data and historical diagnostic data of the target field, and determines the micro-region division scale according to the field diagnostic accuracy and the coverage capability of low-power sensing nodes. Based on the micro-region division scale, the target field is divided into multiple field micro-regions, and a micro-region identifier and a low-power sensing node association identifier are configured for each field micro-region. Micro-region association edges are generated according to the spatial proximity relationship, management operation consistency and anomaly propagation similarity between field micro-regions, and association weights are configured for the micro-region association edges. Based on the crop status baseline, environmental status baseline, node credibility and micro-region association edges of each field micro-region, a field micro-region health status map is constructed.

[0081] In some embodiments, Figure 2 The generation module 202 enables low-power sensing nodes to acquire basic sensing data of corresponding field micro-regions according to a low duty cycle sampling strategy, and performs time-series stability evaluation on the basic sensing data to obtain micro-region state deviation characterization; reads the node energy consumption status of low-power sensing nodes, and determines the available energy consumption constraint for the abnormal sensing process based on the node energy consumption status; acquires risk transmission information associated with the corresponding field micro-region based on the field micro-region health status map, and generates associated risk characterization; and integrates micro-region state deviation characterization, available energy consumption constraint and associated risk characterization to generate an abnormal wake-up factor for triggering the hierarchical acquisition strategy.

[0082] In some embodiments, Figure 2 The acquisition module 203 matches the abnormal wake-up factor with the preset wake-up level rules to determine the abnormal trigger level of the corresponding field micro-region; it generates a hierarchical acquisition strategy based on the abnormal trigger level, the available energy consumption constraints of the low-power sensing node, and the diagnostic requirements of crop status data; it controls the low-power sensing node to switch between different acquisition intensities according to the hierarchical acquisition strategy and determines the corresponding acquisition window and acquisition accuracy; and it acquires crop status data that matches the current abnormal trigger level based on the acquisition window and acquisition accuracy.

[0083] In some embodiments, Figure 2The diagnostic module 204 performs node-side quality verification and feature compression on crop status data to obtain crop status features suitable for low-power inference; it calls a lightweight diagnostic model deployed on low-power sensing nodes to perform anomaly type discrimination and confidence calculation on the crop status features, generating candidate diagnostic results; it extracts key features, acquisition conditions, and model output information that contribute to the diagnostic discrimination based on the candidate diagnostic results, forming a diagnostic basis representation; it associates and encapsulates the diagnostic basis representation with node identifiers, micro-region identifiers, and time identifiers to generate a diagnostic evidence summary.

[0084] In some embodiments, Figure 2 The fusion module 205 starts with the map nodes of the target field micro-region in the field micro-region health status map, determines the associated micro-regions that meet the synergy conditions based on the micro-region association edges and association weights, and forms a synergy diagnosis range; obtains the diagnostic evidence summary of the target field micro-region and the synergy diagnosis range, and filters the diagnostic evidence summary for validity based on node credibility and evidence timeliness; performs association propagation consistency analysis on the filtered diagnostic evidence summary based on the field micro-region health status map to generate a synergy evidence representation; and performs fusion reasoning with the candidate diagnostic results of the target field micro-region to generate a marginal synergy diagnosis confidence level.

[0085] In some embodiments, Figure 2 The update module 206 determines the source of diagnostic uncertainty in the target field micro-region based on the difference between the edge collaborative diagnosis confidence and the confirmation conditions; it determines the supplementary sampling target and supplementary sampling constraints by combining the source of diagnostic uncertainty, the available energy consumption constraints of the low-power sensing nodes, and the evidence gap within the collaborative diagnosis range; it generates a minimum supplementary sampling instruction based on the supplementary sampling target and supplementary sampling constraints, and sends the minimum supplementary sampling instruction to the corresponding low-power sensing nodes; it receives the supplementary sampling results generated by the low-power sensing nodes based on the minimum supplementary sampling instructions, and converts the supplementary sampling results into a supplementary diagnostic evidence summary; it re-executes fusion inference based on the supplementary diagnostic evidence summary to update the edge collaborative diagnosis confidence.

