Intelligent bullfrog disease identification system

Abstract

The invention discloses an intelligent bullfrog disease recognition system, and relates to the technical field of intelligent culture and aquatic product informatization. The technical problems that a traditional manual observation mode is low in efficiency, and early risk early warning and accurate positioning cannot be carried out before macroscopic abnormality occurs to bullfrog individuals are solved. According to the invention, the pre-symptom early warning module monitors water body metabolism markers in real time and constructs a dynamic health baseline to generate early warning; the subclinical screening module is used for positioning suspected individuals with abnormal spectral characteristics by using hyperspectral imaging non-contact scanning groups; the pathogen detection module is used for driving automatic sampling, carrying out on-site rapid molecular detection and analyzing specific pathogen gene segments; and the system control and decision module fuses multi-source information, performs data association analysis based on a disease knowledge graph, and finally generates a bullfrog biosafety risk multi-modal data fusion analysis report, thereby realizing whole-course automatic and intelligent monitoring and early warning of biosafety risks in the bullfrog breeding process.

Description

Technical Field

[0001] This invention belongs to the field of smart aquaculture and aquatic information technology, specifically an intelligent system for identifying bullfrog diseases. Background Technology

[0002] Bullfrogs, as an aquaculture species with significant economic value, have seen large-scale, intensive farming become an important part of the agricultural industry. However, disease is one of the key risk factors restricting the healthy development of the industry during the farming process. Traditional disease monitoring and control mainly rely on the experience and observation of farmers, and intervention is usually only carried out after individuals show obvious clinical symptoms (such as skin lesions and abnormal behavior). By this time, the disease has often spread to the entire population, resulting in high treatment costs, increased mortality, and decreased farming efficiency. Therefore, achieving early risk monitoring, precise location, and rapid analysis is of great significance for improving the disease prevention and control capabilities of bullfrog farming, ensuring farming safety, and promoting the sustainable development of the industry.

[0003] In recent years, with the rapid development of technologies such as the Internet of Things, spectral imaging, molecular detection, and artificial intelligence, their application in aquaculture health monitoring has become a research hotspot. By monitoring changes in the aquaculture environment and the organism's own indicators, it is hoped that early signals of disease occurrence can be captured before symptoms appear. In particular, by analyzing specific metabolic markers in water bodies and combining them with non-contact hyperspectral imaging technology, multi-scale health status assessments can be achieved from the population level to the individual level. The introduction of on-site rapid molecular detection technology provides a reliable technical means for further pathogen analysis. Against this backdrop, developing an intelligent monitoring system that can integrate multi-source information and cover the entire chain of "early warning-screening-analysis-decision making" has become an important technical requirement for promoting the intelligent, precise, and forward-looking development of disease prevention and control in bullfrog farming.

[0004] The following problems exist in the existing technology: Traditional assessments of bullfrog health status rely heavily on visual observation or examinations after clinical symptoms appear, which cannot detect diseases in their incubation or subclinical stages, leading to delays in prevention and control and making it easier for diseases to spread within the population. Even if abnormalities in a population are detected through water bodies or environmental indicators using traditional methods, there is still a lack of efficient technical means to quickly and accurately screen individuals in the early stages of disease within the population, which affects the implementation of targeted interventions. Current pathogen analysis typically requires manual sampling and laboratory testing, which involves many steps and a long cycle, failing to meet the needs of breeding sites for rapid acquisition of pathogen information and delaying the timing of prevention and control response. Metabolic data, imaging data, pathogen data, and other information from multiple sources during the breeding process often exist independently, lacking effective integration and correlation analysis, making it difficult to form a systematic and intelligent monitoring and control decision support. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent bullfrog disease identification system to solve the above-mentioned technical problem.

[0006] The first aspect of this invention provides an intelligent system for identifying bullfrog diseases, comprising the following modules: Pre-symptom warning module: Monitors the concentrations of at least two specific physiological and pathological metabolic markers released into the water by bullfrogs in the breeding area, and simultaneously collects environmental parameter data of the breeding water; Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a generalized autoregressive conditional heteroscedasticity model is used to construct a dynamic baseline of the healthy population; By analyzing the statistical deviation of metabolic marker concentrations from the dynamic baseline of the healthy population, a population metabolic warning signal is generated before individuals show visible symptoms; Subclinical screening module: In response to the population metabolic early warning signal, the hyperspectral imaging device is activated to perform non-contact scanning of the bullfrog population in the corresponding breeding area to obtain hyperspectral image cube data; by analyzing and comparing the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, the target individuals showing abnormal early pathological spectral characteristics are located from the population, and a list of spatial coordinates of the target individuals is output. Pathogen detection module: Based on the target individual spatial coordinate list, it drives the automated sampling device to perform guided automated sampling of the target individual; using the on-site rapid molecular detection device, it performs molecular detection on the biological samples collected from the target individual for preset bullfrog pathogen-specific gene fragments, and outputs a pathogen detection report; System Control and Decision Module: By querying the system's built-in disease knowledge graph, the module performs fusion analysis and correlation reasoning on population metabolic early warning signals, target individual spatial coordinate lists, and pathogen detection reports to generate a multimodal data fusion analysis report on bullfrog biosafety risks.

[0007] Preferably, in the pre-symptom warning module, the concentrations of at least two specific physiological and pathological metabolic markers released into the water by the bullfrog's metabolism within the breeding area are monitored, and environmental parameter data of the breeding water are collected simultaneously. Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a dynamic baseline of the healthy population is constructed using a generalized autoregressive conditional heteroscedasticity model, including the following steps: The farm is divided into at least two breeding areas. Monitoring devices are deployed in each breeding area, and each breeding area is assigned a unique identifier, which is then linked to the data collected by the corresponding monitoring device. Time-series data of the concentrations of at least two preset specific physiological and pathological metabolic markers released by bullfrog metabolism in the water of the breeding area are continuously collected, and time-series data of water environmental parameters in the breeding area are collected simultaneously. The time-series data of metabolic marker concentrations and environmental parameters are preprocessed. Using preprocessed environmental parameter time-series data as covariates, a generalized autoregressive conditional heteroscedasticity model was employed to jointly model the preprocessed metabolic biomarker concentration time-series data. Through iterative calculations of the model, the dynamic health baseline values ​​of each metabolic biomarker under corresponding environmental conditions were determined. and its dynamic standard deviation The dynamic standard deviation The iterative calculation process is as follows: in, , indicating the first Metabolic markers, The total number of metabolic biomarkers monitored and ≥2; Indicates the first Index of each sampling time; for The difference between time-series data of metabolic biomarker concentrations at a given time and their corresponding dynamic healthy baseline concentration. Indicates the first Metabolic markers in Concentration time-series data after time-series preprocessing; , , For parameters; Indicates the first Environmental parameters in Preprocessed data at any given time, , The number of environmental parameter types; It's about environmental parameters. Preset functions; Environmental stress For the The influence coefficient of volatility of various metabolic markers; The dynamic health concentration baseline With dynamic standard deviation Together they constitute the dynamic baseline of the healthy population.

[0008] Preferably, in the pre-symptom warning module, by analyzing the statistical deviation of metabolic marker concentrations relative to the dynamic baseline of a healthy population, a population metabolic warning signal is generated before an individual develops visible symptoms, including the following steps: Dynamic health concentration baseline based on dynamic baseline of healthy population With dynamic standard deviation For the first Metabolic markers, calculating their role in Standardized instantaneous deviation at time ,in, for Time-series data of metabolic biomarker concentrations after preprocessing. It is a tiny positive number; Based on the standardized instantaneous deviation, a dynamic model characterizing the correlation between different metabolic biomarkers is constructed and updated online, and the multi-scale Mahalanobis distance is calculated as a comprehensive statistical deviation index. : in, For delay The standardized instantaneous deviation vector of the step. For the delay extracted from the dynamic model The correlation matrix at step time, For exponentially decaying weights, This represents the maximum delay steps. Comprehensive statistical deviation index The input includes a Hidden Markov Model (HMM) encompassing healthy, subclinical, and clinical states, which is decoded to obtain a population health state sequence. When the decoded state sequence indicates that the population health state has been consistently in a subclinical state over multiple consecutive sampling periods, a warning confidence level is calculated based on Bayesian inference. ; If and only if the population health status remains subclinical, with a warning confidence level Exceeding the preset threshold Furthermore, when the key environmental parameter values ​​of the aquaculture area corresponding to the warning signal exceed the corresponding preset health threshold range, a group metabolism warning signal is generated, which encapsulates the identification information of the aquaculture area, the health status of the group, the confidence level, and the time series data of the environmental parameters.

[0009] Preferably, in the subclinical screening module, in response to the population metabolic early warning signal, a hyperspectral imaging device is activated to perform a non-contact scan of the bullfrog population in the corresponding breeding area to acquire hyperspectral image cube data. The process involves analyzing and comparing the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, including the following steps: Upon receiving the population metabolism early warning signal encapsulated with aquaculture area identification information, the hyperspectral imaging device is activated to perform a non-contact scan of the bullfrog population within the corresponding aquaculture area, acquiring hyperspectral image cube data with dimensions of [missing information]. ,in This represents the total number of pixels in the height direction of the image. This represents the total number of pixels in the image width direction. The total number of spectral bands; define the two-dimensional spatial coordinates of each pixel location in the image as its row index in the image. With column index ,in ; The system calls upon a pre-generated health spectral intrinsic space and its associated reference statistics based on a set of historical healthy bullfrog hyperspectral image data. The health spectral intrinsic space is extracted from the historical healthy bullfrog hyperspectral image data using tensor decomposition. The reference statistics include the mean vector and covariance matrix calculated from the projection coefficients of all pixels in the historical healthy bullfrog hyperspectral image data, as well as the variance calculated based on the reconstruction error of all pixels. The health spectral intrinsic space and reference statistics are stored in a health spectral database. For pixel locations in hyperspectral image cube data The corresponding pixel spectral vector Calculated through projection operation Projection coefficient in the eigenspace of the health spectrum The reconstruction error was calculated. ; Calculate projection coefficients The Mahalanobis distance relative to the mean vector and covariance matrix in the reference statistic ,in and These are the mean vector and covariance matrix of the projection coefficients calculated from historical health spectral data, respectively. By fusing reconstruction error and Mahalanobis distance, the position of each pixel in the hyperspectral image cube data is generated. Comprehensive spectral anomaly index ,in For the reconstruction error variance in the reference statistic, For degrees of freedom The 95th percentile of the chi-square distribution, These are the weighting coefficients.

