Monitoring analysis system and method for potato breeding

Abstract

The invention discloses a monitoring analysis system and method for potato breeding, and relates to the technical field of agricultural information technology and intelligent breeding, and the system comprises a multi-modal data acquisition module which is configured to synchronously obtain greenhouse environment parameters, potato plant multispectral imaging data and leaf metabolite concentration data; and the space-time causal modeling module is connected to the multi-modal data acquisition module, and is used for constructing a dynamic correlation model among environmental fluctuation, gene expression and phenotypic characteristics, and deducing and quantifying the genetic effect of the environmental abnormal event based on anti-fact intervention. According to the monitoring analysis system and method for potato breeding, through multi-modal data fusion and space-time causal modeling, the accuracy and early warning timeliness of potato breeding monitoring are improved, the limitation of traditional visible light imaging is broken through, and the incubation period disease recognition accuracy is improved compared with a traditional method.

Description

Technical Field

[0001] The present invention relates to the fields of agricultural information technology and intelligent breeding technology, and in particular to a monitoring and analysis system and method for potato breeding. Background Art

[0002] Potato is an important food and cash crop in the world, but the problem of seed potato degeneration seriously restricts the sustainable development of the industry. Although traditional breeding monitoring technology has introduced IoT sensor equipment, it still faces systemic defects. In terms of disease monitoring, traditional visible light imaging technology can only identify overt symptoms, such as leaf yellowing and curling, but lacks sensitivity to latent diseases. Studies have shown that potato seed degeneration is positively correlated with virus accumulation, but existing technologies do not integrate multispectral imaging (near-infrared, hyperspectral) and gene expression data (disease resistance gene markers), making it impossible to extract early pathological features and predict degeneration risks. In addition, most monitoring platforms use data sampling at a single time node and have not established a time series-based environmental fluctuation traceability mechanism. For example, continuous high temperatures may affect seed potato stress resistance by activating heat shock protein expression, but existing systems find it difficult to trace the temporal relationship between abnormal environmental events and changes in gene expression, resulting in inaccurate attribution analysis of seed degeneration. Summary of the Invention

[0003] (1) Technical problems solved

[0004] In response to the shortcomings of the existing technology, the present invention provides a monitoring and analysis system and method for potato breeding, which solves the problems of insufficient detection of latent diseases in potato breeding, lack of temporal correlation between environmental parameters and gene expression, fragmentation of phenotype-genotype data, and delayed early warning of species degeneration.

[0005] (2) Technical solution

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a monitoring and analysis system for potato breeding, comprising:

[0007] a multimodal data acquisition module configured to simultaneously acquire greenhouse environmental parameters, potato plant multispectral imaging data, and leaf metabolite concentration data;

[0008] a spatiotemporal causal modeling module, connected to the multimodal data acquisition module, for constructing a dynamic association model between environmental fluctuations, gene expression, and phenotypic characteristics, and quantifying the genetic effects of environmental anomalies based on counterfactual interventions;

[0009] a degradation warning and control module, connected to the spatiotemporal causal modeling module, for generating a multimodal risk index and triggering adaptive environmental control instructions;

[0010] The data storage and interaction module is used to store multi-source heterogeneous data and model parameters, and provide a cross-platform data interface.

[0011] Wherein, the multimodal data acquisition module includes:

[0012] A distributed environmental sensor array integrates temperature, humidity, light intensity, and CO2 concentration sensors, uploading data every ten minutes;

[0013] A dual-band excitation multispectral imaging device is configured to simultaneously scan the plant canopy in the near-infrared (900-1700nm) and short-wave infrared (1700-2500nm) bands, and extract stomatal density, secondary metabolite distribution, and pathogenic bacteria characteristic spectra.

[0014] The microfluidic metabolome sampling unit uses surface-enhanced Raman spectroscopy (SERS) to detect the concentrations of salicylic acid (SA) and jasmonic acid (JA) in leaf exudates in real time.

[0015] Preferably, the dual-band excitation multispectral imaging device further comprises:

[0016] The laser-induced fluorescence excitation module uses a 405nm wavelength laser light source to stimulate leaf autofluorescence and capture the distribution characteristics of flavonoids;

[0017] The spectroscopic filter wheel is configured to switch between near-infrared and short-wave infrared filters in a single scan, acquiring dual-channel spectral images via an InGaAs sensor. A germanium-based filter is used for the near-infrared band (900-1700 nm), while a zinc sulfide-coated filter is used for the short-wave infrared band (1700-2500 nm). This switching is achieved within 0.5 seconds by a stepper motor. The InGaAs sensor, a Xenics XEVA-2.5-320, acquires dual-band images at 5 frames per second with a spatial resolution of 0.1 mm / pixel.

