Blue tongue disease monitoring and early warning system based on media monitoring and multi-source data checking

Abstract

The application discloses a blue tongue disease monitoring and early warning system based on medium monitoring and multi-source data checking, relates to the fields of animal disease monitoring and early warning, medium insect monitoring, artificial intelligence data fusion and spatial information processing, and comprises a spatial evidence chain sample library, a prior control quantity generation module, a gate ticket generation module, a medium sentinel terminal, a collaborative evidence chain arranger, a primitive recognition and reasoning engine and a closed-loop updating and layout controller. The prior control quantity generation module outputs a suitable grade grid and an uncertainty grid; the gate ticket generation module generates prior gate tickets and convergent gate tickets based on a calibration loss sequence. The medium sentinel terminal generates a medium evidence event package, and the arranger generates a sampling script transaction; the reasoning engine outputs a consistency score and a conflict type and triggers resampling until a convergent proof mark is generated; and the closed-loop updating and layout controller forms a strategy version package and updates the next round of gate control and layout.

Description

Technical Field

[0001] This invention relates to the fields of animal disease monitoring and early warning, vector insect monitoring, artificial intelligence data fusion and spatial information processing, specifically a bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification. Background Technology

[0002] Bluetongue disease in bovine animals is a typical vector-borne animal disease. Its risk identification and control require consideration of the spatial heterogeneity of vector activity, environmental variables, and changes in the status of farms. Current technologies typically employ various solutions, such as collecting partial sensor data via IoT devices and providing threshold-based early warnings, constructing risk distribution maps of vectors or diseases based on raster data of environmental variables for manual inspections and immunization programs, or using trapping devices to count / classify vectors and report them according to event thresholds.

[0003] However, the aforementioned existing technologies still have certain limitations: triggering and stopping thresholds mostly rely on empirical settings or static rules, making it difficult to calibrate and update them according to changes in the error structure, which can easily lead to false alarms / false negatives and threshold drift; multi-source evidence is often accessed in parallel without a traceable evidence chain object and consistency verification mechanism, making it difficult to locate evidence gaps and generate supplementary evidence paths when there are conflicts between media-side evidence and site-side evidence; sampling and processing are mostly single-trigger or passive responses, lacking forward multi-round collaborative simulation, multi-round benefit evaluation, and strategy update mechanisms under convergence constraints, making it difficult to complete the evidence chain and shorten the convergence path with the fewest rounds while ensuring reliability; there is a lack of a closed-loop update mechanism for convergence proof marking—database entry—strategy version package, which can easily cause low-quality evidence to pollute the sample database, and the adjustment of thresholds / script parameters / deployment density lacks version management and grid-level applicability mapping.

[0004] Therefore, there is an urgent need for a bovine bluetongue disease monitoring and early warning system that can organize media entry events, station status parameter streams, and environmental variable raster data into a traceable, calibrable, and sustainably optimized evidence chain closed loop process within the same closed control chain. This system would enable proactive collaborative generation of sampling script transactions, conflict-driven resampling, and convergence proof storage under the constraints of prior gating tickets and convergence gating tickets. Furthermore, it would complete the versioned reconfiguration and reverse correction of thresholds and deployments through policy version packages. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification to solve the above-mentioned problems.

[0006] The objective of this invention is achieved through the following technical solution: a bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification, comprising: The spatial evidence chain sample library is used to store evidence units. The evidence units include spatiotemporal index, media evidence digest, station evidence digest, conflict type sequence, convergence proof flag, policy version number and event hash fingerprint. The convergence proof flag is used to indicate that the evidence unit meets the stopping threshold defined by the convergence gate ticket. The prior control quantity generation module is used to train a species distribution model with Culicopters as the target species based on spatiotemporal index and environmental variable raster data, and output the suitability level raster and uncertainty raster. The gated ticket generation module is used to form a calibration loss sequence based on the error metric and generate a quantile interval weight vector to obtain the upper bound random quantity of risk, and take the confidence quantile threshold of the upper bound random quantity of risk to generate prior gated tickets and convergent gated tickets. The media sentinel terminal is deployed at the ventilation openings of the aquaculture farm and includes an optical checkpoint detection unit for the ventilation openings. The optical checkpoint detection unit outputs an event availability score and a Culicoides midge candidate confidence score, which are then bound to a fan operating condition label to form a media entry event stream. The media sentinel terminal performs wing vibration fingerprint purification and media classification on the media entry event stream to output a media classification confidence score. When the prior gating ticket allows and the event availability score, Culicoides midge candidate confidence score, and media classification confidence score meet the event gating conditions, it performs DNA signal detection within the media to output a DNA signal confidence score within the media, thereby generating a media evidence event package. A collaborative evidence chain orchestrator is used to receive media evidence event packets and generate sampling script transactions based on prior gated tickets, event availability scores, Coxsmid candidate confidence and convergence gated tickets. The collaborative evidence chain orchestrator performs forward multi-round collaborative simulation on candidate sampling script transactions to generate evidence chain trajectories and calculates multi-round payoff values ​​based on the evidence chain trajectories to update the sampling script transaction generation strategy. The primitive recognition and reasoning engine is used to drive the output of the station state parameter stream from multi-view camera devices and multi-sensor devices according to the sampling script transaction. It extracts primitive vectors and primitive chain completeness from the station state parameter stream, generates herd state micro-drift feature vectors and state consistency deviation and its trend, and performs consistency verification with the media evidence event package to output consistency score and conflict type. Based on the conflict type, it generates resampling script transactions to trigger resampling until the stopping threshold corresponding to the convergence gate ticket is met and generates convergence proof mark. The closed-loop update and deployment controller is used to write the corresponding evidence unit into the spatial evidence chain sample library after the convergence proof marker is generated. Based on the conflict type sequence and error structure, it performs version reconfiguration of the wing vibration fingerprint purification threshold, the DNA signal detection trigger condition in the medium, the sampling script transaction parameters, and the trap deployment density to form a strategy version package, which is used for the generation of prior gating tickets and the update of event gating conditions for the next round of medium entry events.

[0007] Culicoides midges include the sharp-beaked Culicoides oxystoma, the field Culicoides homotomus, the Ryukyu Culicoides actoni, the banded Culicoides tainanus, the skeletal Culicoides imicola, the pale yellow Culicoides fulvus, the short-tarsed Culicoides brevitarsis, and the spotted Culicoides... (jacobsoni), the media evidence summary includes event availability score, Culicoides candidate confidence, media classification confidence, DNA signal confidence within the media, time-series counting features, and wind turbine operating condition labels. The site evidence summary includes statistical descriptions of primitive vectors, primitive chain integrity identifiers, and statistical descriptions of cattle state micro-drift feature vectors, along with state consistency deviation and its trend quantity identifiers. Furthermore, the event hash fingerprint is jointly generated by the spatiotemporal index, media evidence summary, site evidence summary, conflict type sequence, and policy version number for evidence deduplication, tracing, and database verification.

[0008] The species distribution model is a maximum entropy species distribution model. The prior control quantity generation module performs resolution unification and spatial purification processing on the environmental variable raster data to form a training raster package. Based on the training raster package and spatiotemporal index, it generates occurrence point samples to train the maximum entropy species distribution model and outputs the suitability level raster and uncertainty raster.

[0009] The ventilation opening optical checkpoint detection unit includes a laser light curtain emitter spanning the ventilation opening cross-section and a photoelectric receiving array. Based on the scattering waveform output by the photoelectric receiving array, it generates an event availability score and a Cuglossy midge candidate confidence score. The fan operating condition label includes a ventilation opening identifier, fan speed parameters, and airflow setting parameters. These are bound to the event availability score and Cuglossy midge candidate confidence score within the same time window to form a media entry event stream.

[0010] The wing vibration fingerprint cleanup includes extracting wing vibration fingerprint features from the entry signal of the media entry event stream and calculating a cleanup score, classifying the wing vibration fingerprint features into media to generate a media classification confidence score, and prohibiting the corresponding media entry event from participating in the event gating condition determination and marking the corresponding media entry event as a low-confidence event that cannot be used to trigger sampling script transactions when the cleanup score is lower than the cleanup threshold.

[0011] DNA signal detection within the media is initiated when the prior gating ticket allows and the event gating conditions are met. The event gating conditions are jointly limited by the event availability score, Culicoides candidate confidence, media classification confidence, and time-series counting features. The media evidence event package includes a detection time window identifier and a detection process version identifier for association with the policy version number and for time-series alignment in consistency verification.