[0010] Preferably, in the subclinical screening module, locating target individuals exhibiting abnormal early pathological spectral characteristics from the population and outputting a list of spatial coordinates of the target individuals includes the following steps: Based on the pixel position in the hyperspectral image cube data Comprehensive spectral anomaly index Multiple candidate connected regions are obtained through image segmentation and connected component analysis, and those with areas exceeding a preset threshold are selected. The candidate connected regions are further segmented to obtain sub-connected regions; all connected regions, including the unsegmented candidate connected regions and the segmented sub-connected regions, are collectively formed into a set of connected regions; each connected region is processed as an independent target individual; the connected region refers to a continuous region composed of spatially adjacent pixels that are all marked as suspected anomalies; For each connected region in the set of connected regions Calculate its confidence score The calculation formula is as follows: in, Connected region Total number of pixels included. Connected region The actual area occupied in the resulting image. The standard individual imaging area value is obtained by statistically analyzing historical hyperspectral image data of healthy bullfrogs. Confidence score Compare with the preset confidence threshold to filter out all For connected regions with a confidence threshold, each selected connected region is identified as a target individual exhibiting abnormal early pathological spectral characteristics. The geometric mean of all pixel coordinates within the connected region corresponding to each target individual is calculated and used as the spatial coordinates of that target individual. The spatial coordinates of all target individuals exhibiting abnormal early pathological spectral characteristics are summarized, and a list of target individual spatial coordinates is generated and output.

[0011] Preferably, in the pathogen detection module, the guided automated sampling device is driven to perform automated sampling of the target individual based on the target individual spatial coordinate list, including the following steps: Based on the target individual spatial coordinate list, obtain Spatial coordinates of the target individual to be sampled Simultaneously, obtain the spatial coordinates of non-target individuals, denoted as... ,common One, of which The total number of non-target individuals; With the goal of minimizing the total sampling path length and reducing interference with non-target individuals, a motion path traversing all target individuals is planned for the automated sampling device; this planning is achieved by solving the following composite objective function. Minimization problem implementation: in, To characterize the coordinate sequence of the spatial path points that the automated sampling device is to traverse; To estimate the total sampling time; For the sampling device at time Spatial location; For the first The location of a non-target individual ; , , For weights and scale parameters; The minimization problem is solved using a cooperative adaptive particle swarm optimization algorithm to obtain the optimal path coordinate sequence. That is, an optimal motion trajectory consisting of a series of spatial path points; based on the optimal motion trajectory, the automated sampling device is driven to move sequentially to the position of each target individual and complete automated sampling.

[0012] Preferably, the pathogen detection module utilizes a rapid on-site molecular detection device to perform molecular detection on biological samples collected from the target individual targeting a preset bullfrog pathogen-specific gene fragment, and outputs a pathogen detection report, including the following steps: Receive and process biological samples collected from the target individual by the automated sampling device; inject the biological samples into a pre-set container. In a centrifugal microfluidic chip containing a primer set specific to a bullfrog pathogen, the biological sample is distributed to each independent reaction chamber during chip rotation; the chip contains... A separate reaction chamber, the first The reaction chamber is pre-set to the first A specific primer set for one pathogen, used for the specific detection of the first pathogen. Pathogens, ; Parallel isothermal nucleic acid amplification reactions were performed on biological samples allocated to each reaction chamber under isothermal conditions, where, corresponding to the first... The first pathogen Each reaction chamber has an amplification reaction kinetics described by a modified Logistic model: in, For the first reaction chamber The concentration of pathogen-specific double-stranded DNA products, , , The first The maximum rate of amplification reaction of a pathogen, its saturation concentration, and its degradation rate constant; Based on the modified Logistic kinetic model, the fluorescence signals of each reaction chamber acquired in real time are analyzed to determine the... The characteristic time at which the amplification curve of each reaction chamber reaches the preset fluorescence threshold. and its confidence interval; Based on the characteristic time and its confidence interval and preset effective signal reference time The comparison generates a target for the first. Detection data of pathogen-specific gene fragments; if The upper limit of the confidence interval is earlier than Then, the estimated initial template nucleic acid concentration is calculated based on the standard curve. : ,in, and The coefficients are those calibrated through preliminary experiments using standard samples; The detection data of all L pathogen-specific gene fragments are summarized, including characteristic time, confidence interval and corresponding initial template nucleic acid concentration estimates, to generate a structured pathogen detection report.

[0013] Preferably, the system control and decision-making module includes the following steps: The system receives a population metabolic early warning signal from the pre-symptom early warning module, a list of target individual spatial coordinates from the subclinical screening module, and a pathogen detection report from the pathogen detection module; it extracts the breeding area identifier, environmental parameters, and comprehensive deviation index of metabolic markers from the population metabolic early warning signal, extracts the spatial distribution information of the target individuals from the list of target individual spatial coordinates, and extracts the pathogen type, detection data, and corresponding initial template nucleic acid concentration estimate from the pathogen detection report. The system queries a built-in disease knowledge graph, which includes five types of nodes: disease type, pathogen, clinical symptoms, metabolic markers, and environmental conditions, as well as edges representing the causal, correlation, susceptibility, and manifestation relationships among them. The extracted metabolic markers are combined with deviation indices, initial template nucleic acid concentration estimates, environmental parameters, and spatial distribution information of target individuals, and matched with the corresponding nodes and relationships in the graph. Through graph query and data association algorithms, the system outputs the disease knowledge graph subgraphs identified by the matching operation and the corresponding data association weight information for each subgraph. Based on the output disease knowledge graph subgraph, data association weight information, and pre-stored treatment reference information in the disease knowledge graph, a structured bullfrog biosafety risk multimodal data fusion analysis report is generated.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention, by continuously monitoring water metabolic markers and constructing a dynamic health baseline, can identify metabolic abnormalities in bullfrog populations before visible symptoms appear, significantly advancing the warning time, providing a crucial window for early intervention, and effectively curbing the spread of diseases. This invention utilizes hyperspectral imaging technology and comparison with a health spectral database to non-contactly and automatically identify individuals with early pathological spectral characteristics from a population and output their spatial coordinates. This invention integrates an automated sampling device with a rapid on-site molecular detection device to achieve full automation from target localization to pathogen analysis, significantly shortening detection time, improving analysis efficiency, and supporting timely risk assessment. This invention utilizes a built-in disease knowledge graph to perform fusion analysis and correlation reasoning on metabolic early warning, individual location, and pathogen detection reports, outputting a structured multimodal data fusion analysis report, providing farmers with a scientific, systematic, and operable basis for assessing biosafety risks. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the module flow of the present invention. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 This invention is an intelligent system for identifying bullfrog diseases, comprising the following modules: Pre-symptom warning module: Monitors the concentrations of at least two specific physiological and pathological metabolic markers released into the water by bullfrogs in the breeding area, and simultaneously collects environmental parameter data of the breeding water; Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a generalized autoregressive conditional heteroscedasticity model is used to construct a dynamic baseline of the healthy population; By analyzing the statistical deviation of metabolic marker concentrations from the dynamic baseline of the healthy population, a population metabolic warning signal is generated before individuals show visible symptoms; Subclinical screening module: In response to the population metabolic early warning signal, the hyperspectral imaging device is activated to perform non-contact scanning of the bullfrog population in the corresponding breeding area to obtain hyperspectral image cube data; by analyzing and comparing the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, the target individuals showing abnormal early pathological spectral characteristics are located from the population, and a list of spatial coordinates of the target individuals is output. Pathogen detection module: Based on the target individual spatial coordinate list, it drives the automated sampling device to perform guided automated sampling of the target individual; using the on-site rapid molecular detection device, it performs molecular detection on the biological samples collected from the target individual for preset bullfrog pathogen-specific gene fragments, and outputs a pathogen detection report; System Control and Decision Module: By querying the system's built-in disease knowledge graph, the module performs fusion analysis and correlation reasoning on population metabolic early warning signals, target individual spatial coordinate lists, and pathogen detection reports to generate a multimodal data fusion analysis report on bullfrog biosafety risks.

[0018] Specifically, the pre-symptom warning module first continuously monitors the breeding area, collecting the concentrations of at least two specific physiological and pathological metabolic markers released by bullfrog metabolism in the water, and simultaneously recording environmental parameters such as water temperature and pH. Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a generalized autoregressive conditional heteroscedasticity model is used to construct a dynamic baseline for the healthy bullfrog population, and statistical methods are used to analyze the deviation between the real-time monitoring data and this baseline. Once a sustained abnormal deviation in metabolic markers is detected, i.e., before individual bullfrogs show visible symptoms, a population metabolic warning signal is generated, indicating a potential health risk in the area. After the warning signal is generated, the subclinical screening module is immediately activated. This module controls a hyperspectral imaging device to perform non-contact scanning of the bullfrog population within the warning area, acquiring hyperspectral image data containing spectral information. By comparing and analyzing the spectral features of each pixel with a pre-established healthy bullfrog spectral database, areas with abnormal spectral features are identified, thereby accurately locating suspected abnormal individuals exhibiting early pathological characteristics from the population, and outputting a list of the coordinates of these target individuals. Next, the pathogen detection module, based on the target individual's location coordinates, drives the automated sampling device to the designated location to automatically and guidedly collect biological samples from the target bullfrog individual. The collected samples are then sent to a rapid molecular detection device on-site. This device uses a microfluidic chip pre-loaded with various bullfrog pathogen-specific primers to perform parallel isothermal amplification of nucleic acids on the samples. By analyzing fluorescence signals, it analyzes the presence and concentration estimates of specific pathogen gene fragments in the samples, ultimately outputting a structured pathogen detection report. Finally, the system control and decision-making module, as the core hub, receives and integrates all information output from the first three modules, including early warning signals, abnormal individual coordinates, and pathogen detection reports. By querying the system's built-in disease knowledge graph, which covers the relationships between diseases, pathogens, symptoms, metabolites, and environmental factors, it performs multidimensional information fusion analysis and correlation reasoning. Based on the reasoning results, it automatically generates a structured bullfrog biosafety risk multimodal data fusion analysis report, thus completing the closed-loop process from early warning to risk analysis.