[0086] In some embodiments, Figure 2 When the updated edge collaborative diagnosis confidence meets the confirmation conditions, the output module 207 determines the abnormal diagnosis category and risk level of the target field micro-region based on the fusion inference results; it generates a diagnostic evidence chain by associating the diagnostic evidence summary, supplementary sampling results, collaborative diagnosis range, and model version involved in the fusion inference; it generates field diagnosis results based on the abnormal diagnosis category, risk level, and diagnostic evidence chain; it obtains manual review feedback or treatment feedback corresponding to the field diagnosis results, updates the health status and association weights in the field micro-region health status map based on the consistency of feedback, and simultaneously corrects the abnormal awakening factor, hierarchical collection strategy, and diagnostic control parameters corresponding to the fusion inference process.

[0087] In some embodiments, Figure 2 The output module 207 determines the adaptation requirements of the lightweight diagnostic model on the low-power sensing node side based on the deviation relationship between the diagnostic feedback and the field diagnostic results; extracts summary samples that match the adaptation requirements from the diagnostic evidence chain, and generates incremental adaptation data for model updates based on the summary samples; determines the timing of incremental adaptation data delivery based on the node energy consumption status and communication status of the low-power sensing node through the field edge gateway; and writes the incremental adaptation data to the corresponding low-power sensing node when the delivery conditions are met, so that the lightweight diagnostic model can complete the node-side adaptive update based on the diagnostic feedback of the field micro-region.

[0088] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0089] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0090] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0091] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0092] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0095] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A low-power diagnostic method for field edge collaboration, characterized in that, include: The target field is divided into multiple field micro-regions, and a field micro-region health status map is constructed based on the spatial correlation, agricultural correlation and historical abnormal correlation of each field micro-region. Basic sensing data is acquired by low-power sensing nodes deployed in field micro-plots. Abnormal wake-up factors are generated based on the basic sensing data, node energy consumption status, and associated risks in the field micro-plot health status map. The hierarchical acquisition strategy for the corresponding low-power sensing node is determined based on the abnormal wake-up factor and energy consumption constraints, and crop status data is collected according to the hierarchical acquisition strategy. Lightweight diagnostics of crop status data are performed on the low-power sensing node side to generate candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis. The collaborative diagnosis range of the target field micro-region is determined by the field edge gateway based on the field micro-region health status map, and the diagnostic evidence summary of the target field micro-region and the collaborative diagnosis range is integrated to generate the edge collaborative diagnosis confidence. When the confidence level of edge collaborative diagnosis does not meet the confirmation conditions, a minimum supplementary sampling instruction is generated based on the diagnostic uncertainty and the additional energy consumption cost, and the confidence level of edge collaborative diagnosis is updated according to the supplementary sampling result. Based on the updated edge co-diagnostic confidence level, the field diagnostic results carrying the chain of evidence are output, and the field micro-plot health status map and diagnostic control parameters are updated based on the diagnostic feedback.

2. The method according to claim 1, characterized in that, The process of dividing the target field into multiple micro-regions and constructing a health status map of each micro-region based on spatial correlation, agricultural correlation, and historical anomaly correlation includes: Acquire basic management data and historical diagnostic data of the target field, and determine the micro-area division scale according to the field diagnostic accuracy and the coverage capability of low-power sensing nodes; The target field is divided into multiple field micro-regions based on the micro-region division scale, and each field micro-region is configured with a micro-region identifier and a low-power sensing node association identifier. Micro-region association edges are generated based on the spatial proximity between field micro-regions, the consistency of management operations, and the similarity of anomaly propagation, and association weights are assigned to the micro-region association edges; Based on the crop status baseline, environmental status baseline, node credibility, and micro-region association edges of each field micro-region, a field micro-region health status map is constructed.