[0012] The gated ticket generation module sets the quantile interval weight vector to a weight sampling vector that satisfies the Dirichlet distribution constraint, and generates a sample set of risk upper bound random quantities based on multiple weight sampling vectors and the sorted calibration loss sequence. The confidence quantile threshold of the sample set forms the trigger threshold constraint of the prior gated ticket and the stopping threshold constraint of the convergence gated ticket, which are used to limit the triggering, resampling stop and storage of DNA signal detection in the media, respectively.

[0013] The collaborative evidence chain orchestrator includes a collaborative simulator, a multi-round payoff evaluator, and a policy updater. The collaborative simulator performs forward multi-round simulations of candidate sampling script transactions based on media evidence event packages, prior gating tickets, uncertainty grids, and herd state micro-drift feature vectors and state consistency deviations and their trend values ​​to generate evidence chain trajectories. The multi-round payoff evaluator calculates the convergence efficiency and sampling cost of the evidence chain trajectory and generates multi-round payoff values, where the multi-round payoff value is a weighted combination of the convergence efficiency and sampling cost. The policy updater updates the sampling script transaction generation strategy based on the multi-round payoff values ​​so that subsequent sampling script transactions reduce the number of resampling times and shorten the convergence path while satisfying the convergence gating ticket constraints.

[0014] The sampling script transaction limits the collection channel set to a combination of multi-view camera channels and multi-sensor channels, and limits the termination conditions to primitive chain completeness reaching the target threshold, conflict type satisfying the order reduction condition, and sampling budget reaching the upper limit condition. The primitive recognition and inference engine extracts the herd state micro-drift feature vector based on the station state parameter flow within the collection time window, aligns the herd state micro-drift feature vector with the historical baseline feature vector to calculate the state consistency deviation, and performs sliding time window trend analysis on the state consistency deviation to generate trend quantity. The trend quantity is used as the input constraint for generating the resampling script transaction. The herd state micro-drift feature vector includes at least the activity spectrum feature, trajectory dwell hotspot feature, gait rhythm feature, feeding event frequency and duration feature, drinking event frequency and duration feature, posture stability feature, and herd structure feature.

[0015] The conflict type is determined by a combination of the suitability level and uncertainty level identifiers corresponding to the prior gating tickets, the confidence and temporal count features of DNA signals within the media, the completeness of the primitive chain, the herd state micro-drift feature vector, the state consistency deviation, the trend quantity, and the primitive missing item list. Specifically, a high prior uncertainty conflict is defined as a conflict where the uncertainty level is high and the media-side evidence is weak; a strong media-weak-field conflict is defined as a conflict where the confidence of DNA signals within the media is high and the field-side evidence is weak; and a weak media-strong-field conflict is defined as a conflict where the field-side evidence is high and the media-side evidence is weak. A conflict is further defined as a conflict where the primitive missing item list is not empty and the primitive chain... When the integrity is lower than the target threshold, it is judged as an incomplete chain conflict. The conflict type is used to generate supplementary evidence paths for resampling script transactions. The strategy version package includes a wing vibration fingerprint purification threshold version, a DNA signal detection trigger condition version in the vector body version, a gated ticket confidence quantile threshold version, a sampling script transaction parameter version, and a trap deployment density version, and is mapped to the grid-level applicable range. The grid-level applicable range is determined by the grid-level reconfiguration intensity, which is calculated by the conflict type sequence statistical count and the average error metric to update the prior gated ticket generation rules, event gating conditions, and incremental training sample selection rules of the species distribution model.

[0016] The beneficial effects of this invention are: The spatial evidence chain sample library stores evidence units including spatiotemporal indexes, media evidence summaries, site evidence summaries, conflict type sequences, convergence proof markers, policy version numbers, and event hash fingerprints. It uses event hash fingerprints to achieve evidence deduplication, traceability, and database verification, enabling data generated from different batches, devices, and time windows to be consistently indexed, compared, and reused under a unified data object structure. This avoids resource waste caused by repeated collection and calculation and improves the manageability of the system under multi-terminal and multi-site deployment conditions.

[0017] The prior control quantity generation module trains a species distribution model with Culicoides midge as the target species based on the spatiotemporal index of evidence units and environmental variable raster data, and outputs a suitability level raster and an uncertainty raster. This enables subsequent gating and scheduling to differentiate based on spatial heterogeneity and prior uncertainty, thereby achieving targeted control over resource allocation and sampling paths even in scenarios with uneven risk spatial distribution and rapid environmental changes, reducing the bandwidth, energy consumption and manpower costs brought about by full-coverage sampling.

[0018] The gated ticket generation module generates a calibration loss sequence based on the error metric of the evidence unit and generates a quantile interval weight vector. After sorting the calibration loss sequence, it linearly combines it with the quantile interval weight vector to obtain a random upper bound of risk. Then, it generates a priori gated tickets and convergence gated tickets by taking the confidence quantile threshold. This ensures that the triggering and stopping of key actions have a calibrable threshold constraint source, avoiding drift and uninterpretable problems caused by relying solely on empirical thresholds. The priori gated tickets are used to limit the triggering of DNA signal detection and sampling script transactions in the media, while the convergence gated tickets are used to limit the stopping of resampling and storage. This can suppress unnecessary detection and sampling while ensuring reliability, improve closed-loop convergence efficiency, and reduce the proportion of invalid samples.

[0019] The media sentinel terminal is deployed at the ventilation openings of the aquaculture farm and includes an optical checkpoint detection unit. It outputs an event availability score and a midge candidate confidence score, which are then bound to a fan condition tag to form a media entry event stream. The stream is further processed by wing vibration fingerprint purification and media classification to output a media classification confidence score. When prior gating passes allow and the event availability score, midge candidate confidence score, and media classification confidence score meet the event gating conditions, DNA signal detection within the media is performed to output a DNA signal confidence score, thus generating a structured media evidence event package. This process implements a hierarchical processing mechanism of screening before detection. Wing vibration fingerprint purification at the terminal reduces the probability of low-confidence events entering subsequent processes. Then, when the pass allowance and event gating conditions are met, DNA signal detection within the media is triggered, focusing high-cost detection on high-value events and reducing the costs associated with false triggers and redundant detection. Simultaneously, the fan condition tag and time-series counting features enhance the characterization of event background and temporal stability, improving the availability and consistency of media-side evidence.

[0020] The collaborative evidence chain orchestrator receives media evidence event packets and generates sampling script transactions based on prior gating tickets, event availability scores, Cumming midge candidate confidence levels, and convergence gating tickets. It then performs forward multi-round collaborative simulations on candidate sampling script transactions to generate evidence chain trajectories and calculates multi-round payoff values ​​based on these trajectories to update the sampling script transaction generation strategy. This solidifies the sampling object binding rules, collection channel set, collection time window, sampling intensity parameters, termination conditions, and convergence objectives into executable scheduling instructions in a transactional script format. This enables the automatic generation of supplementary evidence paths when evidence is insufficient or conflicting, and iterative optimization of the sampling paths based on multi-round payoff values. This transforms sampling from passive triggering to proactive scheduling oriented towards the convergence objective, reducing blind and duplicate sampling and shortening the path length from triggering to convergence.

[0021] The primitive recognition and reasoning engine drives the output of station state parameter streams from multi-view camera devices and multi-sensor devices based on sampling script transactions. It extracts primitive vectors and primitive chain completeness from the station state parameter streams, generating herd state micro-drift feature vectors and state consistency deviation and its trend. The primitive vectors, primitive chain completeness, herd state micro-drift feature vectors, and state consistency deviation and its trend are then compared with the media evidence event package for consistency verification to output a consistency score and conflict type. This structure expresses the fusion results of multi-source data as interpretable objects such as primitive vectors and primitive chain completeness. Through consistency scores and conflict types, it provides a unified conflict characterization of media-side evidence and station-side evidence, enabling the system to locate the source of evidence gaps and generate resampling script transactions to trigger resampling. This avoids the problem of outputting only a single conclusion without supplementary evidence, improving the interpretability and verifiability of the decision-making chain.