[0019] In one embodiment of the present invention, the pre-symptom warning module monitors the concentrations of at least two specific physiological and pathological metabolic markers released into the water by the bullfrog's metabolism within the breeding area, and simultaneously collects environmental parameter data of the breeding water. Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a dynamic baseline of the healthy population is constructed using a generalized autoregressive conditional heteroscedasticity model, including the following steps: The farm is divided into at least two breeding areas. Monitoring devices are deployed in each breeding area, and each breeding area is assigned a unique identifier, which is then linked to the data collected by the corresponding monitoring device. Time-series data of the concentrations of at least two preset specific physiological and pathological metabolic markers released by bullfrog metabolism in the water of the breeding area are continuously collected, and time-series data of water environmental parameters in the breeding area are collected simultaneously. The time-series data of metabolic marker concentrations and environmental parameters are preprocessed. Using preprocessed environmental parameter time-series data as covariates, a generalized autoregressive conditional heteroscedasticity model was employed to jointly model the preprocessed metabolic biomarker concentration time-series data. Through iterative calculations of the model, the dynamic health baseline values ​​of each metabolic biomarker under corresponding environmental conditions were determined. and its dynamic standard deviation The dynamic standard deviation The iterative calculation process is as follows: in, , indicating the first Metabolic markers, The total number of metabolic biomarkers monitored and ≥2; Indicates the first Index of each sampling time; for The difference between time-series data of metabolic biomarker concentrations at a given time and their corresponding dynamic healthy baseline concentration. Indicates the first Metabolic markers in Concentration time-series data after time-series preprocessing; , , For parameters; Indicates the first Environmental parameters in Preprocessed data at any given time, , The number of environmental parameter types; It's about environmental parameters. Preset functions; Environmental stress For the The influence coefficient of volatility of various metabolic markers; The dynamic health concentration baseline With dynamic standard deviation Together they constitute the dynamic baseline of the healthy population.

[0020] Specifically, the target farm is physically or logically divided into at least two independent farming areas (e.g., by pond number, farming batch, or spatial partition); each farming area is assigned a unique identifier (such as a region ID). An integrated online water quality monitoring device is deployed in each aquaculture area. This device integrates a continuous environmental parameter sensing module, an automatic metabolic marker analysis module, and a central control and communication module through a modular design. The continuous environmental parameter sensing module consists of industrial-grade sensor probes immersed in the water to continuously monitor basic indicators such as water temperature, dissolved oxygen, pH value, and ammonia nitrogen concentration. The automatic metabolic marker analysis module adopts a technical path of "timed automatic sampling - sample pretreatment - in-situ detection". Water samples are automatically collected through corrosion-resistant micro-peristaltic pumps and multi-way valves. After pretreatment by built-in filtration, metering dilution, and reagent mixing, the concentration of target metabolic markers (such as specific forms of ammonia nitrogen, steroid hormone metabolites, etc.) in the water is quantitatively determined by microfluidic chips or integrated optical detection units using colorimetry, fluorescence, or immunochromatography. All processes are coordinated by an industrial controller in the central control and communication module, and the data is encrypted and transmitted to the central processing system via wired or wireless Internet of Things protocols.

[0021] Continuously collect time-series data on metabolic biomarker concentrations and environmental parameters; the time-series data on metabolic biomarker concentrations are denoted as... ,in Representing the Metabolic markers, ≥2, Representing the Sampling times at equal or unequal intervals; environmental parameter time series data are denoted as... ,in Representing the Environmental parameters. Preprocessing was performed on the original time-series data of metabolic biomarker concentrations and environmental parameters. Preprocessing included data cleaning, noise filtering, and standardization. Data cleaning addressed outliers and missing values ​​in the original data. Noise filtering smoothed the data sequences to remove high-frequency random measurement noise. To eliminate differences in units and magnitudes between different time-series data, standardization was performed on the smoothed metabolic biomarker concentration and environmental parameter time-series data. The preprocessed data... Metabolic markers in The concentration at time is The synchronized environment parameter vector is .

[0022] A generalized autoregressive conditional heteroscedasticity (GARCH-X) model was used to construct a dynamic baseline for the healthy population. Since bullfrogs are poikilothermic animals, their metabolic rate and physiological state are significantly affected by fluctuations in environmental factors such as water temperature and dissolved oxygen. In open-air farming, conventional static baselines or time-series models struggle to effectively distinguish between physiological metabolic fluctuations caused by natural environmental disturbances (such as diurnal temperature variations and sudden weather changes) and pathological metabolic abnormalities caused by the early development of potential diseases, leading to decreased warning specificity. The GARCH-X model used in this invention dynamically incorporates environmental parameters as exogenous variables into the modeling of baseline mean and conditional variance. This aims to quantify and isolate the contribution of environmental fluctuations to the concentration and variability of metabolic markers, thereby enhancing the model's sensitivity and specificity to early, weak pathological signals and solving the technical problem of strong background noise masking early disease signals in open environments.

[0023] For the Metabolic markers, their dynamic health baseline concentrations (i.e., the desired concentration), expressed as: ,in, It is a constant term; It is the autoregressive coefficient of the metabolic biomarker itself at the p-th lag (i.e., a time lag of p sampling intervals); It is the coefficient vector of the qth lag of the environmental parameter vector; and These are the maximum lag orders for the autoregressive term and the environmental impact term, respectively; based on the system sampling frequency (e.g., once per hour) and the physiological timescale of the bullfrog's metabolic response to the environment (typically several hours to 24 hours), a reasonable range for the candidate orders is pre-defined, for example... and The values ​​range from 0 to 12 or 0 to 24. Historical health data (standardized metabolic marker concentration sequences and normalized environmental parameter sequences for healthy aquaculture cycles) are used to determine the values ​​for each group within a reasonable range. , Candidate values ​​are fitted to the GARCH-X model, and the corresponding Akaike Information Criterion (AIC) values ​​are calculated. The value that minimizes the AIC value is selected. , The combination is used as the optimal lag order.

[0024] The conditional variance is described using the GARCH(1,1) structure. The dynamic changes are observed, and environmental parameters are introduced as exogenous variables to enhance explanatory power: ,in, yes The deviation between the time-standardized concentration and its dynamic baseline is called the prediction residual; >0 is a constant term; ≥0 measures the impact of the squared residuals (ARCH term) from the previous time step on the variance at the current time step; ≥0 measures the impact of the previous time step variance (GARCH term) on the current time step variance; it is typically required that... To ensure variance stability; It's about environmental parameters. The preset function, in this embodiment, is taken as... = That is, directly using the normalized environmental parameter values ​​as exogenous inputs; Environmental stress For the The influence coefficient of the volatility of a metabolic biomarker.

[0025] Long-term historical monitoring data obtained during the healthy breeding period of bullfrog populations were used as the training set. The healthy breeding period refers to the complete breeding cycle confirmed by breeding records and staff, and verified by regular pathogen detection to be free of the target disease described in this invention. The data duration typically covers multiple breeding batches and different seasons. The training data consisted of standardized metabolic biomarker concentration sequences obtained through the same preprocessing steps. and the corresponding normalized environmental parameter sequence .

[0026] For each metabolic marker The maximum likelihood estimation (MLE) method was used to estimate all parameters of the GARCH-X model. , , , , , , Estimate; specifically, assume the predicted residuals Follows a mean of 0 and a variance of If the conditional normal distribution is followed by the GARCH-X model, then the conditional log-likelihood function (ignoring the constant term) of the GARCH-X model over the time interval {1,...,T} is: Parameter estimation is achieved by maximizing Implementation; during implementation, the parameters are initialized, including parameters describing the dynamic health concentration baseline. , , The initial values ​​are obtained by regressing the corresponding mathematical expression using ordinary least squares, and the variance equation parameters are ( , , The initial value is set to a small positive number that satisfies the stationarity constraint (e.g., ...). Exogenous influence coefficient The initial value is set to 0; to satisfy GARCH-X constraints on the positivity, non-negativity, and stationarity of parameters, the parameters are transformed during the optimization process, for example, by... Take the exponent to ensure it is positive. and A logistic transformation is employed to ensure that the parameters lie in the interval [0,1) and their sum is less than 1. Subsequently, an unconstrained optimization of the transformed parameters is performed using a quasi-Newton method (e.g., the L-BFGS-B algorithm) to maximize... The termination condition for the optimization iteration is set as follows: the absolute value of the difference between the log-likelihood values ​​of two consecutive iterations is less than 1. Or the Euclidean norm change of the parameter vector is less than .

[0027] After parameter estimation is completed, new data collected in real time will be used ( and The input is fed into the pre-trained GARCH-X model, and through iterative calculation, the current time step is output in real time. dynamic health concentration baseline value and its corresponding dynamic standard deviation For each monitored metabolic marker The GARCH-X model was independently constructed and its parameters were estimated, ultimately yielding the results for the aquaculture area at time [time value missing]. The dynamic baseline of a healthy population is composed of the dynamic baselines of all biomarkers. A set that is formed together.

[0028] In one embodiment of the present invention, the pre-symptom warning module generates a population metabolic warning signal before an individual develops visible symptoms by analyzing the statistical deviation of metabolic marker concentrations relative to the dynamic baseline of a healthy population, including the following steps: Dynamic health concentration baseline based on dynamic baseline of healthy population With dynamic standard deviation For the first Metabolic markers, calculating their role in Standardized instantaneous deviation at time ,in, for Time-series data of metabolic biomarker concentrations after preprocessing. It is a tiny positive number; Based on the standardized instantaneous deviation, a dynamic model characterizing the correlation between different metabolic biomarkers is constructed and updated online, and the multi-scale Mahalanobis distance is calculated as a comprehensive statistical deviation index. : in, For delay The standardized instantaneous deviation vector of the step. For the delay extracted from the dynamic model The correlation matrix at step time, For exponentially decaying weights, This represents the maximum delay steps. Comprehensive statistical deviation index The input includes a Hidden Markov Model (HMM) encompassing healthy, subclinical, and clinical states, which is decoded to obtain a population health state sequence. When the decoded state sequence indicates that the population health state has been consistently in a subclinical state over multiple consecutive sampling periods, a warning confidence level is calculated based on Bayesian inference. ; If and only if the population health status remains subclinical, with a warning confidence level Exceeding the preset threshold Furthermore, when the key environmental parameter values ​​of the aquaculture area corresponding to the warning signal exceed the corresponding preset health threshold range, a group metabolism warning signal is generated, which encapsulates the identification information of the aquaculture area, the health status of the group, the confidence level, and the time series data of the environmental parameters.