[0018] The spectral feature fusion unit performs noise suppression and feature alignment on dual-band images based on the generative adversarial network (GAN), and outputs a latent disease probability map.

[0019] Preferably, the spatiotemporal causal modeling module includes:

[0020] The bidirectional causal temporal convolutional network (BiC-TCN) uses an expanded convolution kernel to extract the long-term dependency between environmental parameter time series and gene expression data, and calculates the impact weights of environmental events on the expression pathways of the heat shock protein HSP70 and the pathogenesis-related protein PR1 through counterfactual intervention. It should be further explained that the input layer of the BiC-TCN network is a 72-hour environmental time series, including temperature, humidity, and CO2. The hidden layer uses a causal convolution kernel with an expansion factor of 2∧0 to 2∧5, and the output layer is connected to the gene expression level, that is, the HSP70 mRNA level measured by qPCR. Among them, the counterfactual intervention inference process includes: assuming that a certain temperature mutation event is removed, by comparing the actual gene expression with the simulated data, the contribution of the environmental impact is calculated. For example, the high temperature event contributes to the upregulation of HSP70 by 42%.

[0021] The three-dimensional phenotype-genome association analysis unit reconstructs the tuber point cloud model based on structured light three-dimensional scanning, extracts curvature entropy and volume growth rate phenotypic parameters, and performs genome-wide association analysis (GWAS) with SNP marker data to locate QTL sites related to stress resistance.

[0022] Preferably, the three-dimensional phenotype-genome association analysis unit performs:

[0023] The curvature entropy of the tuber point cloud was calculated, and the surface morphological complexity was quantified using the Gaussian curvature differential algorithm. Tubers with a curvature entropy greater than 0.35 were selected as candidate samples for stress resistance.

[0024] Dynamic growth rate modeling, fitting the logistic growth curve based on time series point cloud data, extracting the maximum growth rate Vmax and the inflection point date Tinflection;

[0025] The stress resistance QTL mapping engine inputs the curvature entropy and Vmax phenotypic parameters into a mixed linear model MLM, and combines the SolCAP chip SNP data to identify the StCIPK23 gene on chromosome 5 and the StWRKY45 gene on chromosome 8 as key resistance loci.

[0026] Preferably, the degradation warning and control module includes:

[0027] A multimodal risk index MRI calculation engine, configured to fuse the probability of latent disease, environmental stress index, and genetic vulnerability score, and dynamically generate risk levels through fuzzy cognitive map (FCM);

[0028] The adaptive control strategy generator, based on the Deep Deterministic Policy Gradient (DDPG) algorithm, optimizes greenhouse equipment control parameters based on real-time MRI values, dynamically adjusting the lighting duration and temperature control thresholds during the tuber expansion phase. It should be noted that the DDPG algorithm's state space includes MRI values ​​and tuber growth stages (with tuber diameter > 20 mm as the indicator for the expansion phase), and its action space includes the lighting on / off duration (0-24 hours) and the temperature setpoint (15-25°C).

[0029] Preferably, the MRI computing engine performs:

[0030] Environmental pressure index calculation, mapping the contribution of environmental events output by BiC-TCN into a 0-1 standardized score;

[0031] Gene vulnerability score, based on the genotype of QTL loci mapped by 3D-PGAS, calculates the variety-specific disease resistance attenuation coefficient;

[0032] Risk level decisions: When the probability of disease during the incubation period is >0.6, the environmental stress index is >0.5, and the genetic vulnerability score is >0.7, a high-risk warning is triggered and targeted regulation is initiated. It should be further explained that the MRI decision threshold was determined through receiver operating characteristic (ROC) curve analysis. When the probability of disease during the incubation period is >0.6, it corresponds to a sensitivity of 82% and a specificity of 76%. An environmental stress index >0.5 corresponds to a cumulative temperature stress of >150°C·h. A genetic vulnerability score >0.7 indicates at least two high-risk QTL loci.

[0033] Preferably, the microfluidic metabolomics sampling unit comprises:

[0034] The in situ exudate capture chip integrates a nanoporous membrane and a microfluidic structure to collect leaf exudate at a flow rate of 5 μL / min;

[0035] The SERS enhanced substrate uses a gold-titanium dioxide core-shell nanostructure to improve the detection sensitivity of the characteristic peaks of salicylic acid SA and jasmonic acid JA. -1 , characteristic peak of jasmonic acid at 1650 cm -1 ;

[0036] The metabolism-spectral spatiotemporal alignment module matches the SERS detection results with the multispectral imaging data according to the acquisition timestamp and spatial coordinates to construct a metabolite-spectral response association database.