[0022] The closed-loop update and deployment controller only writes the corresponding evidence unit into the spatial evidence chain sample library after the convergence proof marker is generated. Based on the conflict type sequence and error structure, it re-versions the wing vibration fingerprint purification threshold, the DNA signal detection trigger condition within the medium, the sampling script transaction parameters, and the trap deployment density to form a strategy version package. This package is used for the generation of prior gating tickets and the updating of event gating conditions for the next round of medium entry events. This post-convergence storage mechanism prevents non-converged or low-quality evidence from contaminating the sample library, ensuring the usability of evidence units. Simultaneously, the strategy version package manages the adjustment of key thresholds and parameters in a versioned manner and maps them to a grid-level applicable range. This allows the system to implement differentiated deployment and gating for uncertainties and conflict distributions in different regions, forming a sustainable self-correction and self-optimization closed loop, improving long-term robustness and adaptability. Attached Figure Description

[0023] Figure 1 The process of this invention Figure 1 ; Figure 2 The process of this invention Figure 2 ; Figure 3 The process of this invention Figure 3 . Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1 like Figure 1 As shown, this embodiment provides a bluetongue disease monitoring and early warning system based on media monitoring and multi-source data verification. The system consists of a spatial evidence chain sample library, a priori control quantity generation module, a gated ticket generation module, a media sentinel terminal, a collaborative evidence chain orchestrator, a primitive recognition and reasoning engine, and a closed-loop update and deployment controller. Each module is connected via wired Ethernet or a 4G / 5G private network, and uses a unified spatiotemporal indexing standard and version number standard to achieve end-to-end traceability. This embodiment uses a farm as the deployment target, divided into several spatial grid units with a grid resolution of 50m×50m or 100m×100m. The spatiotemporal index is represented by a combination of grid ID and collection start and end timestamps, where the timestamps are UTC millisecond timestamps. During system operation, the sequence of media entry event → evidence scripting sampling → primitive reasoning and consistency verification → conflict-driven resampling → convergence proof → database entry and version reconfiguration is executed continuously, with uninterrupted data processing.

[0026] The spatial evidence chain sample library is used to store evidence units. These units are stored in a relational database or key-value database in structured record form. Simultaneously, a large volume of raw waveforms, image / video clips, and raw sensor streams are stored in object storage and linked via pointers. Each evidence unit includes at least: a spatiotemporal index. The document outlines the following components: a media evidence summary; a site evidence summary; a conflict type sequence (a chronologically ordered sequence of conflict type codes); a convergence proof marker; a policy version number; and an event hash fingerprint. The media evidence summary includes at least an event availability score, a Coxswain candidate confidence level, a media classification confidence level, a DNA signal confidence level within the media, time-series counting features, and a wind turbine operating condition label. The site evidence summary includes at least a statistical description of primitive vectors, a primitive chain integrity identifier, a statistical description of cattle herd state micro-drift feature vectors, and a state consistency deviation and its trend identifier. The policy version number is a string or an incrementing integer version (e.g., V20260120.001), used to identify the configuration version of control parameters such as the current gating threshold, sampling script parameters, and trap deployment density. The event hash fingerprint is used for evidence deduplication, tracing, and database verification. In this embodiment, SHA-256 is used to hash the normalized serialization result of the spatiotemporal index + media evidence summary + site evidence summary + conflict type sequence + policy version number, outputting a 64-bit hexadecimal string. Culicoides include Culicoides oxystoma, Culicoides homotomus, Culicoides actoni, Culicoides tainanus, Culicoides imicola, Culicoides fulvus, Culicoides brevitarsis, and Culicoides jacoboni.

[0027] The data sources for the prior control quantity generation module include two categories: one is the spatiotemporal index of evidence units already included in the spatial evidence chain sample library; the other is environmental variable raster data. Environmental variable raster data can come from publicly available or self-built geographic raster products and meteorological raster products. In this embodiment, the environmental variable raster data includes at least elements such as temperature, relative humidity, rainfall, wind speed, vegetation index, and distance to water bodies. The data format is GeoTIFF or equivalent raster format, the spatial reference system is WGS84 or the National Geodetic Coordinate System, and the raster time resolution is 1 hour or 3 hours.

[0028] The prior control generation module performs resolution unification and time alignment processing on the environmental variable raster data: resampling each raster to the same resolution as the grid cell (50m or 100m), and aligning the raster time slices to the system's event time window (e.g., using...). (The time slice is aligned forward to the nearest grid cell). Subsequently, the prior control quantity generation module constructs occurrence point samples based on the spatiotemporal index: the grid ID and time window corresponding to the media-related events marked as valid after convergence proof in the evidence unit are used as occurrence points, and other grid cells in the same region and season are used as background points to train a maximum entropy species distribution model with Culicoides midge as the target species. In this embodiment, the maximum entropy species distribution model outputs a suitability level grid and an uncertainty grid. The suitability level grid is a floating-point grid with a value range of [0,1] and is discretized into a suitability level identifier according to a threshold (e.g., 0-0.33 is low, 0.33-0.66 is medium, and 0.66-1 is high). The uncertainty grid is a floating-point grid with a value range of [0,1] and is discretized into an uncertainty level identifier according to a threshold (e.g., 0-0.4 is low, 0.4-0.7 is medium, and 0.7-1 is high).

[0029] The gated ticket generation module uses the evidence units already included in the spatial evidence chain sample library as the calibration data source, calculates an error metric for each evidence unit, and forms a calibration loss sequence. To ensure feasibility, this embodiment defines the error metric as a complementary quantity to the consistency score: let the consistency score corresponding to the i-th evidence unit be... Then the error metric (calibration loss) is The gated ticket generation module will use the calibration loss sequence Sorted from smallest to largest And generate quantile interval weight vector ,in and In this embodiment It is generated from a weighted sampling vector that satisfies the Dirichlet distribution constraint. The gated ticket generation module linearly combines the sorted calibration loss sequence with the quantile interval weight vector to obtain the upper bound random variable of risk. Its calculation formula is in when and order The gated ticket generation module uses repeated sampling. And calculate The sample set that forms the upper bound of the risk random quantity ,in Take values ​​from 1000 to 10000; at the confidence level Below (for example) or The empirical quantile of the sample set is taken as the confidence quantile threshold. The gated ticket generation module will Write prior gating tickets and convergence gating tickets: Prior gating tickets are used to limit the triggering threshold constraints of DNA signal detection and sampling script transactions within the medium, while convergence gating tickets are used to limit the stopping threshold constraints of resampling and storage. Each gating ticket must include at least a ticket number. Confidence quantile threshold Confidence level The applicable time and space range (one or more grid IDs and their corresponding validity periods), and the policy version number.

[0030] The media sentinel terminal is deployed at the ventilation openings of the aquaculture farm and includes an optical checkpoint detection unit for the ventilation openings. The optical checkpoint detection unit has a laser light curtain emitter and a photoelectric receiving array on both sides of the ventilation opening cross-section. The sampling frequency of the photoelectric receiving array is 10kHz to 50kHz, outputting a scattering waveform. Within each detection time window (e.g., 100ms to 500ms), the system extracts features from the scattering waveform and calculates the event availability score and Culex midge candidate confidence level, both ranging from [0,1]. The fan operating condition tag consists of a ventilation opening identifier, fan speed parameters, and airflow setting parameters. The fan speed parameters can be read by the fan controller (e.g., 0-3000rpm), and the airflow setting parameters are control setpoints (e.g., 0-100%). The media sentinel terminal binds the fan operating condition tag with the event availability score and Culex midge candidate confidence level according to the same detection time window to form a media entry event stream.

[0031] The media sentinel terminal performs wing-vibration fingerprint cleanup and media classification on the media entry event stream to output a media classification confidence score. In this embodiment, wing-vibration fingerprint cleanup includes: extracting wing-vibration fingerprint features from the entry signal, where the wing-vibration fingerprint features include at least the dominant frequency, harmonic energy ratio, and spectral entropy; calculating a cleanup score (value [0,1]) based on the wing-vibration fingerprint features; comparing the cleanup score with a wing-vibration fingerprint cleanup threshold; and prohibiting the corresponding media entry event from participating in event gating condition determination and marking it as a low-confidence event that cannot be used to trigger sampling script transactions when the cleanup score is lower than the wing-vibration fingerprint cleanup threshold. For events that pass cleanup, the media sentinel terminal performs media classification on the wing-vibration fingerprint features and outputs a media classification confidence score (value [0,1]).

[0032] When the prior gating ticket allows and the event gating conditions are met, the media sentinel terminal performs DNA signal detection within the media to output the confidence level of the DNA signal within the media. The event gating conditions are jointly defined by the event availability score, Culicoides candidate confidence level, media classification confidence level, and time-series counting features. The time-series counting features are obtained by statistically analyzing the valid event count sequences within multiple consecutive detection time windows, and include at least the valid event count per unit time. (e.g., the number of events that were cleaned and classified as Culicoides midges in the past 5 minutes), the moving average of the counts. (e.g., the mean of the past 10 time windows) and the coefficient of variation of the count (Standard deviation divided by mean); In this embodiment, the event gating condition adopts a threshold joint judgment method: the event availability score is not lower than the first threshold, the candidate confidence of Culicoides is not lower than the second threshold, the media classification confidence is not lower than the third threshold, and the time series counting feature meets the counting lower limit requirement (e.g. The first threshold, second threshold, third threshold, and lower count limit are determined by the trigger threshold constraint of the prior gating ticket and the current policy version number. The detection time window for DNA signal detection in the media is 1 min to 10 min. The detection process version identifier is used to identify the reagent batch, amplification process, or interpretation model version used. The media sentinel terminal outputs the confidence level of DNA signal in the media and writes the detection time window identifier and the detection process version identifier into the media evidence event package for association with the policy version number and for timing alignment in consistency verification.