[0029] Specifically, based on the dynamic health concentration baseline in the dynamic baseline of the healthy population. With dynamic standard deviation For the first ( ) metabolic biomarkers, calculate their role in Standardized instantaneous deviation at time ,in, for Time-series data of metabolic biomarker concentrations after preprocessing. To prevent small positive numbers from being divided by zero (such as...) ); defined at time Below, delay step( The standardized instantaneous deviation vector is: ; To capture the time-varying relationships among metabolic biomarkers, the exponentially weighted moving average (EWMA) method was used to estimate the covariance matrix under different delays online. and correlation matrix Regarding the delay The covariance matrix of the step is updated by the following formula: ;in ∈(0,1) is the forgetting factor (e.g. =0.95), used to control the weights of historical data; initial covariance matrix It can be set as a diagonal matrix or estimated based on initial health data. Correlation matrix From the covariance matrix After standardization, the element in the p-th row and q-th column is: ;in yes The corresponding element in the middle.

[0030] Comprehensive statistical deviation index The formula used to quantify the degree of abnormality of current and recent metabolic markers deviating from the healthy baseline is as follows: ,in, The maximum delay step is set based on the bullfrog's physiological response time (e.g., several hours to 24 hours) and the system sampling frequency. For example, if sampling is performed once per hour, then set... =12; For exponentially decaying weights, defined as ,in ∈(0,1] (e.g.) =0.85), where j is the summation index variable, ranging from 0 to... ; For delay The inverse of the step correlation matrix.

[0031] We use a predefined Hidden Markov Model (HMM) that contains three hidden states: (healthy), (Subclinical) and (Clinical); its observed variable is the composite statistical deviation index calculated in real time. Model parameters include the initial state probability distribution. The state transition probability matrix A (where elements) Let represent the probability of transitioning from state i to state j and the observation probability distribution, where the observation probability distribution is specifically, assuming that in state j... (m=1,2,3) Observations Follows a gamma distribution Use historical health period (corresponding) ) and known disease occurrence periods (early corresponding Clear period corresponding )of The sequence is used as training data, and the model parameters are iteratively optimized using the Baum-Welch algorithm (an expectation-maximization algorithm) until convergence, resulting in a trained Hidden Markov Model (HMM). During online monitoring, for real-time observed sequences... Hidden state sequence decoded using the Viterbi algorithm This sequence is the hidden state sequence corresponding to the observation sequence, where each Set a time window length. (For example, 6 consecutive sampling periods), when the decoded hidden state sequence is in the most recent The patient remained in a subclinical state throughout the cycle (i.e., When [a certain condition] is met, the early warning confidence level calculation is triggered. The early warning confidence level is calculated based on Bayesian inference and reflects the degree of credibility in currently judging the group to be in a subclinical state; it is defined as [a condition] in the most recent [time period]. Observations Under these conditions, the population is in a subclinical state. The posterior probability is the confidence level: Using the trained HMM model, calculate the forward variables. Where t is the time index, and its value range is 1≤t≤k, then ,in Based on from arrive All observation data were used to calculate the result at time [time]. In state The forward probability.

[0032] Generating the final population metabolic early warning signal requires simultaneously meeting three conditions: First, the state condition, that is, the population health state decoded based on the hidden Markov model must be continuous. Within each sampling period (e.g.) =6) Persistently in a subclinical state ( First, it characterizes the persistence of metabolic abnormalities rather than transient fluctuations; second, it considers the confidence level, calculated through Bayesian inference, as the warning confidence level for a current state that is subclinical. It must exceed the preset confidence threshold. (For example =0.85), filtering out low-confidence judgments caused by model uncertainty or random noise; third, environmental conditions, that is, in the aquaculture area that triggers the warning, the real-time value or short-term trend (such as moving average) of at least one key environmental parameter (such as water temperature, dissolved oxygen, ammonia nitrogen concentration) exceeds the preset health threshold range for the area. The health threshold range is determined based on the statistical distribution of long-term historical health monitoring data (for example, taking the 5th percentile to the 95th percentile of a certain environmental parameter value during the healthy period as the range) or industry-recognized aquaculture standards (for example, the healthy water temperature range can be set to 20℃-28℃). A structured population metabolism warning signal will be generated only when all three conditions are met. This signal encapsulates the aquaculture area identification information (ID), population health status, warning confidence level, and synchronized environmental parameter time series data.

[0033] In one embodiment of the present invention, the subclinical screening module, in response to the population metabolic early warning signal, activates a hyperspectral imaging device to perform a non-contact scan of the bullfrog population in the corresponding breeding area to acquire hyperspectral image cube data. The module then analyzes and compares the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, including the following steps: Upon receiving the population metabolism early warning signal encapsulated with aquaculture area identification information, the hyperspectral imaging device is activated to perform a non-contact scan of the bullfrog population within the corresponding aquaculture area, acquiring hyperspectral image cube data with dimensions of [missing information]. ,in This represents the total number of pixels in the height direction of the image. This represents the total number of pixels in the image width direction. The total number of spectral bands; define the two-dimensional spatial coordinates of each pixel location in the image as its row index in the image. With column index ,in ; The system calls upon a pre-generated health spectral intrinsic space and its associated reference statistics based on a set of historical healthy bullfrog hyperspectral image data. The health spectral intrinsic space is extracted from the historical healthy bullfrog hyperspectral image data using tensor decomposition. The reference statistics include the mean vector and covariance matrix calculated from the projection coefficients of all pixels in the historical healthy bullfrog hyperspectral image data, as well as the variance calculated based on the reconstruction error of all pixels. The health spectral intrinsic space and reference statistics are stored in a health spectral database. For pixel locations in hyperspectral image cube data The corresponding pixel spectral vector Calculated through projection operation Projection coefficient in the eigenspace of the health spectrum The reconstruction error was calculated. ; Calculate projection coefficients The Mahalanobis distance relative to the mean vector and covariance matrix in the reference statistic ,in and These are the mean vector and covariance matrix of the projection coefficients calculated from historical health spectral data, respectively. By fusing reconstruction error and Mahalanobis distance, the position of each pixel in the hyperspectral image cube data is generated. Comprehensive spectral anomaly index ,in For the reconstruction error variance in the reference statistic, For degrees of freedom The 95% quantile of the chi-square distribution is the weight coefficient.

[0034] Specifically, after receiving the population metabolism warning signal encapsulated with specific aquaculture area identification information, the hyperspectral imaging device deployed in the corresponding area is scheduled according to the identification information; under the preset scanning parameters, the hyperspectral imaging device performs non-contact line scanning or area array scanning on the bullfrog population in the aquaculture area from top to bottom to obtain the original hyperspectral image data; after radiometric calibration and reflectance conversion of the data, hyperspectral image cube data is formed, and its dimension is H×W×L, where H is the total number of pixels in the image height direction, W is the total number of pixels in the image width direction, and L is the total number of spectral bands; each pixel point in the image is uniquely identified by its spatial coordinates (h, w), where h∈[1, H] is the row index and w∈[1, W] is the column index, and this pixel point corresponds to a spectral vector containing L band reflectance values, denoted as .

[0035] Hyperspectral image data collected from multiple historical healthy bullfrog individuals (confirmed without clinical symptoms) under similar environmental conditions. Based on the hyperspectral image data of the same group of historical healthy bullfrogs, a healthy spectral eigen-space and its associated reference statistics are pre-generated, and this group of data is collected from a confirmed healthy bullfrog population. The tensor decomposition method is used to extract the healthy spectral eigen-space from the hyperspectral image data of the historical healthy bullfrogs. Specifically, multiple healthy hyperspectral image data are regarded as a third-order tensor , and by performing Tucker decomposition on the spectral dimension (the 3rd order), a core tensor and factor matrices are found such that , and the basis matrix is the healthy spectral eigen-space, where is the preset eigen-dimension( <L). At the same time, the reference statistics are calculated; the spectral vector of each healthy pixel is projected onto this eigen-space to obtain the projection coefficient , and the mean vector and covariance matrix are calculated based on the projection coefficients of all healthy pixels; the reconstruction error of each healthy pixel spectral vector is calculated, and its variance is calculated based on all reconstruction errors. The healthy spectral eigen-space and the reference statistics It is stored as a health spectral database. Among them, compared with conventional dimensionality reduction methods (such as principal component analysis), the core advantage of the tensor decomposition method is that it can simultaneously maintain and utilize the inherent high-order correlation of hyperspectral image cubes in both spatial and spectral dimensions. In the bullfrog farming scenario, water surface reflection, ripple interference, and body surface mucus coverage can introduce strong and complex optical noise. At this time, the tensor decomposition method can robustly extract the weak spectral distortion features caused by early subcutaneous lesions (such as weak bleeding points, tissue fluid exudation, or early ulcers) and hidden in the noise background from high-dimensional data contaminated by such noise. These features are difficult to effectively separate and identify using conventional RGB image processing and ordinary dimensionality reduction methods that rely solely on spatial morphology or single spectral curves.