3. The method according to claim 1, characterized in that, The process involves acquiring basic sensing data through low-power sensing nodes deployed in field micro-plots, and generating abnormal wake-up factors based on the basic sensing data, node energy consumption status, and associated risks in the field micro-plot health status map. This includes: The low-power sensing nodes acquire basic sensing data of the corresponding field micro-regions according to the low duty cycle sampling strategy, and evaluate the temporal stability of the basic sensing data to obtain the micro-region state deviation characterization. Read the node energy consumption status of the low-power sensing node and determine the available energy consumption constraints for the anomaly sensing process based on the node energy consumption status. Based on the field micro-plot health status map, risk transmission information associated with the corresponding field micro-plot is obtained, and associated risk characterization is generated; By integrating micro-area state deviation representation, available energy consumption constraint, and associated risk representation, an abnormal wake-up factor is generated to trigger the hierarchical acquisition strategy.

4. The method according to claim 1, characterized in that, The step of determining the hierarchical acquisition strategy for the corresponding low-power sensing node based on the abnormal wake-up factor and energy consumption constraints, and acquiring crop status data according to the hierarchical acquisition strategy, includes: The abnormal awakening factor is matched with the preset awakening level rule to determine the abnormal triggering level of the corresponding field micro-region; A tiered data acquisition strategy is generated based on the anomaly trigger level, the available energy consumption constraints of low-power sensing nodes, and the diagnostic requirements of crop status data. The low-power sensing nodes are controlled to switch between different acquisition intensities according to the hierarchical acquisition strategy, and the corresponding acquisition time window and acquisition accuracy are determined. Crop status data matching the current anomaly trigger level is obtained based on the acquisition time window and acquisition accuracy.

5. The method according to claim 1, characterized in that, The lightweight diagnosis of crop state data at the low-power sensing node side, generating candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis, includes: Node-side quality verification and feature compression are performed on crop state data to obtain crop state features suitable for low-power inference. The lightweight diagnostic model deployed on the low-power sensing node is invoked to identify the anomaly type and calculate the confidence level of crop status characteristics, and generate candidate diagnostic results. Based on the candidate diagnostic results, extract the key features, data collection conditions, and model output information that contribute to the diagnostic judgment to form a diagnostic basis representation; The diagnostic evidence is associated and encapsulated with node identifiers, micro-region identifiers, and time identifiers to generate a summary of diagnostic evidence.

6. The method according to claim 1, characterized in that, The process involves determining the collaborative diagnostic range of a target field micro-region based on a field micro-region health status map using a field edge gateway, and fusing diagnostic evidence summaries from the target field micro-region and the collaborative diagnostic range to generate edge collaborative diagnostic confidence, including: Starting from the map node of the target field micro-region in the field micro-region health status map, the associated micro-regions that meet the synergy conditions are determined according to the micro-region association edges and association weights, thus forming the synergistic diagnosis range; Obtain diagnostic evidence summaries from the target field micro-regions and the scope of collaborative diagnosis, and screen the diagnostic evidence summaries for validity based on node credibility and evidence timeliness; Based on the field micro-region health status map, a correlation propagation consistency analysis was performed on the screened diagnostic evidence summary to generate a collaborative evidence characterization. By fusing collaborative evidence representation with candidate diagnostic results of the target field micro-region, a marginal collaborative diagnostic confidence score is generated.