[0033] The media evidence event package, as the input data object of the collaborative evidence chain orchestrator, is encapsulated in this embodiment using a structured message format (JSON, Protobuf, or equivalent) and includes: a priori gated tickets ( , The system includes: scope of application, strategy version number, event availability score, Coccinelli candidate confidence, wind turbine operating condition label, media classification confidence, DNA signal confidence in the media, time series counting features, spatiotemporal index, detection time window identifier, and detection process version identifier.

[0034] The collaborative evidence chain orchestrator receives media evidence event packets and generates sampling script transactions based on prior gating tickets, event availability scores, Cumming midge candidate confidence scores, and convergence gating tickets. In this embodiment, the sampling script transaction is an executable scheduling transaction record, including sampling object binding rules, collection channel set, collection time window, sampling intensity parameters, termination conditions, and convergence objectives. The sampling object binding rule is used to bind the sampling object to the spatiotemporal index. In this embodiment, it includes a set of cattle identifiers and a sampling area identifier. The cattle identifiers can come from electronic ear tags or visual ReID. The sampling channel set includes at least a multi-view camera channel and a multi-sensor channel. The multi-view camera channel includes a fixed top-view camera and a side-view camera, with a frame rate of 10fps to 25fps and a resolution of 1080p or higher. The multi-sensor channel includes at least a feeding event sensor, a drinking event sensor, and an environmental sensor. The sampling time window is 5min to 60min. The sampling intensity parameters include the camera frame rate, the sensor sampling frequency (1Hz to 10Hz), and the data upload frequency. The termination conditions include at least the primitive chain integrity reaching the target threshold, the conflict type meeting the order reduction condition, and the sampling budget reaching the upper limit condition. The convergence target is used to constrain the resampling stop and the data entry. In this embodiment, it is composed of the stop threshold constraint of the convergence gating ticket and the preset primitive chain integrity target threshold.

[0035] The collaborative evidence chain orchestrator performs forward multi-round collaborative simulation on candidate sampling script transactions to generate evidence chain trajectories, and calculates multi-round payoff values ​​based on the evidence chain trajectories to update the sampling script transaction generation strategy. For ease of implementation, in this embodiment, the collaborative evidence chain orchestrator defines the multi-round payoff value as a computable combination of numerical indicators, including convergence efficiency and sampling cost. The convergence efficiency reflects the number of resampling attempts and the time required to reach the stopping threshold, while the sampling cost reflects the amount of data, energy consumption, and bandwidth usage generated by multi-view cameras and multi-sensor sampling. The formula for calculating the multi-round payoff value is as follows: in The number of rounds required to achieve convergence, Weights for convergence efficiency (e.g., 0.7). The sampling cost weight (e.g., 0.3), For the first The amount of data generated by round sampling (in MB) For the first Energy consumption per round of sampling (in Wh), and The normalization coefficient for data volume and energy consumption (e.g.) , ); Collaborative evidence chain orchestrator selection Larger sampling script transactions are issued and executed, while the parameter search range of subsequent candidate script transactions is updated, so that subsequent sampling script transactions reduce the number of resampling times and shorten the convergence path while satisfying the convergence gate ticket constraint.

[0036] The primitive recognition and inference engine drives the output of the farm's state parameter stream from multi-view cameras and multi-sensor devices based on the sampling script transaction. The data sources for the farm's state parameter stream include: video frame sequences, event sequences output from feeding and drinking event sensors, and numerical sequences output from environmental sensors. The primitive recognition and inference engine performs individual detection and tracking on the video frame sequences, generating a trajectory sequence for each cow, and extracts activity spectrum features, trajectory dwell hotspot features, gait rhythm features, posture stability features, and herd structure features within the acquisition time window; it also statistically analyzes the frequency and duration of feeding and drinking events in the feeding and drinking event sequences; and arranges these features according to a preset feature dictionary to form a herd state micro-drift feature vector. The primitive recognition and inference engine further calculates the state consistency deviation and its trend: in this embodiment, historical baseline feature vectors are selected. (For example, take the mean vector of the same time period over the past 7 days), and calculate in For the current time window The herd state micro-drift feature vector, For historical baseline feature vectors, For the feature deviation vector, The deviation from state consistency. It is the Euclidean norm.

[0037] The primitive recognition and inference engine uses a sliding time window (e.g., a window length of 6 acquisition time windows) to... Trend analysis is performed to obtain trend indicators (e.g., rising / stable / falling) and corresponding trend values ​​(e.g., slope or growth rate). The primitive vectors and primitive chain completeness are output together. In this embodiment, primitive chain completeness is defined as the ratio of the number of successfully extracted primitives to the preset total number of primitives, with a value range of [0,1]. A list of missing primitives is also output for subsequent conflict type determination.

[0038] The primitive recognition and inference engine performs consistency checks on primitive vectors, primitive chain completeness, cattle herd state micro-drift feature vectors, and state consistency deviation and its trend quantity with the media evidence event package to output a consistency score and conflict type. In this embodiment, consistency checks are based on time alignment: the detection time window identifier in the media evidence event package is mapped to the acquisition time window of the site state parameter stream, and within the same time window, the key quantities of the media evidence event package (event availability score, Culicoides candidate confidence, media classification confidence, media DNA signal confidence, time-series counting features, wind turbine operating condition label) are compared with the key quantities of the site evidence summary (primitive chain completeness, ... The media-side evidence vector is obtained by vectorizing and combining trend indicators and micro-drift characteristic vector statistics. Evidence vector from the station side Calculate the consistency score between the two. Its calculation formula is in Let be the Euclidean distance between the evidence vectors from the media side and the evidence vectors from the station side. This is a scaling parameter (values ​​range from 0.5 to 2.0). For the current primitive chain completeness, The reference completeness is 0.85; the smaller the difference and the closer the chain completeness is to the reference value, the higher the consistency score. The conflict type is determined by the following information: the susceptibility level identifier and uncertainty level identifier corresponding to the prior gate ticket, the confidence level and temporal count characteristics of the DNA signal in the media evidence event package, the primitive chain completeness, the herd state micro-drift feature vector, the state consistency deviation degree and its trend, and the list of missing primitives; the conflict type is represented by discrete coding, and in this embodiment, it includes at least the following categories and their codes: prior high susceptibility high uncertainty conflict (code 1), strong media weak field conflict (code 2), weak media strong field conflict (code 3), chain incomplete conflict (code 4), temporal inconsistency conflict (code 5), and can be extended to combined conflict coding.

[0039] The primitive recognition and inference engine generates resampling script transactions based on conflict types to trigger resampling until the stopping threshold corresponding to the convergence gate ticket is met and a convergence proof flag is generated. In this embodiment, the supplementary evidence path of the resampling script transaction is determined by the conflict type: when the conflict type is an incomplete chain conflict, missing channels in the acquisition channel set are added or the acquisition time window is extended; when the conflict type is a strong media-site weak conflict, the coverage of multi-view camera channels is increased or the sampling frequency of multi-sensor channels is increased to improve the primitive chain completeness; when the conflict type is a weak media-site strong conflict, the sampling intensity parameter of the media sentinel terminal is increased or the observation angle of the vent position is adjusted to improve the event availability score; when the conflict type is a temporal inconsistency conflict, the alignment offset between the detection time window identifier and the acquisition time window is adjusted and sampling is retried. After each round of resampling, the primitive recognition and inference engine updates the consistency score. And calculate the current error metric. ,when When the primitive chain completeness is not lower than the target threshold (e.g., 0.85), the stopping threshold is determined to be met, and a convergence proof marker is generated. In this embodiment, the convergence proof marker is a structured marker, which includes at least a convergence timestamp, the round number that met the stopping threshold, and the corresponding... And the strategy version number.