[0036] For each pixel (h, w) in the hyperspectral image cube data, its spectral vector is: ;calculate In the intrinsic space of the health spectrum Projection coefficients in And calculate the reconstruction error. ; Calculate the projection coefficient Mahalanobis distance relative to the reference statistic By fusing reconstruction error and Mahalanobis distance, a comprehensive spectral anomaly index is generated. ,in, (0≤ ≤1) is the preset weighting coefficient. For degrees of freedom The 95th percentile of the chi-square distribution; The method for determining the value is as follows: using historical healthy bullfrog hyperspectral image data (i.e., the same set of data used to construct the intrinsic space of the healthy spectrum), the reconstruction error and Mahalanobis distance of each pixel are calculated on this set of healthy spectral data. By analyzing the joint distribution of the reconstruction error and Mahalanobis distance, the value is set. The value is determined so that the distribution of the final comprehensive spectral anomaly index A(h,w) on this set of health spectral data meets the preset statistical characteristics requirements. For example, the value is determined by... The goal is to set the percentage of pixels in the health spectral data whose A(h,w) exceeds a certain threshold (e.g., the 99th percentile of the distribution of A values ​​in the health spectral data) (i.e., the false positive rate) to be below an acceptable level (e.g., 1%). The larger the value, the more significant the difference between the spectral characteristics of that pixel location and the health database, and the higher the possibility of early pathological abnormalities.

[0037] In one embodiment of the present invention, the subclinical screening module locates target individuals exhibiting abnormal early pathological spectral characteristics from the population and outputs a list of spatial coordinates of the target individuals, including the following steps: Based on the pixel position in the hyperspectral image cube data Comprehensive spectral anomaly index Multiple candidate connected regions are obtained through image segmentation and connected component analysis, and those with areas exceeding a preset threshold are selected. The candidate connected regions are further segmented to obtain sub-connected regions; all connected regions, including the unsegmented candidate connected regions and the segmented sub-connected regions, are collectively formed into a set of connected regions; each connected region is processed as an independent target individual; the connected region refers to a continuous region composed of spatially adjacent pixels that are all marked as suspected anomalies; For each connected region in the set of connected regions Calculate its confidence score The calculation formula is as follows: in, Connected region Total number of pixels included. Connected region The actual area occupied in the resulting image. The standard individual imaging area value is obtained by statistically analyzing historical hyperspectral image data of healthy bullfrogs. Confidence score Compare with the preset confidence threshold to filter out all For connected regions with a confidence threshold, each selected connected region is identified as a target individual exhibiting abnormal early pathological spectral characteristics. The geometric mean of all pixel coordinates within the connected region corresponding to each target individual is calculated and used as the spatial coordinates of that target individual. The spatial coordinates of all target individuals exhibiting abnormal early pathological spectral characteristics are summarized, and a list of target individual spatial coordinates is generated and output.

[0038] Specifically, an initial anomaly threshold is set based on the comprehensive spectral anomaly index A(h,w) of each pixel location (h,w) in the hyperspectral image cube data. This threshold is determined based on the statistical distribution of A(h,w) in historical healthy bullfrog hyperspectral image data, for example, taking the 99th percentile of A(h,w) values ​​in the data; and comparing A(h,w) of each pixel with... In comparison, if If a pixel is found to be abnormal, it is marked as a suspected anomalous pixel (assigned a value of 1); otherwise, it is considered a normal pixel (assigned a value of 0), resulting in binary image B. A morphological closing operation is then performed on binary image B using a structuring element of a preset size (e.g., 3×3), first dilating and then eroding to fill the small holes in the anomalous region, thus obtaining the image. ;right Perform 8-connected region labeling, meaning that if two suspected anomalous pixels are adjacent in the horizontal, vertical, or diagonal direction, they are considered to be the same region; after labeling, multiple candidate connected regions are obtained, denoted as... Each region is a set of spatially adjacent suspected abnormal pixels.

[0039] Calculate each candidate connected component pixel area (Number of pixels), and converted to the actual physical area Area based on the spatial resolution of the imaging system. Preset maximum reasonable single-unit area; This value is determined based on the physical projection area of ​​a single healthy bullfrog individual obtained from the historical hyperspectral image data of healthy bullfrogs (e.g., taking the mean of this area distribution plus three times the standard deviation). For each ,like If the region is not found, then retain it and add it directly to the final connected component set; if Then, the watershed segmentation algorithm is used to further segment the region: calculate the distance transformation graph within the region, find local maxima as markers, and perform watershed segmentation to obtain several sub-connected regions. All sub-regions are added to the final set. All retained candidate connected regions and the sub-connected regions generated by the split are merged to form the final set of connected regions. Each of them As an independent candidate region for a target individual.

[0040] For each connected region Calculate its confidence score : ,in, For the region The total number of pixels; The actual physical area of ​​the region (unit: cm²) is obtained by multiplying the number of pixels by the ground sampling distance (imaging system calibration parameter); The average imaging area of ​​standard individuals is statistically obtained from historical hyperspectral image data of healthy bullfrogs. A pre-set confidence threshold is used. For each connected region ,like If the pattern is correct, then the region is determined to correspond to a target individual exhibiting abnormal early pathological spectral characteristics; otherwise, it is determined to be a non-target (e.g., noise, background abnormalities, etc.). The value is determined based on historical hyperspectral image data of healthy bullfrogs. This data is processed using the same image processing and connected component extraction steps, and the confidence scores of all connected components are calculated to form a set of health baseline scores. A high statistical percentile (e.g., the 95th or 99th percentile) of this set is taken as the baseline score. The value of .

[0041] For each connected region identified as the target individual Calculate its spatial coordinates in the coordinate system of the actual aquaculture area. The coordinates are defined as the mapping positions in real space of the geometric centers (centroids) of all pixels within the connected region. Calculating the connected region... centroid in the image coordinate system : Using pre-calibrated imaging system intrinsic and extrinsic parameters (camera position, attitude, focal length, etc.), the image coordinates are transformed through perspective projection inverse transformation. Mapping to the actual ground coordinate system with a fixed point in the aquaculture area (such as a corner of the pond) as the origin, we obtain the actual spatial coordinates. The spatial coordinates of all target individuals are aggregated to generate a structured list of target individual spatial coordinates. Each record in the list is assigned a unique target individual identifier (ID, e.g., 1, 2, 3, ...) in sequence and contains the corresponding spatial coordinates. .

[0042] In one embodiment of the present invention, the pathogen detection module drives an automated sampling device to perform guided automated sampling of the target individual based on the target individual spatial coordinate list, including the following steps: Based on the target individual spatial coordinate list, obtain Spatial coordinates of the target individual to be sampled Simultaneously, obtain the spatial coordinates of non-target individuals, denoted as... ,common One, of which The total number of non-target individuals; With the goal of minimizing the total sampling path length and reducing interference with non-target individuals, a motion path traversing all target individuals is planned for the automated sampling device; this planning is achieved by solving the following composite objective function. Minimization problem implementation: in, To characterize the coordinate sequence of the spatial path points that the automated sampling device is to traverse; To estimate the total sampling time; For the sampling device at time Spatial location; For the first The location of a non-target individual ; , , For weights and scale parameters; The minimization problem is solved using a cooperative adaptive particle swarm optimization algorithm to obtain the optimal path coordinate sequence. That is, an optimal motion trajectory consisting of a series of spatial path points; based on the optimal motion trajectory, the automated sampling device is driven to move sequentially to the position of each target individual and complete automated sampling.

[0043] Specifically, based on the list of target individual spatial coordinates, M target individuals identified as exhibiting abnormal early pathological spectral characteristics are obtained; the two-dimensional spatial coordinates of the m-th target individual are extracted from the list, denoted as... Where m = 1, 2, ..., M; the coordinates of all target individuals constitute a set. Simultaneously, the location information of non-target individuals (i.e., bullfrog individuals that are not judged to be abnormally healthy or have not reached the warning threshold) within the same breeding area is obtained; let the total number of identified non-target individuals be N, and the two-dimensional spatial coordinates of the nth non-target individual be denoted as . Where n = 1, 2, ..., N; the coordinates of all non-target individuals constitute a set. .

[0044] Automated sampling devices (e.g., a robot equipped with a robotic arm that can move along a track on the water's surface or the edge of a pond) need to start from a starting point (such as a fixed base station) and sequentially visit the collection. All target points The goal of the plan is to complete the sampling and eventually return to the starting point or reach the designated endpoint; the objective is to find an optimal trajectory that minimizes the total distance traveled while maximizing the distance from the non-target individual set. To minimize disturbance to healthy bullfrog populations, the movement and noise of automated sampling devices in high-density bullfrog farming conditions can easily disturb the population, potentially triggering stress responses in non-target individuals and escape behaviors in target individuals, thus increasing sampling uncertainty. Furthermore, strong stress responses may alter the physiological state of individuals, causing the concentration of metabolic markers in the collected biological samples to deviate from their true steady-state levels, introducing interference into subsequent pathogen analysis. Therefore, to ensure sampling success rate and the physiological authenticity of the samples, this path planning method must minimize the trajectory length while constraining the proximity of the sampling device to non-target individuals as a core cost. A composite objective function J is defined to quantify the cost of a candidate path: ,in, It is the sequence of critical path points that defines the path. These are the coordinates of the k-th (k=0,1,…,K) path point, and the path is smoothly constructed by line segments connecting these points. Starting from, The endpoint; It is the estimated total time required to complete the entire sampling task, and the total length of the sampling task path. and the average moving speed of the sampling device and the fixed sampling operation time at each target point Related, expressed as ; The spatial coordinates of the sampling device at time t (obtained through the navigation and positioning system of the sampling device, such as fusion of GPS, vision and inertial measurement unit). Indicates the device position at time t Location of the nth non-target individual The Euclidean distance between them; ( >0) is the weight coefficient of the interference penalty term. The larger the coefficient, the more the optimization algorithm tends to plan a path that is far away from all non-target individuals. ( ≥0) is the time cost weighting coefficient, used to weigh the cost between path length and total time cost; and The specific value needs to be preset according to the sampling task requirements (such as tolerance for disturbance and time urgency), usually in... ∈[0.1,5.0], Select from the range [0, 0.5]. ( >0) is the scale parameter of the disturbance effect. Physically, it can be understood as the safe distance scale at which an individual bullfrog feels the disturbance. It is usually set to be slightly larger than the average body width of the bullfrog (e.g., 0.15 meters).

[0045] The cooperative adaptive particle swarm optimization algorithm is used to solve the path planning problem. Each particle i represents a candidate path. Each particle contains an access sequence of length M. This indicates that M target individuals are visited. The order; between every two adjacent target individuals, R additional intermediate obstacle avoidance path points are allowed to be inserted. Randomly generated paths contain a preset number (denoted as...). Typically, an initial population of 50-200 particles is selected; each particle i has a position vector. and velocity vector Initialize the individual's historical best position. And the global historical best position gbest; set the maximum number of iterations in the optimization algorithm solution process. (Usually 100-500).