[0037] It should be further explained that the gold-titania core-shell structure of the SERS substrate was prepared by chemical vapor deposition. The gold layer thickness was 5nm and the titanium dioxide layer was 10nm. The detection limit of SA reached 0.1μM and the JA was 0.05μM. Specifically, the SERS substrate synthesis process was as follows: HAuCl4 chloroauric acid and tetrabutyl titanate were mixed in a molar ratio of 1:3, sodium citrate was added as a reducing agent, and the mixture was reacted in an 80°C water bath for 2 hours. After centrifugation, gold-titania core-shell nanoparticles with a particle size of 50±5nm were obtained. Subsequently, the mixture was calcined in a tube furnace under argon protection and kept at 300°C for 1 hour to form a crystalline titanium dioxide coating with a thickness of 10±2nm.

[0038] A monitoring and analysis method for potato breeding comprises the following steps:

[0039] S1. Synchronously acquire environmental parameters, multispectral images, and metabolite concentration data through a multimodal data acquisition module;

[0040] S2. Use the spatiotemporal causal modeling module to construct the BiC-TCN network and 3D-PGAS model to analyze the dynamic association between environment, gene and phenotype;

[0041] S3. Generate an MRI risk index based on the degradation warning control module and trigger an adaptive environment control instruction;

[0042] S4. The model prediction accuracy was verified through double-blind closed-loop validation and gene editing reverse experiments.

[0043] Preferably, the step S2 further comprises:

[0044] S21. A bidirectional causal temporal convolutional network (BiC-TCN) was used to extract the temporal causal relationship between environmental parameters and gene expression, and to calculate the probability of upregulation of HSP70 gene expression by continuous high temperature.

[0045] S22. Extracting tuber curvature entropy and volume growth rate based on structured light 3D scanning data, and mapping stress resistance QTLs through genome-wide association analysis.

[0046] S23. Through counterfactual intervention simulation, quantify the contribution of abnormal environmental events to seed potato degradation and generate an attribution analysis report.

[0047] Preferably, step S4 includes:

[0048] S41. We simulated eight temperature-humidity combinations in an artificial climate chamber and used a Bayesian network to infer the pathways by which environmental parameters influence gene expression, verifying that the causal model error was <8%.

[0049] S42. Use CRISPR-Cas9 to knock out the StWRKY45 gene and compare the phenotypic parameters of the knockout strain with those of the wild-type strain to verify the biological significance of the 3D-PGAS model (p < 0.05).

[0050] S43. Visualize MRI decision-making evidence through Layer-wise Relevance Propagation (LRP) to ensure that warning results conform to agronomic empirical rules.

[0051] (3) Beneficial effects

[0052] The present invention provides a monitoring and analysis system and method for potato breeding. It has the following beneficial effects:

[0053] (I) The monitoring and analysis system and method used in potato breeding significantly improves the accuracy and timeliness of potato breeding monitoring through multimodal data fusion and spatiotemporal causal modeling. First, the system integrates dual-band multispectral imaging, microfluidic metabolic detection, and three-dimensional phenotypic analysis, breaking through the limitations of traditional visible light imaging and achieving an improvement in the accuracy of latent disease identification compared to traditional methods. Based on structured light scanning and curvature entropy calculation, it defines three-dimensional phenotypic parameters of stress resistance, reducing the error in tuber morphology quantification. Secondly, through a bidirectional causal temporal convolutional network and counterfactual intervention deduction, it analyzes the dynamic association between environmental fluctuations and gene expression, quantifies the contribution of high temperature events to the upregulation of heat shock protein HSP70 expression, and combines genome-wide association analysis to locate the key stress resistance gene StWRKY45, providing a molecular mechanism explanation for degradation attribution.