[0040] The closed-loop update and deployment controller only writes the corresponding evidence unit into the spatial evidence chain sample library after the convergence proof marker is generated. Based on the conflict type sequence and error structure, it performs versioned reconfiguration of the wing vibration fingerprint purification threshold, the DNA signal detection trigger condition within the medium, the sampling script transaction parameters, and the trap deployment density to form a strategy version package. In this embodiment, the error structure consists of the conflict type sequence and each round... Composed of the primitive chain integrity sequence and the key quantity sequence of the media evidence event package, the closed-loop update and deployment controller statistically corresponds to the threshold sensitivity direction for different conflict types: if the chain is incomplete and conflicts occur frequently, the coverage of the collection channel set is increased or the default lower limit of the sampling intensity parameter is increased; if the media is strong and the field is weak and conflicts occur frequently, the primitive chain integrity target threshold is increased or the collection time window is extended; if the media is weak and the field is strong and conflicts occur frequently, the count lower limit in the event gating condition is lowered or the triggering condition for DNA signal detection in the media is adjusted; if the prior high uncertainty conflicts occur frequently, the trap deployment density is increased in the high uncertainty grid and the grid-level applicable range is written into the strategy version package. The strategy version package includes at least the wing vibration fingerprint purification threshold version, the triggering condition for DNA signal detection in the media version, the gate ticket confidence quantile threshold version, the sampling script transaction parameter version, and the trap deployment density version, and is mapped to the grid-level applicable range to update the next round of prior gate ticket generation rules, event gating conditions, and incremental training sample selection rules for the species distribution model. This completes the closed control chain from the media entry event to resampling convergence, then to backfilling update and reverse correction, and begins the next round of continuous operation of the media entry event.

[0041] Example 2 like Figure 1 and Figure 2 As shown, based on the overall system framework described in Embodiment 1, this embodiment further provides the hardware structure and data processing flow of the media sentinel terminal, focusing on disclosing the end-side implementation details of the ventilation optical checkpoint detection unit, wing vibration fingerprint purification, media classification, and DNA signal detection within the media, as well as its continuous data link with prior gate tickets, event gate conditions, detection time window identifiers, and detection process version identifiers.

[0042] The media sentinel terminal is deployed at the ventilation openings of the farm. The ventilation openings can be either negative or positive pressure ventilation structures, with a cross-sectional width of 0.3m to 1.2m and a height of 0.3m to 1.2m. The media sentinel terminal includes a ventilation opening optical checkpoint detection unit, which comprises a laser light curtain emitter and a photoelectric receiving array spanning the cross-section of the ventilation opening. The laser light curtain emitter is positioned on one side of the ventilation opening, and the photoelectric receiving array is positioned on the opposite side. The two emit light onto each other on the same horizontal or vertical plane, forming a light curtain with a thickness of 2mm to 10mm. In this embodiment, the laser light curtain emitter uses a semiconductor laser or a narrowband infrared emitter, with a center wavelength selected as [missing information]. or The output power is 5mW to 30mW, and a linear light curtain covering the cross-section of the ventilation opening is formed through a linear beam expander lens. In this embodiment, the photoelectric receiver array consists of... Composed of a photodiode or CMOS linear array Choose 32, 64, or 128, sampling frequency The light intensity voltage value is output at each sampling point from 20kHz to 50kHz and then converted into a discrete sequence through analog-to-digital conversion, thereby outputting the scattering waveform.

[0043] The optical bayonet detection unit for ventilation openings slices the scattering waveform using a fixed detection time window, with the detection time window length being... A time interval of 200ms is used, with window intervals ranging from 50ms to 200ms. For each detection time window... The original sampling sequence of each receiving channel First, perform baseline correction and amplitude normalization: select the background segment 20ms before the window to calculate the baseline. with standard deviation The normalized sequence is obtained. in For sampling point index, This represents the total number of sampling points within the detection time window. To prevent division by zero regularization constants, For the first The receiving channel is in the first The original light intensity voltage value of each sampling point The baseline mean of the background segment. The standard deviation of the background segment is then used. Subsequently, event detection is performed on the normalized sequence to determine whether there are occlusion / scattering events crossing the light curtain. In this embodiment, the energy aggregation amount across channels is used as the event triggering statistic, and is defined as follows: in This represents the total number of channels in the optoelectronic receiver array. Sampling frequency, To detect the length of the time window, This is the normalization threshold (used to filter out low-amplitude noise). This indicates the effective signal amplitude exceeding the threshold. When When the detection time window is determined to contain candidate crossing events, among which The energy trigger threshold is adaptively calibrated based on the background noise level of the vent and the operating conditions of the fan.

[0044] After determining the existence of a candidate crossing event, the ventilation vent optical checkpoint detection unit generates an event availability score and a Culex midge candidate confidence score. The event availability score is used to characterize the signal quality and readability of the event, avoiding inputting saturated, incompletely obstructed, or multi-target superimposed events into subsequent links; in this embodiment, the event availability score is jointly determined by the signal-to-noise ratio and the effective obstruction duration. First, the peak amplitude within the window is calculated. Reference value of standard deviation of background noise Constructing signal-to-noise ratio index At the same time The proportion of non-zero samples approximates the proportion of event duration. The event availability score is defined as follows: in For the Sigmoid function, the parameter takes... , , , This results in low signal-to-noise ratio or excessively short event scores approaching 0, and high-quality event scores approaching 1.

[0045] The Culex midge candidate confidence score is used to characterize the probability that the crossing event was caused by Culex midges. In this embodiment, it is input from the morphological features of the scattering waveform and the multi-channel spatial consistency features to the output of the binary classification model. The morphological features include at least: energy. Peak width (Percentage of consecutive sample lengths exceeding the threshold), kurtosis Consistency with Channel (Inverse normalization of multi-channel peak time difference). The above features are combined to form a feature vector. Subsequently, the candidate confidence level of Culicoides was defined as follows: in and The output is the model parameters obtained by training on offline labeled data. .

[0046] The media sentinel terminal synchronously collects wind turbine operating condition tags and binds them to event assessment results to form a media entry event stream. Wind turbine operating condition tags include vent identification. Fan speed parameters With air volume setting parameters ,in The data is read by the wind turbine frequency converter or controller via Modbus / RS485, with a sampling frequency of 1Hz. For the fan control setpoint (0% to 100%), and Synchronous recording. The media sentinel terminal records the start and end timestamps of the detection time window. and Perform time window binding and generate media entry event stream records. Each record must include at least a spatiotemporal index, a detection time window identifier, and , Related to the operating condition label for the wind turbine.

[0047] The media sentinel terminal performs wing-vibration fingerprint cleanup and media classification on the media entry event stream to output a media classification confidence score. Wing-vibration fingerprint cleanup is achieved on the terminal side by extracting wing-vibration fingerprint features from the entry signal; the entry signal can be a single-channel high signal-to-noise ratio waveform of the photoelectric receiving array when a candidate event occurs, or a multi-channel principal component projection waveform. The entry signal is denoted as... ,right The amplitude spectrum is obtained by applying the Hanning window and calculating the discrete Fourier transform. In the frequency range Searching for the main peak frequency As the dominant frequency of wing vibration, among which , Simultaneously calculate the harmonic energy ratio. With spectral entropy The harmonic energy ratio is defined as follows: Take values ​​from 3 to 5; spectral entropy Calculated from the normalized spectral distribution and normalized to Pulse continuity index The periodic stability of the wing vibration waveform is characterized by the following formula: in For the input signal The sequence of time intervals between adjacent zero crossing points, For standard deviation calculation, For mean calculation, Used to prevent division by zero; The closer the value is to 1, the better the pulse continuity. Composition of wing vibration fingerprint feature vector .

[0048] In this embodiment, the cleanliness score is used to determine whether the event has stable wing-beat characteristics and is not a significant noise or mechanical vibration artifact. The cleanliness score is defined as follows: in and Output parameters that are preset or obtained by fitting labeled data. Fingerprint purification threshold of wing vibration The value is given by the configuration corresponding to the current policy version number, and ranges from 0.4 to 0.8; when When this happens, the media sentinel terminal prohibits the corresponding media from participating in the event gating condition determination and marks the event as a low-confidence event that cannot be used to trigger sampling script transactions. For events that pass the cleanup, the media sentinel terminal performs media classification, which in this embodiment uses a multi-class classifier. and scattering waveform morphology characteristics The combined features are used for classification, and the posterior probability of the Culex midge class is used as the mediating classification confidence score, which is calculated as follows: in Fingerprint features of wing vibration and scattering waveform morphology characteristics The concatenated vector, Total number of media categories (e.g.) ), and The classifier parameters corresponding to the Culicoides midge category. and For the first Classifier parameters for the class, Media Sentinel Terminal will Write the medium into the event stream record.