7. The method according to claim 1, characterized in that, When the confidence level of edge collaborative diagnosis does not meet the confirmation conditions, a minimum supplementary sampling instruction is generated based on the diagnostic uncertainty and the additional energy consumption cost, and the confidence level of edge collaborative diagnosis is updated according to the supplementary sampling result, including: Based on the difference between the confidence level of edge co-diagnosis and the confirmation criteria, the sources of diagnostic uncertainty in the target field micro-regions were determined; By combining the sources of diagnostic uncertainty, the available energy consumption constraints of low-power sensing nodes, and the evidence gaps within the scope of collaborative diagnosis, the targets and constraints for supplementary sampling are determined. The minimum replenishment command is generated based on the replenishment target and replenishment constraints, and then sent to the corresponding low-power sensing node. Receive the supplementary sampling results generated by the low-power sensing node according to the minimum supplementary sampling instruction, and convert the supplementary sampling results into a supplementary diagnostic evidence summary; Based on the supplementary diagnostic evidence summary, fusion inference is re-executed to update the confidence of edge co-diagnostic findings.

8. The method according to claim 1, characterized in that, The process of outputting field diagnostic results carrying a chain of evidence based on the updated edge-cooperative diagnostic confidence level, and updating the field micro-plot health status map and diagnostic control parameters based on diagnostic feedback, includes: When the updated confidence level of the edge co-diagnosis meets the confirmation criteria, the abnormal diagnosis category and risk level of the target field micro-region are determined based on the fusion inference results. The diagnostic evidence chain is generated by associating the diagnostic evidence summary, supplementary sampling results, collaborative diagnostic scope, and model version involved in the fusion reasoning; Field diagnostic results are generated based on the abnormal diagnostic category, risk level, and diagnostic evidence chain. Obtain manual review feedback or treatment feedback corresponding to the field diagnosis results, update the health status and correlation weight in the field micro-plot health status map based on the consistency of feedback, and simultaneously correct the abnormal awakening factor, hierarchical collection strategy and diagnostic control parameters corresponding to the fusion reasoning process.

9. The method according to claim 8, characterized in that, After updating the field micro-plot health status map and diagnostic control parameters based on diagnostic feedback, the following is also included: Based on the discrepancy between diagnostic feedback and field diagnostic results, the adaptation requirements of the lightweight diagnostic model on the low-power sensing node side are determined. Extract summary samples that match the adaptation requirements from the diagnostic evidence chain, and generate incremental adaptation data for model updates based on the summary samples; The field edge gateway determines the timing of incremental adaptation data delivery based on the node energy consumption and communication status of the low-power sensing nodes. When the distribution conditions are met, incremental adaptation data is written to the corresponding low-power sensing node, enabling the lightweight diagnostic model to complete node-side adaptive updates based on diagnostic feedback from field micro-regions.

10. A low-power diagnostic system for field edge collaboration, characterized in that, include: The module is used to divide the target field into multiple field micro-regions and construct a field micro-region health status map based on the spatial association, agricultural association and historical anomaly association of each field micro-region; The generation module is used to acquire basic sensing data through low-power sensing nodes deployed in field micro-regions, and generate abnormal wake-up factors based on the basic sensing data, node energy consumption status, and associated risks in the field micro-region health status map. The acquisition module is used to determine the hierarchical acquisition strategy for the corresponding low-power sensing node based on the abnormal wake-up factor and energy consumption constraints, and to acquire crop status data according to the hierarchical acquisition strategy. The diagnostic module is used to perform lightweight diagnostics on crop status data at low-power sensing nodes, and generate candidate diagnostic results and a summary of diagnostic evidence to characterize the diagnostic basis. The fusion module is used to determine the collaborative diagnosis range of the target field micro-region based on the field micro-region health status map through the field edge gateway, and to fuse the diagnostic evidence summary of the target field micro-region and the collaborative diagnosis range to generate edge collaborative diagnosis confidence. The update module is used to generate a minimum supplementary sampling instruction based on diagnostic uncertainty and additional energy consumption cost when the confidence level of edge collaborative diagnosis does not meet the confirmation conditions, and to update the confidence level of edge collaborative diagnosis according to the supplementary sampling results. The output module is used to output field diagnostic results carrying the chain of evidence based on the updated edge co-diagnostic confidence, and to update the field micro-plot health status map and diagnostic control parameters based on the diagnostic feedback.

Images