[0054] (2) The monitoring and analysis system and method used for potato breeding can identify degradation risks in advance and dynamically control greenhouse parameters in combination with the DDPG reinforcement learning algorithm to improve starch accumulation efficiency; through gene editing reverse verification and double-blind field trials, the reliability of the model prediction was confirmed, and the virus detection rate in the experimental group was reduced and the yield was increased; by constructing a three-dimensional "phenotype-environment-genotype" database, data-driven decision support is provided for the selection and breeding of stress-resistant varieties, reducing the cost of blind seed adjustment, and promoting the transformation of potato breeding from experience-based to intelligent and precise. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the overall structure / framework of the present invention;

[0056] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] See also Figure 1 and Figure 2 The present invention provides a technical solution: a monitoring and analysis system for potato breeding, comprising:

[0059] a multimodal data acquisition module configured to simultaneously acquire greenhouse environmental parameters, potato plant multispectral imaging data, and leaf metabolite concentration data;

[0060] The spatiotemporal causal modeling module, connected to the multimodal data acquisition module, is used to construct a dynamic association model between environmental fluctuations, gene expression, and phenotypic characteristics, and to quantify the genetic effects of environmental anomalies based on counterfactual interventions;

[0061] The degradation warning and control module is connected to the spatiotemporal causal modeling module to generate a multimodal risk index and trigger adaptive environmental control instructions;

[0062] The data storage and interaction module is used to store multi-source heterogeneous data and model parameters and provides a cross-platform data interface. It should be further explained that the data storage and interaction module uses MongoDB to store unstructured data, including spectral images and point cloud models. Time series data is stored in InfluxDB. It provides a RESTful API interface and supports third-party devices such as drip irrigation systems to access MRI warning data.

[0063] Among them, the multimodal data acquisition module includes:

[0064] The distributed environmental sensor array integrates temperature, humidity, light intensity, and CO2 concentration sensors, and uploads data every ten minutes. It should be further explained that the multimodal data acquisition module collects greenhouse environmental parameters in real time through the distributed environmental sensor array. The specific configuration is as follows:

[0065] The temperature and humidity sensor uses the SHT35 chip, with a measurement range of -40 to 125°C and an accuracy of ±0.2°C. The data is transmitted to the central server via LoRa wireless every 10 minutes.

[0066] The light sensor uses the BH1750FVI chip with a detection range of 0 to 65535 lx, dynamically tracking canopy photosynthetically active radiation PAR;

[0067] The CO2 concentration sensor used is SenseAir K30, with a range of 0 to 10,000 ppm and a resolution of ±50 ppm, and is used to monitor the intensity of respiration.

[0068] A dual-band excitation multispectral imaging device is configured to simultaneously scan the plant canopy in the near-infrared and short-wave infrared bands and extract stomatal density, secondary metabolite distribution, and pathogen characteristic spectra;

[0069] The microfluidic metabolome sampling unit uses surface-enhanced Raman spectroscopy (SERS) to detect the concentrations of salicylic acid (SA) and jasmonic acid (JA) in leaf exudates in real time.

[0070] The dual-band excitation multispectral imaging device further comprises:

[0071] The laser-induced fluorescence excitation module uses a laser light source with a wavelength of 405 nm: a laser diode with a power of 50 mW. The 405 nm laser is collimated through a lens to form a 5 cm diameter spot. The incident angle is 60° to avoid interference from mirror reflections, which stimulates leaf autofluorescence. The fluorescence signal is separated by a dichroic mirror (cut-off wavelength 450 nm) and collected by an EMCCD camera (Andor iXon Ultra 888) with an exposure time of 200 ms. The bandpass filter is 500-600 nm to capture the distribution characteristics of flavonoids. The flavonoid distribution map is extracted by a spectral unmixing algorithm (non-negative matrix decomposition) with a spatial resolution of 50 μm.

[0072] The spectroscopic filter wheel is configured to switch between near-infrared and short-wave infrared filters in a single scan and acquire dual-channel spectral images using an InGaAs sensor. The filters are mounted on a 15cm diameter aluminum alloy turntable and driven by a stepper motor with a 0.9° step angle, resulting in a switching time of less than 0.3 seconds. The near-infrared channel uses an InGaAs sensor with a response wavelength of 900 to 1700nm, while the short-wave infrared channel uses a HgCdTe detector with a response wavelength of 1700 to 2500nm, cooled to -40°C to reduce noise. The dual-channel spectral images, or dual-band images, are registered using feature point matching (SIFT algorithm) to correct for displacement errors caused by mechanical switching, achieving a registration accuracy of 0.5 pixels.

[0073] It is important to note that the filter wheel switching control logic uses an STM32F407 controller to drive a stepper motor (28BYJ-48) with a pulse frequency set to 500 Hz. A photoelectric encoder provides real-time feedback on the filter position, with a positioning accuracy of ±0.1° and a switching time of <0.3 seconds. Spectral image registration uses a modified SIFT algorithm, with a feature point matching distance ratio threshold of 0.6 and 1000 RANSAC iterations. Anomalous matching points with a projection error greater than 1 pixel are eliminated.