[0049] When the prior gating ticket allows and the event gating condition is met, the media sentinel terminal performs DNA signal detection within the media to output the confidence level of the DNA signal within the media. The prior gating ticket is sent to the media sentinel terminal by the gating ticket generation module and includes at least a confidence quantile threshold. Confidence level The applicable time and space range and policy version number are considered. After receiving a ticket, the media sentinel terminal matches the current event's time and space index. Only events within the applicable scope and validity period of the ticket are allowed to enter the DNA detection link. Event gating conditions are determined by event availability scoring. Cumin candidate confidence Media Classification Confidence Together with the time-series counting feature, which is defined as the number of valid events per unit time, this feature is used to define the time-series counting feature. express, For the past The number of events that pass purification and are classified as Culicoides midges within seconds. To ensure the feasibility of the determination, this embodiment uses a joint threshold condition: when If both conditions are met simultaneously, the event gating condition is deemed satisfied. , , , These are the event availability score threshold, the Coober midge candidate confidence threshold, the media classification confidence threshold, and the lower limit of the effective event count. These thresholds are determined by the configuration corresponding to the policy version number and can be updated in a closed loop and reconfigured by the deployment controller versioning.

[0050] In this embodiment, the detection of DNA signals within the vector can be achieved using a microfluidic sampling and amplification detection integrated module or an equivalent nucleic acid signal detection module. The input is the vector sample or vector sample lysis buffer collected through a trapping structure, and the detection time window is 1 to 10 minutes. Fluorescent or electrochemical signal sequences are acquired during the detection process. Background subtraction is performed, and the peak response after background subtraction is defined. Standard deviation of background noise Constructing a normalized response And map it to the confidence level of DNA signals in the mediator. in For the Sigmoid function, This is the slope control parameter. For the starting point parameters, For normalized response, To prevent division by zero regularization constants, the Media Sentinel terminal will combine the detection time window identifier (start and end timestamps) and the detection process version identifier (used to identify the chip batch / reagent batch / interpretation model version) corresponding to this DNA test with... Record them together to ensure that subsequent consistency checks can perform timing alignment and version tracing.

[0051] After completing the above processing, the media sentinel terminal generates a media evidence event package and outputs it to the collaborative evidence chain orchestrator. In this embodiment, the media evidence event package includes at least: a priori gating tickets (…). , , (Applicable time and space range, policy version number), event availability score Cumin candidate confidence Fan operating condition labels (ventilation outlet markings, fan speed parameters, airflow setting parameters), media classification confidence level Confidence of DNA signals in the mediator Time-series counting characteristics The system includes a spatiotemporal index, a detection time window identifier, and a detection process version identifier. After receiving the media evidence event packet, the collaborative evidence chain orchestrator uses the prior gating documents, event availability score, and Culex midge candidate confidence level as gating inputs to generate a sampling script transaction, which is then sent to the site acquisition terminal. This enables a continuous data processing process, starting from the scattering waveform generated by the ventilation vent optical checkpoint detection unit, through event assessment, wing vibration fingerprint purification, media classification, DNA signal detection, and finally, the output of the media evidence event packet.

[0052] Example 3 like Figures 1 to 3As shown, based on the overall system framework and end-side acquisition link described in Embodiments 1 and 2, this embodiment further provides an integrated implementation method for the prior control quantity generation module, the gated ticket generation module, and the collaborative evidence chain orchestrator, which integrates prior modeling, ticket calibration, collaborative simulation, policy update, conflict-driven supplementary evidence, and versioned reconfiguration. This enables the determination of data sources, processing procedures, and output objects to realize the adaptation level grid, uncertainty grid, prior gated tickets, convergence gated tickets, sampling script transactions, evidence chain trajectory, multi-round payoff values, conflict types, supplementary evidence paths, and policy version packages, while ensuring that the data processing process is coherent and uninterrupted.

[0053] The data sources for the prior control quantity generation module include environmental variable raster data and spatiotemporal indexes of evidence units with convergence proof markers already generated in the spatial evidence chain sample library. The environmental variable raster data is provided in a gridded form by time slice, and includes at least elements such as temperature, relative humidity, rainfall, wind speed and direction, vegetation index, distance to water bodies, altitude, and land use type. The data format can be GeoTIFF, NetCDF, or equivalent raster format; the spatial reference system is uniformly WGS84 or the National Geodetic Coordinate System, the timestamp uses UTC, and the time resolution is 1 hour to 24 hours. The prior control quantity generation module first performs resolution unification processing on the environmental variable raster data, resampling all elements to a spatial resolution consistent with the business grid (e.g., 50m×50m or 100m×100m), and uses nearest neighbor resampling for discrete elements (such as land use type), and bilinear or bicubic resampling for continuous elements (such as temperature and humidity); then it performs spatial purification processing. In this embodiment, spatial purification aims to reduce spatial autocorrelation, and performs minimum distance screening on the occurrence point samples, with a minimum distance threshold of 100m to 500m, so that adjacent occurrence points are not less than this threshold in space; and for temporal purification, it adopts the rule that the same grid retains at most one occurrence record within the same natural day.

[0054] Occurrence point samples are generated from the spatiotemporal index of the evidence unit: when the convergence proof flag of the evidence unit is valid, its corresponding grid ID and time window are used as occurrence point samples. ,in Indicates the appearance, This is the feature vector of the environmental variable raster within this time window; background point samples are randomly sampled from the same geographic area, with the number of background points ranging from 5 to 20 times the number of occurrence points, and background point labels. To avoid model bias caused by class imbalance, this embodiment uses downsampling for background points or upsampling for occurrence points, with the sample weights ranging from 1 to 10.

[0055] The prior control generation module trains a maximum entropy species distribution model targeting Culicoides midge. For feasibility, this embodiment implements the maximum entropy species distribution model as a logistic regression with a regularization term, and uses environmental element vectors. A linear combination of these probabilities is used as the logarithmic odds to obtain the fitness probability output. : in For environmental element vectors, For its dimensions, For the Sigmoid function, For the weight vector, This is the bias term. The training objective is to minimize the band ratio. Regular negative log-likelihood loss: in The total number of training samples, The regularization coefficient is . For the weight vector Norm squared. Training is performed using gradient descent or a quasi-Newton method, with a maximum number of iterations ranging from 200 to 2000, or when the validation set loss decreases by less than [value missing] within 20 consecutive iterations. The process stops early. To ensure the output can be used for subsequent gating and conflict determination, the prior control quantity generation module calculates the output on each grid and time slice. Forming a suitable habitat hierarchy grid and will Discretize the results into a suitable level identifier based on a threshold, which can be either 0.33 or 0.66.

[0056] Uncertainty raster To characterize the spatial and temporal instability of prior outputs, this embodiment uses the binary entropy of the predicted probability as a measure of uncertainty, and defines... in It is a binary entropy, and its value range is... ,when hour (Highest uncertainty). To ensure cross-grid comparability, for Performing global normalization yields normalized uncertainty. in and Within the entire region The minimum and maximum values, To prevent division by zero regularization constants. Write uncertainty raster The results are then discretized into uncertainty level identifiers based on a threshold, which can be either 0.4 or 0.7. The suitability level identifier and uncertainty level identifier are stored together with the grid ID, time slice ID, and policy version number, and serve as inputs for subsequent gating ticket generation and conflict type determination.

[0057] The data source for the gated ticket generation module is the evidence units already in the database and their error metrics. To ensure the calibration sequence is continuously updated, in this embodiment, each evidence unit records a final consistency score. And define the error metric as This forms a calibration loss sequence. The gated ticket generation module maintains the calibration set using a rolling time window, with a window length ranging from 7 to 90 days, or a fixed retention period for the most recent calibration. One piece of evidence unit, Use a range of 500 to 50,000; when the calibration set data is insufficient, use the minimum calibration size. If the requirement is insufficient, the ticket will not be updated; the previous version of the ticket will be used instead.

[0058] The gated ticket generation module generates a quantile interval weight vector from the calibration loss sequence and constructs a random upper bound for risk, thereby obtaining threshold constraints for prior gated tickets and convergent gated tickets. Specifically, the calibration loss sequence... Sort to get And generate quantile interval weight vector ,in and and make Satisfying the Dirichlet distribution constraint, i.e. Obtain the parameter as The Dirichlet distribution has the following probability density function: in For the Gamma function, Let be the Dirichlet parameter vector. For its first Each component. Definition when and Then the upper bound random variable of risk is defined as Repeat sampling And calculate Get the sample set , Take values ​​from 1000 to 10000; at the confidence level The empirical quantile is used as the confidence quantile threshold. The gated ticket generation module will Prior gating tickets and convergence gating tickets are written, where prior gating tickets are used to limit the triggering threshold constraints of DNA signal detection and sampling script transactions within the medium, and convergence gating tickets are used to limit the stopping threshold constraints of resampling and storage; each ticket includes a ticket number, , The applicable grid range (one or more grid ID sets or spatial polygon mappings to grid sets), validity period (start and end timestamps), and policy version number are included with the suitability level identifier and uncertainty level identifier when issued, so that the end side and the orchestration side can make consistent gating decisions.