[0046] The initial population is divided into two co-evolving subpopulations: an exploration subpopulation and an exploratory subpopulation. The exploration subpopulation focuses on fine-grained searches in known promising regions, and its particle velocity updates use a lower inertial weight. and a high social learning factor The exploratory subpopulation focuses on exploring new spatial regions, and the velocity updates of particles in this subpopulation use a higher inertial weight. and higher cognitive learning factor The inertia weight parameter satisfies 0 < < <1, its specific value is determined based on the empirical range of the standard particle swarm optimization algorithm, and is usually taken as... ∈[0.3,0.5], ∈[0.7,0.9]; the learning factor parameters satisfy >0, >0, and its specific value is determined based on the empirical range of the standard particle swarm optimization algorithm and the requirements of co-evolution, usually taken as 0. ∈[1.8,2.2].

[0047] Define population aggregation degree ( (∈[0,1]) is a normalized measure of the average distance from all particles to the global optimal particle gbest; two clustering thresholds are preset, and the lower threshold is... and upper threshold Satisfying 0 < < <1, its specific value is determined based on the algorithm's desired balance between exploration and exploitation, and is usually taken as ∈[0.1,0.3], ∈[0.7,0.9]; Execute dynamic population management strategy based on population aggregation degree ζ; When ≤ (When the population density is high, it may get stuck in a local optimum.) Increase the proportion of exploring subpopulations and set the inertia weight of all particles to [value missing]. ;when ≥ When the population is dispersed, the proportion of developing subpopulations is increased, and the inertial weight of all particles is set to... ;when < < At the same time, the current subpopulation ratio and inertial weight remain unchanged.

[0048] In the In each iteration, for each dimension d of each particle i: , ,in, , It is a random number within the interval [0,1]; coefficient , , The value of is determined by the subpopulation type to which the particle currently belongs. If the particle belongs to the development subpopulation, then is taken as . = , = , = If it belongs to the exploratory subpopulation, then take... = , = , = The basic learning factor and Based on the empirical balance between cognitive and social components in the standard particle swarm optimization algorithm, it is usually determined that... = =0.5, and can also be adjusted within the range of 0,2] depending on the specific problem.

[0049] For each particle, based on its position vector Access order of Chinese encoding The coordinates of intermediate obstacle avoidance path points can be used to directly reconstruct the corresponding candidate paths. Calculate the path The composite objective function value The fitness value of a particle is defined as follows: ,in It is a very small positive number designed to prevent division by zero; the higher the fitness, the better the path; then each particle's... And the gbest of the entire population. Repeat the steps until the maximum number of iterations is reached. After the algorithm terminates, the globally optimal particle gbest is decoded to obtain the optimal path coordinate sequence. This sequence defines a path that starts from the starting point and efficiently and with low interference traverses all target individuals. The optimal trajectory for reaching the destination.

[0050] The obtained optimal path coordinate sequence The motion control system of the automated sampling device is sent to the device; These commands are then converted into specific motor control instructions, driving the device (such as an unmanned surface vessel) to move autonomously along a planned trajectory. The device then uses a navigation and positioning system (such as a system integrating GPS, vision, and inertial measurement units) to determine its own position relative to the coordinates of a target object. The distance is less than the preset sampling trigger distance (For example, at a distance of 0.3 meters), the device decelerates and precisely positions itself. The device then paused its movement and activated its onboard sampling mechanism (such as a multi-degree-of-freedom robotic arm). Following pre-set safety sampling procedures (such as gentle contact with the body surface and minimally invasive aspiration of body fluids or mucus), the robotic arm automatically sampled the target bullfrog. After sampling, the biological sample was automatically stored in a numbered, cryogenically preserved sample tube. The sampling device then continued along the planned path. Move to the next target individual location and repeat the location-sampling process until sampling of all M target individuals is completed; finally, all collected biological samples are sent to the on-site rapid molecular detection device of the pathogen detection module for subsequent pathogen-specific gene fragment analysis.

[0051] In one embodiment of the present invention, the pathogen detection module utilizes a rapid on-site molecular detection device to perform molecular detection on biological samples collected from the target individual targeting a preset bullfrog pathogen-specific gene fragment, and outputs a pathogen detection report, including the following steps: Receive and process biological samples collected from the target individual by the automated sampling device; inject the biological samples into a pre-set container. In a centrifugal microfluidic chip containing a primer set specific to a bullfrog pathogen, the biological sample is distributed to each independent reaction chamber during chip rotation; the chip contains... A separate reaction chamber, the first The reaction chamber is pre-set to the first A specific primer set for one pathogen, used for the specific detection of the first pathogen. Pathogens, ; Parallel isothermal nucleic acid amplification reactions were performed on biological samples allocated to each reaction chamber under isothermal conditions, where, corresponding to the first... The first pathogen Each reaction chamber has an amplification reaction kinetics described by a modified Logistic model: in, For the first reaction chamber The concentration of pathogen-specific double-stranded DNA products, , , The first The maximum rate of amplification reaction of a pathogen, its saturation concentration, and its degradation rate constant; Based on the modified Logistic kinetic model, the fluorescence signals of each reaction chamber acquired in real time are analyzed to determine the... The characteristic time at which the amplification curve of each reaction chamber reaches the preset fluorescence threshold. and its confidence interval; Based on the characteristic time and its confidence interval and preset effective signal reference time The comparison generates a target for the first. Detection data of pathogen-specific gene fragments; if The upper limit of the confidence interval is earlier than Then, the estimated initial template nucleic acid concentration is calculated based on the standard curve. : ,in, and The coefficients are those calibrated through preliminary experiments using standard samples; The detection data of all L pathogen-specific gene fragments are summarized, including characteristic time, confidence interval and corresponding initial template nucleic acid concentration estimates, to generate a structured pathogen detection report.

[0052] Specifically, the device receives biological samples (such as bullfrog body surface mucus or tissue fluid) collected and transported to the detection area by an automated sampling device. The on-site rapid molecular detection device includes a centrifugal microfluidic chip with a disk structure. The chip has a central common sample loading chamber and L independent microreaction chambers evenly distributed around it (L is the preset number of pathogens to be detected). Each microreaction chamber is pre-freeze-dried and embedded with a nucleic acid isothermal amplification reaction system for a specific bullfrog pathogen. This system includes at least: primer pairs for the pathogen-specific gene fragment, DNA polymerase, deoxyribonucleoside triphosphate, buffer, and fluorescent dye (such as SYBR Green I). During operation, a certain volume (e.g., 10 μL) of biological sample is injected into the central common sample loading chamber of the chip. The chip is then placed on the centrifuge rotor of the device. The centrifugation program is started, and under the action of centrifugal force, the sample liquid is evenly distributed to the L independent microreaction chambers around it through the microchannels. Each reaction chamber receives an equal volume of sample. This process realizes the parallel distribution and loading of a single sample for L pathogens. The chip with the sample added is transferred to the temperature control module, where the reaction is carried out under isothermal conditions (e.g., 65°C, the specific temperature depending on the isothermal amplification technique used, such as loop-mediated isothermal amplification). The multi-channel fluorescence detection system built into the on-site rapid molecular detection device simultaneously monitors the fluorescence signal intensity of L reaction chambers in real time. ( The fluorescence intensity is related to the concentration of the double-stranded DNA product in the reaction chamber. They are directly proportional, and their relationship is pre-calibrated through a calibration experiment as follows: Among them, the proportionality coefficient >0 and background fluorescence All of these are preset constants, determined through standard experiments; The results were obtained by linear regression after amplification experiments using target DNA standards of known concentrations. The fluorescence signal was determined by a stable baseline value obtained from a template-free (NTC) control experiment. Real-time fluorescence intensity sequences were acquired for each reaction chamber. Perform preprocessing (such as background subtraction and smoothing filtering).

[0053] For the The first pathogen corresponding to the One reaction chamber ( The amplification reaction kinetics are described using the following modified Logistic model: ,in, Indicates the first In the reaction chamber, at time t, the reaction is directed towards the first reaction chamber. The concentration of double-stranded DNA products generated by specific primer set amplification of the pathogen. ;>0 is the first The maximum intrinsic growth rate constant of the amplification reaction of a pathogen; >0 is the first The saturation concentration of each reaction system; ≥0 is the product degradation rate constant; parameter set The specific values ​​are not fixed in advance, but rather the modified Logistic model is combined with the preprocessed real-time fluorescence intensity data. Nonlinear least squares fitting (e.g., using the Levenberg-Marquardt algorithm) is performed to automatically find a set of optimal parameter values.

[0054] Based on the modified Logistic kinetic model, the fluorescence signals of each reaction chamber acquired in real time are analyzed to determine the... The characteristic time at which the amplification curve of each reaction chamber reaches the preset fluorescence threshold. and its confidence interval. A fluorescence threshold is preset for each reaction chamber. By fitting the modified Logistic model (i.e. The fluorescence signal can be calculated by inverse solution of the functional form of t. Reaching the threshold Corresponding characteristic time Due to the presence of noise in the data, the Logistic model fit is uncertain. Therefore, the bootstrap method or an error propagation method based on the covariance matrix of the fitting parameters is used to calculate... Confidence intervals, such as 95% confidence intervals. The threshold was determined based on a large number of control experiments using healthy samples (samples excluding the target pathogen). The distribution of the maximum fluorescence signal values ​​for all healthy control samples throughout the entire detection period (e.g., 60 minutes) was statistically analyzed, and the 99th percentile of this distribution was taken as the threshold value. This is to ensure that the probability of the signal exceeding this threshold in healthy samples is less than 1%. Typical The value range is within the background fluorescence intensity. Between 1.5 and 3 times.

[0055] Preset an effective signal reference time For the first Each reaction chamber, based on its characteristic time upper limit of confidence interval and Analyze the comparative relationships; if If no valid signal is detected, the output will show the detection data status as "no valid signal detected," and the initial template concentration estimate will not be calculated; if If a valid signal is detected, the output will show the detection data status as "valid signal detected," and the estimated initial template nucleic acid concentration will be calculated based on the standard curve. : ; where, standard curve coefficients and It was determined through pre-established calibration experiments; for each pre-defined bullfrog pathogen (the first... For each pathogen, a known concentration of nucleic acid standard (e.g., serially diluted plasmid DNA or in vitro transcribed RNA containing the target gene fragment) was used to perform multiple repeated isothermal amplification assays, and the average characteristic time corresponding to each known concentration of standard was recorded. The value is then fitted using least squares linear regression. and The linear relationship between them is used to obtain the coefficient unique to this pathogen. and The standard curve parameters for all L pathogens were calibrated before system deployment and stored in the processor or system database of the detection device. It provides quantitative information on pathogen load to assess the relative extent or intensity of infection. It summarizes detection data for all L pathogen-specific gene fragments, including characteristic time, confidence intervals, detection data status, and corresponding initial template nucleic acid concentration estimates, generating a structured pathogen detection report.

[0056] In one embodiment of the present invention, the system control and decision-making module includes the following steps: The system receives a population metabolic early warning signal from the pre-symptom early warning module, a list of target individual spatial coordinates from the subclinical screening module, and a pathogen detection report from the pathogen detection module; it extracts the breeding area identifier, environmental parameters, and comprehensive deviation index of metabolic markers from the population metabolic early warning signal, extracts the spatial distribution information of the target individuals from the list of target individual spatial coordinates, and extracts the pathogen type, detection data, and corresponding initial template nucleic acid concentration estimate from the pathogen detection report. The system queries a built-in disease knowledge graph, which includes five types of nodes: disease type, pathogen, clinical symptoms, metabolic markers, and environmental conditions, as well as edges representing the causal, correlation, susceptibility, and manifestation relationships among them. The extracted metabolic markers are combined with deviation indices, initial template nucleic acid concentration estimates, environmental parameters, and spatial distribution information of target individuals, and matched with the corresponding nodes and relationships in the graph. Through graph query and data association algorithms, the system outputs the disease knowledge graph subgraphs identified by the matching operation and the corresponding data association weight information for each subgraph. Based on the output disease knowledge graph subgraph, data association weight information, and pre-stored treatment reference information in the disease knowledge graph, a structured bullfrog biosafety risk multimodal data fusion analysis report is generated.

[0057] Specifically, the system simultaneously receives population metabolic early warning signals from the pre-symptom warning module, a list of target individual spatial coordinates from the subclinical screening module, and structured pathogen detection reports from the pathogen detection module. The population metabolic early warning signals are analyzed to extract specific aquaculture area identifiers, time-series data of aquatic environmental parameters triggered by the warning (including at least real-time or moving average values ​​of key parameters such as water temperature, dissolved oxygen, pH, and ammonia nitrogen concentration), and time-series data of the comprehensive statistical deviation index. The target individual spatial coordinate list is analyzed to extract spatial distribution information, i.e., the total number of target individuals identified as exhibiting abnormal early pathological spectral characteristics, and the two-dimensional spatial coordinates of each target individual in the actual coordinate system of the aquaculture area. The structured pathogen detection reports are analyzed to extract detection data for each bullfrog pathogen preset in the system, including characteristic time, confidence interval, and corresponding initial template nucleic acid concentration estimates.

[0058] The system incorporates a disease knowledge graph, a structured semantic network, which serves as the core knowledge base for data association analysis. This disease knowledge graph contains five types of entity nodes: the first type is disease type nodes, representing specific bullfrog diseases such as red leg disease and skin rot; the second type is pathogen nodes, representing known pathogenic microorganisms such as Aeromonas hydrophila and frog iridovirus; the third type is clinical symptom nodes, representing pathological manifestations visible to the naked eye; the fourth type is metabolic marker nodes, representing specific metabolites monitored by the system that are related to the physiological processes of the disease; and the fifth type is environmental condition nodes, representing aquatic environmental factors that influence the occurrence and development of the disease. These nodes are connected by various directed edges to characterize causal relationships, inducing relationships, manifestation relationships, and statistical associations between nodes. The knowledge system of this disease knowledge graph originates from three core components: First, it integrates authoritative knowledge from publicly available veterinary pathology literature databases to form a deterministic framework of associations between diseases, pathogens, and typical symptoms; second, it introduces a multimodal control dataset of "symptom-metabolism-spectrum" accumulated through artificially induced disease experiments in a laboratory environment to establish a correlation model between early, quantifiable pathophysiological indicators and disease development stages; and third, it integrates real historical disease record logs from cooperative farms, extracting statistical patterns and risk characteristics of disease occurrence and development under actual complex farming conditions through retrospective analysis of environmental time-series data, population anomaly records, and final diagnostic results. In the specific construction of the disease knowledge graph, entity identification, relation extraction, and confidence assessment are performed on heterogeneous information from the knowledge system's sources, and persistent storage and management are achieved using a graph database, ultimately forming a disease knowledge graph that can be queried and used for associative reasoning.

[0059] The extracted multi-source information is matched with corresponding nodes in the disease knowledge graph. Detection data from pathogen detection reports is matched with pathogen nodes in the disease knowledge graph; environmental parameters exceeding preset health thresholds in population metabolic early warning signals (e.g., water temperature above 30 degrees Celsius) are matched with environmental condition nodes in the disease knowledge graph; and the overall abnormal trend of actual monitored metabolic markers (reflected by a comprehensive statistical deviation index) is matched with metabolic marker nodes in the disease knowledge graph. After completing the matching of data with graph nodes, a graph query and data association algorithm is used for processing. The disease knowledge graph subgraph refers to a locally connected substructure dynamically extracted from the system's built-in global disease knowledge graph through a matching algorithm. This subgraph takes one or more disease type nodes currently matched as its core, and also includes pathogen nodes, metabolic marker nodes, environmental condition nodes, and clinical symptom nodes directly connected to these core disease nodes and activated in this round of data matching, while preserving the association edges between these nodes. The system outputs one or more disease knowledge graph subgraphs identified through the matching operation, along with the corresponding data association weight information for each subgraph. Specifically, for each pathogen that detects a valid signal, its association weight is set based on its initial template nucleic acid concentration estimate; the higher the concentration estimate, the greater the weight. For each abnormal environmental condition, its association weight is set based on its deviation from the healthy threshold. For overall abnormalities in metabolic markers, its association weight is quantified based on the degree of deviation of the comprehensive statistical deviation index relative to the healthy baseline. Data association weight information for each disease type node is generated by calculating the weighted association degree of all information connected to each associated disease type node.

[0060] Based on the output disease knowledge graph subgraphs, data association weight information, and pre-stored treatment reference information in the disease knowledge graph, a structured multimodal data fusion analysis report on bullfrog biosafety risks is generated. This report typically includes the following parts: First, a risk overview summary, listing one or more potential risk types with high association weights and their weight information; second, multi-source data association analysis, describing the association between various data types and knowledge graph nodes from aspects such as pathogen data, environmental data, metabolic abnormality data, and spatial distribution characteristics of target individuals; third, risk treatment reference suggestions, which, based on the treatment reference information in the association knowledge graph, provide tiered suggestions for the identified potential risk types and specific situations, including environmental control schemes (such as adjusting water temperature and increasing oxygenation), and operational guidelines for subsequent monitoring. The generated report is pushed to aquaculture managers in real time through the system's human-computer interaction interface for decision support and is simultaneously stored in the system database for historical query and decision traceability.

[0061] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An intelligent system for identifying bullfrog diseases, characterized in that, Includes the following modules: Pre-symptom warning module: Monitors the concentrations of at least two specific physiological and pathological metabolic markers released into the water by bullfrogs in the breeding area, and simultaneously collects environmental parameter data of the breeding water; Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a generalized autoregressive conditional heteroscedasticity model is used to construct a dynamic baseline of the healthy population; By analyzing the statistical deviation of metabolic marker concentrations from the dynamic baseline of the healthy population, a population metabolic warning signal is generated before individuals show visible symptoms; Subclinical screening module: In response to the population metabolic early warning signal, the hyperspectral imaging device is activated to perform non-contact scanning of the bullfrog population in the corresponding breeding area to obtain hyperspectral image cube data; by analyzing and comparing the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, the target individuals showing abnormal early pathological spectral characteristics are located from the population, and a list of spatial coordinates of the target individuals is output. Pathogen detection module: Based on the list of target individual spatial coordinates, drives the automated sampling device to perform guided automated sampling of the target individual; Using a rapid on-site molecular detection device, molecular detection of pre-set bullfrog pathogen-specific gene fragments is performed on biological samples collected from target individuals, and a pathogen detection report is output. System Control and Decision Module: By querying the system's built-in disease knowledge graph, the module performs fusion analysis and correlation reasoning on population metabolic early warning signals, target individual spatial coordinate lists, and pathogen detection reports to generate a multimodal data fusion analysis report on bullfrog biosafety risks.

2. The intelligent bullfrog disease identification system according to claim 1, characterized in that, The pre-symptom warning module monitors the concentrations of at least two specific physiological and pathological metabolic markers released into the water by the bullfrogs in the breeding area, and simultaneously collects environmental parameter data of the breeding water. Based on the continuously monitored metabolic marker concentrations and environmental parameter data, a generalized autoregressive conditional heteroscedasticity model is used to construct a dynamic baseline of the healthy population, including the following steps: The farm is divided into at least two breeding areas. Monitoring devices are deployed in each breeding area, and each breeding area is assigned a unique identifier, which is then linked to the data collected by the corresponding monitoring device. Time-series data of the concentrations of at least two preset specific physiological and pathological metabolic markers released by bullfrog metabolism in the water of the breeding area are continuously collected, and time-series data of water environmental parameters in the breeding area are collected simultaneously. The time-series data of metabolic marker concentrations and environmental parameters are preprocessed. Using preprocessed environmental parameter time-series data as covariates, a generalized autoregressive conditional heteroscedasticity model was employed to jointly model the preprocessed metabolic biomarker concentration time-series data. Through iterative calculations of the model, the dynamic health baseline values ​​of each metabolic biomarker under corresponding environmental conditions were determined. and its dynamic standard deviation The dynamic standard deviation The iterative calculation process is as follows: in, , indicating the first Metabolic markers, The total number of metabolic biomarkers monitored and ≥2; Indicates the first Index of each sampling time; for The difference between time-series data of metabolic biomarker concentrations at a given time and their corresponding dynamic healthy baseline concentration. Indicates the first Metabolic markers in Concentration time-series data after time-series preprocessing; , , For parameters; Indicates the first Environmental parameters in Preprocessed data at any given time, , The number of environmental parameter types; It's about environmental parameters. The default function; Environmental stress For the first The influence coefficient of volatility of various metabolic markers; The dynamic health concentration baseline With dynamic standard deviation Together they constitute the dynamic baseline of the healthy population.