[0059] The collaborative evidence chain orchestrator includes a collaborative simulator, a multi-round payoff evaluator, and a policy updater. After receiving the media evidence event packet, the collaborative evidence chain orchestrator constructs a collaborative state vector as input for generating candidate sampling script transactions. The collaborative state vector includes at least: a spatiotemporal index, a suitability level identifier, an uncertainty level identifier, and prior gating documents. and The event package contains event availability scores, Culicoides candidate confidence scores, media classification confidence scores, DNA signal confidence scores within the media, time-series counting features, and the current policy version number and remaining sampling budget. Based on this, a collaborative evidence chain orchestrator generates a set of candidate sampling script transactions. Each sampling script transaction is a structured transaction object, specifying sampling object binding rules, a set of acquisition channels, an acquisition time window, sampling intensity parameters, termination conditions, and convergence targets. The acquisition channel set is limited to a combination of multi-view camera channels and multi-sensor channels; the acquisition time window is 5 to 60 minutes; the sampling intensity parameters include at least camera frame rate, sensor sampling frequency, and data upload frequency; the termination conditions include at least primitive chain integrity reaching a target threshold, conflict type satisfying a reduction condition, and sampling budget reaching an upper limit; and the convergence target includes at least the final error metric not exceeding the convergence gating ticket stopping threshold. Furthermore, the primitive chain completeness is not lower than the target threshold.

[0060] A collaborative simulator is used to perform forward multi-round simulations of candidate sampling script transactions to generate evidence chain trajectories. To ensure feasibility, this embodiment uses discrete rounds. The evidence chain evolution is represented by a sampling script transaction execution in each round, which generates a summary output of the site state parameter stream. The collaborative simulator predicts the conflict type evolution in the next round through the transition table obtained from the statistics of historical evidence units. The transition table records... , indicating the first Round-based sampling script transactions And the current conflict type is At that time, the first Wheel conflict type is The transition probability is calculated; simultaneously, a regression model is used to predict the primitive chain completeness and error metric of the next round, with the prediction formula being: in For the first Wheel primitive chain completeness, For the first Wheel error measurement value For sampling script transactions In conflict type The incremental contribution to the completeness of the primitive chain. This represents the decay rate of the sampling intensity with respect to the error metric. The collaborative simulator generates the trajectory based on this. The trajectory includes at least a sequence. ,in For the first Round sampling script transactions, For the first Wheel error measurement value For the first Wheel collision type encoding, For the first Wheel primitive chain completeness, For the first Round sampling cost. The sampling cost is calculated in a calculable manner, and in this embodiment includes at least a weighted sum of data volume and energy consumption: Let the first round of sampling cost be... The amount of data in each video channel is (Unit: MB), sensor channel data volume is The estimated energy consumption at the end side is (unit: Wh) in Preset weights, value range to , used to standardize units.

[0061] The multi-round payoff evaluator calculates the convergence efficiency and sampling cost of the evidence chain trajectory and generates multi-round payoff values. The convergence efficiency reflects the number of rounds and time required to reach the convergence-gated ticket stopping threshold. Assume that the trajectory first satisfies... and The number of rounds is The convergence efficiency is defined as follows: The sampling cost is defined as follows: Multi-round payout is defined as... in To converge efficiency weights, For sampling cost weights, To achieve the required number of rounds for convergence, The minimum threshold for primitive chain completeness. The policy updater updates the sampling script transaction generation strategy based on multi-round payoff values: for each collaborative state, candidate scripts are repeatedly generated and simulated to obtain a multi-round payoff value set, which is then selected. The largest sampling script transaction is used as the current issued script, and the parameter range of this script transaction (such as the collection time window, upper / lower limit of sampling intensity parameters, and channel combination selection probability) is written into the sampling script transaction parameter version in the strategy version package. This drives the generation of candidate scripts under similar conditions to favor higher-yield configurations, thereby reducing the number of resampling times and shortening the convergence path for subsequent sampling script transactions while satisfying the convergence gate ticket constraints.

[0062] During on-site execution, the collaborative evidence chain orchestrator distributes the selected sampling script to multi-view camera equipment and multi-sensor equipment. The primitive recognition and inference engine extracts primitive vectors and primitive chain completeness from the station state parameter stream, generates cattle state micro-drift feature vectors and state consistency deviation and its trend, and performs consistency verification with the media evidence event package to output a consistency score and conflict type. To meet the supplementary disclosure purpose of this embodiment, the conflict type determination in this embodiment adopts a joint determination of multi-source gating and gap list: the suitability level identifier and uncertainty level identifier are used as prior conditions, the confidence level and time-series counting features of DNA signals in the media are used as the strength of evidence on the media side, and the primitive chain completeness, cattle state micro-drift feature vectors, state consistency deviation and its trend are used as the strength of evidence on the station side, and the supplementary evidence path is determined by combining the primitive missing item list; when the uncertainty level identifier is high and the strength of evidence on the media side is weak, it is determined to be a priori high uncertainty conflict and a supplementary evidence path is generated mainly to increase the density of trap deployment, extend the media observation time window, and expand the coverage of ventilation openings; when the media is in When the DNA signal confidence level is high but the site-side evidence strength is weak, it is determined to be a conflict of strong media and weak site, and a supplementary evidence path is generated, mainly to increase the coverage of multi-view camera channels, improve the acquisition time window and sampling intensity parameters, and improve the completeness of the primitive chain. When the site-side evidence strength is high but the media-side evidence strength is weak, it is determined to be a conflict of weak media and strong site, and a supplementary evidence path is generated, mainly to lower the count lower limit or candidate confidence threshold in the event gating conditions, adjust the DNA signal detection triggering conditions in the media, and improve the sampling intensity at the media end. When the primitive missing item list is not empty and the primitive chain completeness is lower than the target threshold, it is determined to be a chain incomplete conflict, and a supplementary evidence path is generated, mainly to fill in the missing channels or extend the acquisition time window. The collaborative evidence chain orchestrator converts the supplementary evidence path into a resampling script transaction according to the conflict type and triggers resampling until the stopping threshold corresponding to the convergence gating ticket is met and a convergence proof marker is generated.

[0063] The closed-loop update and deployment controller performs versioned reconfiguration and forms a policy version package after each convergence proof marker is generated. This policy version package is then mapped to the grid-level applicable scope to update the next round of prior gating ticket generation rules, event gating conditions, and incremental training sample selection rules for the species distribution model. To ensure the mapping is feasible, this embodiment maintains a statistical count of conflict types and an average error metric for each grid: for each grid... Cumulative conflict type count within a statistical period (e.g., 7 days) with average error This is used to generate the mesh-level reconfiguration strength, calculated as follows: in For grid The reconfiguration strength, and These are the conflict count weights and the average error weights, respectively. For grid Conflict types The count, For grid The average error metric, This represents the maximum value of the average error across the entire grid. To prevent division by zero regularization constants. Reconfiguration strength. A larger value indicates that the grid requires more stringent parameter adjustments. The update of the trap deployment density version uses the grid-level uncertainty level indicator and collision count as inputs: when the uncertainty level indicator is high and... When the threshold is exceeded (e.g., 10), the trap deployment density level of the grid is increased; the gated ticket confidence quantile threshold version is updated globally or regionally. If the overall error is too large, the confidence level should be increased, using the distribution as the input. or improve To obtain a more conservative threshold, the threshold should be reduced if the overall error is too small and the sampling cost is too high. Or adjust To reduce costs, the updates to the wing-vibration fingerprint purification threshold version and the DNA signal detection trigger condition version within the medium are adjusted based on the proportion of weak-field conflicts and strong-field conflicts in the medium, as well as the level of false alarms at the edge. and The update of the sampling script transaction parameter version is output by the policy updater and is consistent with the sampling budget constraints. The closed-loop update and deployment controller writes the above update results into the policy version package and publishes it as a new policy version number, so that the next round of media entry events will perform ticket generation, event gating, script transaction generation and resampling convergence under the new version, thereby realizing the continuous closed-loop operation of prior modeling, ticket calibration, collaborative simulation and version reconfiguration.

[0064] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be modified within the scope of the concept described herein by means of the above teachings or the technology or knowledge in related fields.

Claims

1. A bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification, characterized in that, include: A spatial evidence chain sample library is used to store evidence units, which include a spatiotemporal index, a media evidence digest, a station evidence digest, a conflict type sequence, a convergence proof flag, a policy version number, and an event hash fingerprint. The convergence proof flag is used to indicate that the evidence unit meets the stopping threshold defined by the convergence gating ticket. The prior control quantity generation module is used to train a species distribution model with Culicopters as the target species based on spatiotemporal index and environmental variable raster data, and output the suitability level raster and uncertainty raster. The gated ticket generation module is used to form a calibration loss sequence based on the error metric and generate a quantile interval weight vector to obtain a risk upper bound random quantity, and take the confidence quantile threshold of the risk upper bound random quantity to generate a priori gated tickets and convergent gated tickets. A media sentinel terminal, deployed at the ventilation openings of aquaculture farms, includes an optical checkpoint detection unit for the ventilation openings. The optical checkpoint detection unit outputs an event availability score and a Culicoides midge candidate confidence score, which are then bound to a fan operating condition tag to form a media entry event stream. The media sentinel terminal performs wing vibration fingerprint purification and media classification on the media entry event stream to output a media classification confidence score. When the prior gating ticket allows the event and the event availability score, Culicoides midge candidate confidence score, and media classification confidence score meet the event gating conditions, it performs DNA signal detection within the media to output a DNA signal confidence score within the media, thereby generating a media evidence event package. A collaborative evidence chain orchestrator is used to receive the media evidence event packet and generate sampling script transactions based on prior gated tickets, event availability scores, Coxsmid candidate confidence and convergence gated tickets. The collaborative evidence chain orchestrator performs forward multi-round collaborative simulation on the candidate sampling script transactions to generate evidence chain trajectories and calculates multi-round payoff values ​​based on the evidence chain trajectories to update the sampling script transaction generation strategy. The primitive recognition and reasoning engine is used to drive the multi-view camera and multi-sensor devices to output the station state parameter stream according to the sampling script transaction, and extract primitive vectors and primitive chain completeness from the station state parameter stream, generate cattle state micro-drift feature vectors and state consistency deviation and its trend, and perform consistency verification with the media evidence event package to output consistency score and conflict type, and generate resampling script transaction based on the conflict type to trigger resampling until the stop threshold corresponding to the convergence gate ticket is met and generate the convergence proof mark; The closed-loop update and deployment controller is used to write the corresponding evidence unit into the spatial evidence chain sample library after the convergence proof marker is generated, and to reconfigure the wing vibration fingerprint purification threshold, the DNA signal detection trigger condition in the medium, the sampling script transaction parameters, and the trap deployment density based on the conflict type sequence and error structure to form a strategy version package, which is used for the generation of prior gating tickets and the update of event gating conditions for the next round of medium entry events.

2. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The term "Culicoides" includes *Culicoides oxystoma*, *Culicoides homotomus*, *Culicoides actoni*, *Culicoides tainanus*, *Culicoides imicola*, *Culicoides fulvus*, *Culicoides brevitarsis*, and *Culicoides brevitarsis*. (jacobsoni), the media evidence summary includes the event availability score, the confidence level of the Culicoides candidate, the confidence level of the media classification, the confidence level of the DNA signal in the media, the time-series counting features, and the wind turbine operating condition label. The site evidence summary includes the statistical description of the primitive vector, the primitive chain integrity identifier, and the statistical description of the cattle herd state micro-drift feature vector, as well as the state consistency deviation and its trend quantity identifier. The event hash fingerprint is jointly generated by the spatiotemporal index, the media evidence summary, the site evidence summary, the conflict type sequence, and the policy version number for evidence deduplication, tracing, and database verification.

3. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The species distribution model is a maximum entropy species distribution model. The prior control quantity generation module performs resolution unification and spatial purification processing on the environmental variable raster data to form a training raster package. Based on the training raster package and the spatiotemporal index, it generates occurrence point samples to train the maximum entropy species distribution model and outputs the suitability level raster and the uncertainty raster.

4. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The ventilation opening optical checkpoint detection unit includes a laser light curtain emitter spanning the cross-section of the ventilation opening and a photoelectric receiving array. Based on the scattering waveform output by the photoelectric receiving array, it generates the event availability score and the Culex midge candidate confidence score. The fan condition label includes a ventilation opening identifier, fan speed parameters, and air volume setting parameters. These are bound to the event availability score and the Culex midge candidate confidence score within the same time window to form the media entry event stream.

5. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The wing-vibration fingerprint cleanup includes extracting wing-vibration fingerprint features from the entry signal of the media entry event stream and calculating a cleanup score, classifying the wing-vibration fingerprint features into media to generate the media classification confidence score, and prohibiting the corresponding media entry event from participating in the event gating condition determination when the cleanup score is lower than the cleanup threshold, and marking the corresponding media entry event as a low-confidence event that cannot be used to trigger sampling script transactions.

6. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The detection of DNA signals within the media is initiated when the prior gating ticket allows it and the event gating condition is met. The event gating condition is jointly defined by the event availability score, the confidence level of the Culicoides candidate, the confidence level of the media classification, and the time-series counting features. The media evidence event package includes a detection time window identifier and a detection process version identifier for association with the policy version number and for time-series alignment in the consistency check.

7. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The gated ticket generation module sets the quantile interval weight vector as a weight sampling vector that satisfies the Dirichlet distribution constraint, and generates a sample set of risk upper bound random quantities based on multiple weight sampling vectors and the sorted calibration loss sequence. The confidence quantile threshold of the sample set forms the trigger threshold constraint of the prior gated ticket and the stop threshold constraint of the convergence gated ticket, which are respectively used to limit the triggering, resampling stop and storage of DNA signal detection in the media.

8. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The collaborative evidence chain orchestrator includes a collaborative simulator, a multi-round payoff evaluator, and a policy updater. The collaborative simulator performs forward multi-round simulations of candidate sampling script transactions based on the media evidence event package, the prior gating ticket, the uncertainty grid, the herd state micro-drift feature vector, and the state consistency deviation and its trend value to generate an evidence chain trajectory. The multi-round payoff evaluator calculates the convergence efficiency and sampling cost of the evidence chain trajectory and generates a multi-round payoff value, wherein the multi-round payoff value is a weighted combination of the convergence efficiency and sampling cost. The policy updater updates the sampling script transaction generation strategy based on the multi-round payoff value so that subsequent sampling script transactions reduce the number of resampling times and shorten the convergence path while satisfying the convergence gating ticket constraint.

9. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The sampling script transaction limits the collection channel set to a combination of multi-view camera channels and multi-sensor channels, and limits the termination conditions to primitive chain integrity reaching the target threshold, conflict type satisfying the order reduction condition, and sampling budget reaching the upper limit condition. The primitive recognition and inference engine extracts the herd state micro-drift feature vector based on the station state parameter flow within the collection time window, aligns the herd state micro-drift feature vector with the historical baseline feature vector to calculate the state consistency deviation, and performs sliding time window trend analysis on the state consistency deviation to generate the trend quantity. The trend quantity is used as the input constraint for generating the resampling script transaction. The herd state micro-drift feature vector includes at least activity spectrum features, trajectory dwell hotspot features, gait rhythm features, feeding event frequency and duration features, drinking event frequency and duration features, posture stability features, and herd structure features.

10. The bluetongue disease monitoring and early warning system based on vector monitoring and multi-source data verification as described in claim 1, characterized in that, The conflict type is determined by the adaptability level and uncertainty level identifiers corresponding to the prior gating tickets, the confidence and temporal count features of DNA signals within the media, the completeness of the primitive chain, the herd state micro-drift feature vector, the state consistency deviation, the trend quantity, and the primitive missing item list. Specifically, a high prior uncertainty conflict is defined as follows: a high uncertainty level identifier and weak media-side evidence strength; a strong media-weak conflict as the confidence level of DNA signals within the media and weak field-side evidence strength; a weak media-strong conflict as the field-side evidence strength and weak media-side evidence strength; and a conflict where the primitive missing item list is not empty and the primitive chain completeness is lower than a certain threshold. When the target threshold is reached, it is determined to be a chain incomplete conflict. The conflict type is used to generate supplementary evidence paths for the resampling script transaction. The strategy version package includes a wing vibration fingerprint purification threshold version, a DNA signal detection trigger condition version in the vector body version, a gated ticket confidence quantile threshold version, a sampling script transaction parameter version, and a trap deployment density version, and is mapped to a grid-level applicable range. The grid-level applicable range is determined by the grid-level reconfiguration intensity, which is calculated by the conflict type sequence statistical count and the average error metric, to update the prior gated ticket generation rules, the event gating conditions, and the incremental training sample selection rules of the species distribution model.

Images