3. The intelligent bullfrog disease identification system according to claim 2, characterized in that, The pre-symptom warning module analyzes the statistical deviation of metabolic marker concentrations relative to the dynamic baseline of a healthy population to generate a population metabolic warning signal before an individual develops visible symptoms. This includes the following steps: Dynamic health concentration baseline based on dynamic baseline of healthy population With dynamic standard deviation For the first Metabolic markers, calculating their role in Standardized instantaneous deviation at time ,in, for Time-series data of metabolic biomarker concentrations after preprocessing. It is a tiny positive number; Based on the standardized instantaneous deviation, a dynamic model characterizing the correlation between different metabolic biomarkers is constructed and updated online, and the multi-scale Mahalanobis distance is calculated as a comprehensive statistical deviation index. : in, For delay The standardized instantaneous deviation vector of the step. For the delay extracted from the dynamic model The correlation matrix at step time, For exponentially decaying weights, This represents the maximum delay steps. Comprehensive statistical deviation index The input includes a Hidden Markov Model (HMM) encompassing healthy, subclinical, and clinical states, which is decoded to obtain a population health state sequence. When the decoded state sequence indicates that the population health state has been consistently in a subclinical state over multiple consecutive sampling periods, a warning confidence level is calculated based on Bayesian inference. ; If and only if the population health status remains subclinical, with a warning confidence level Exceeding the preset threshold Furthermore, when the key environmental parameter values ​​of the aquaculture area corresponding to the warning signal exceed the corresponding preset health threshold range, a group metabolism warning signal is generated, which encapsulates the identification information of the aquaculture area, the health status of the group, the confidence level, and the time series data of the environmental parameters.

4. The intelligent bullfrog disease identification system according to claim 1, characterized in that, In the subclinical screening module, in response to the population metabolic early warning signal, a hyperspectral imaging device is activated to perform a non-contact scan of the bullfrog population in the corresponding breeding area to acquire hyperspectral image cube data. By analyzing and comparing the spectral reflectance characteristics of each pixel in the hyperspectral image cube data with the differences in the healthy spectral database, the following steps are included: Upon receiving the population metabolism early warning signal encapsulated with aquaculture area identification information, the hyperspectral imaging device is activated to perform a non-contact scan of the bullfrog population within the corresponding aquaculture area, acquiring hyperspectral image cube data with dimensions of [missing information]. ,in This represents the total number of pixels in the height direction of the image. This represents the total number of pixels in the image width direction. The total number of spectral bands; define the two-dimensional spatial coordinates of each pixel location in the image as its row index in the image. With column index ,in ; The system calls upon a pre-generated health spectral intrinsic space and its associated reference statistics based on a set of historical healthy bullfrog hyperspectral image data. The health spectral intrinsic space is extracted from the historical healthy bullfrog hyperspectral image data using tensor decomposition. The reference statistics include the mean vector and covariance matrix calculated from the projection coefficients of all pixels in the historical healthy bullfrog hyperspectral image data, as well as the variance calculated based on the reconstruction error of all pixels. The health spectral intrinsic space and reference statistics are stored in a health spectral database. For pixel locations in hyperspectral image cube data The corresponding pixel spectral vector Calculate through projection operation Projection coefficient in the eigenspace of the health spectrum The reconstruction error was calculated. ; Calculate projection coefficients The Mahalanobis distance relative to the mean vector and covariance matrix in the reference statistic ,in and These are the mean vector and covariance matrix of the projection coefficients calculated from historical health spectral data, respectively. By fusing reconstruction error and Mahalanobis distance, the position of each pixel in the hyperspectral image cube data is generated. Comprehensive spectral anomaly index ,in For the reconstruction error variance in the reference statistic, For degrees of freedom The 95th percentile of the chi-square distribution, These are the weighting coefficients.

5. The intelligent bullfrog disease identification system according to claim 4, characterized in that, The subclinical screening module identifies target individuals exhibiting abnormal early pathological spectral characteristics from the population and outputs a list of spatial coordinates of these target individuals, including the following steps: Based on the pixel position in the hyperspectral image cube data Comprehensive spectral anomaly index Multiple candidate connected regions are obtained through image segmentation and connected component analysis, and those with areas exceeding a preset threshold are selected. The candidate connected regions are further segmented to obtain sub-connected regions; all connected regions, including the unsegmented candidate connected regions and the segmented sub-connected regions, are collectively formed into a set of connected regions; each connected region is processed as an independent target individual; the connected region refers to a continuous region composed of spatially adjacent pixels that are all marked as suspected anomalies; For each connected region in the set of connected regions Calculate its confidence score The calculation formula is as follows: in, Connected region Total number of pixels included. Connected region The actual area occupied in the resulting image. The standard individual imaging area value is obtained by statistical analysis based on historical hyperspectral image data of healthy bullfrogs; Confidence score Compare with the preset confidence threshold to filter out all For connected regions with a confidence threshold, each selected connected region is identified as a target individual exhibiting abnormal early pathological spectral characteristics. The geometric mean of all pixel coordinates within the connected region corresponding to each target individual is calculated and used as the spatial coordinates of that target individual. The spatial coordinates of all target individuals exhibiting abnormal early pathological spectral characteristics are summarized, and a list of target individual spatial coordinates is generated and output.

6. The intelligent bullfrog disease identification system according to claim 1, characterized in that, In the pathogen detection module, based on the target individual spatial coordinate list, the automated sampling device is driven to perform guided automated sampling of the target individual, including the following steps: Based on the target individual spatial coordinate list, obtain Spatial coordinates of the target individual to be sampled Simultaneously, obtain the spatial coordinates of non-target individuals, denoted as... ,common One, of which The total number of non-target individuals; With the goal of minimizing the total sampling path length and reducing interference with non-target individuals, a motion path traversing all target individuals is planned for the automated sampling device; this planning is achieved by solving the following composite objective function. Minimization problem implementation: in, To characterize the coordinate sequence of the spatial path points that the automated sampling device is to traverse; To estimate the total sampling time; For the sampling device at time Spatial location; For the first The location of a non-target individual ; , , For weights and scale parameters; The minimization problem is solved using a cooperative adaptive particle swarm optimization algorithm to obtain the optimal path coordinate sequence. That is, an optimal motion trajectory consisting of a series of spatial path points; based on the optimal motion trajectory, the automated sampling device is driven to move sequentially to the position of each target individual and complete automated sampling.

7. The intelligent bullfrog disease identification system according to claim 1, characterized in that, The pathogen detection module utilizes a rapid on-site molecular detection device to perform molecular detection on biological samples collected from the target individual, targeting specific gene fragments of a pre-set bullfrog pathogen, and outputs a pathogen detection report, including the following steps: Receive and process biological samples collected from the target individual by the automated sampling device; inject the biological samples into a pre-set container. In a centrifugal microfluidic chip containing a primer set specific to a bullfrog pathogen, the biological sample is distributed to each independent reaction chamber during chip rotation; the chip contains... A separate reaction chamber, the first The reaction chamber is pre-set to the first A specific primer set for one pathogen, used for the specific detection of the first pathogen. Pathogens, ; Parallel isothermal nucleic acid amplification reactions were performed on biological samples allocated to each reaction chamber under isothermal conditions, where, corresponding to the first... The first pathogen Each reaction chamber has an amplification reaction kinetics described by a modified Logistic model: in, For the first reaction chamber The concentration of pathogen-specific double-stranded DNA products, , , The first The maximum rate of amplification reaction of a pathogen, its saturation concentration, and its degradation rate constant; Based on the modified Logistic kinetic model, the fluorescence signals of each reaction chamber acquired in real time are analyzed to determine the... The characteristic time at which the amplification curve of each reaction chamber reaches the preset fluorescence threshold. and its confidence interval; Based on the characteristic time and its confidence interval and preset effective signal reference time The comparison generates a target for the first. Detection data of pathogen-specific gene fragments; if The upper limit of the confidence interval is earlier than Then, the estimated initial template nucleic acid concentration is calculated based on the standard curve. : ,in, and The coefficients are those calibrated through preliminary experiments using standard samples; The detection data of all L pathogen-specific gene fragments are summarized, including characteristic time, confidence interval and corresponding initial template nucleic acid concentration estimates, to generate a structured pathogen detection report.

8. The intelligent bullfrog disease identification system according to claim 1, characterized in that, The system control and decision-making module includes the following steps: The system receives a population metabolic early warning signal from the pre-symptom early warning module, a list of target individual spatial coordinates from the subclinical screening module, and a pathogen detection report from the pathogen detection module; it extracts the breeding area identifier, environmental parameters, and comprehensive deviation index of metabolic markers from the population metabolic early warning signal, extracts the spatial distribution information of the target individuals from the list of target individual spatial coordinates, and extracts the pathogen type, detection data, and corresponding initial template nucleic acid concentration estimate from the pathogen detection report. The system queries a built-in disease knowledge graph, which includes five types of nodes: disease type, pathogen, clinical symptoms, metabolic markers, and environmental conditions, as well as edges representing the causal, correlation, susceptibility, and manifestation relationships among them. The extracted metabolic markers are combined with deviation indices, initial template nucleic acid concentration estimates, environmental parameters, and spatial distribution information of target individuals, and matched with the corresponding nodes and relationships in the graph. Through graph query and data association algorithms, the system outputs the disease knowledge graph subgraphs identified by the matching operation and the corresponding data association weight information for each subgraph. Based on the output disease knowledge graph subgraph, data association weight information, and pre-stored treatment reference information in the disease knowledge graph, a structured bullfrog biosafety risk multimodal data fusion analysis report is generated.

Images