[0074] The spectral feature fusion unit performs noise suppression and feature alignment on the dual-band image based on the Generative Adversarial Network (GAN), and outputs a latent disease probability map. It should be further explained that the GAN model training for spectral feature fusion is as follows:

[0075] The generator input is a dual-band image (256×256 pixels) and outputs a fused image after noise suppression;

[0076] The discriminator uses the PatchGAN structure to judge the authenticity of the image patch (70×70 pixels);

[0077] The loss function includes pixel-level L1 loss (weight 0.7) and adversarial loss (weight 0.3), with a training cycle of 200 epochs and a batch size of 16.

[0078] The spatiotemporal causal modeling module includes:

[0079] The bidirectional causal temporal convolutional network (BiC-TCN) uses a dilated convolution kernel to extract the long-term dependency between environmental parameter time series and gene expression data, and calculates the impact weights of environmental events on the expression pathways of heat shock protein HSP70 and pathogenesis-related protein PR1 through counterfactual intervention. It should be further explained that the specific parameters of the BiC-TCN network (210) are:

[0080] Input layer: 72 hours of environmental data (temperature, humidity, CO2) normalized to Z-score;

[0081] Hidden layer: 6 layers of dilated causal convolution with dilation factors of 1, 2, 4, 8, 16, and 32, and 64 filters per layer;

[0082] Output layer: The fully connected layer is mapped to the gene expression level (log2 transformed value of HSP70 and PR1) and optimized using the Huber loss function.

[0083] Counterfactual intervention calculations:

[0084] Construct a counterfactual series: replace the periods with actual temperature data > 30°C with simulated data of 25°C;

[0085] BiC-TCN was used to predict gene expression levels under counterfactual conditions, and the difference ratio between actual and counterfactual conditions was calculated, such as: HSP70 difference ratio = Δactual / Δcounterfactual × 100%.

[0086] The 3D phenotype-genome association analysis unit reconstructs a tuber point cloud model based on structured light 3D scanning, extracts curvature entropy and volume growth rate phenotypic parameters, and performs genome-wide association analysis (GWAS) with SNP marker data to locate QTL loci associated with stress resistance. It should be further clarified that during the use of the 3D phenotype-genome association analysis unit, the following tasks are included:

[0087] Structured light scanning: A binocular camera (Basler acA2440) is used in conjunction with a DLP4500 projector to generate phase-coded fringes, with a point cloud reconstruction accuracy of ±0.05mm.

[0088] Curvature entropy calculation: Gaussian curvature is calculated for the vertices of the triangle mesh on the tuber surface. The entropy formula is H = -Σ(p i logp i ), where p i is the probability distribution of the curvature interval;

[0089] Logistic modeling: tuber volume V(t)=K / (1+e∧(-r(t-t0))), fitting parameters K: maximum volume, r: growth rate, t0: inflection point date.

[0090] 3D Phenotype-Genome Association Analysis Unit performs:

[0091] The curvature entropy of the tuber point cloud was calculated, and the surface morphological complexity was quantified using the Gaussian curvature differential algorithm. Tubers with a curvature entropy greater than 0.35 were selected as candidate samples for stress resistance.

[0092] Dynamic growth rate modeling, fitting the logistic growth curve based on time series point cloud data, extracting the maximum growth rate Vmax and the inflection point date Tinflection;

[0093] The stress resistance QTL mapping engine inputs the curvature entropy and Vmax phenotypic parameters into the mixed linear model MLM, combined with the SolCAP chip SNP data, and identifies the StCIPK23 gene on chromosome 5 and the StWRKY45 gene on chromosome 8 as key resistance loci.

[0094] It should be further explained that the specific steps of three-dimensional phenotyping analysis include:

[0095] S41. QTL mapping for stress tolerance:

[0096] S411. Phenotypic data standardization: Curvature entropy was Box-Cox transformed, and Vmax was logarithmically transformed;

[0097] S412. Mixed linear model: y = Xβ + Zu + ε, where X is the fixed effect (SNP genotype), Z is the random effect (block design), and ε is the residual;

[0098] S413. Significance threshold: P < 1 × 10 after Bonferroni correction -5 , and the StCIPK23 (Chr5: 32,458,721) and StWRKY45 (Chr8: 15,237,904) sites were screened out.

[0099] S42. Genotype-phenotype verification experiments:

[0100] S421. 200 potato plants (100 wild-type StWRKY45 plants and 100 mutant plants) were grown in an artificial climate chamber.

[0101] S422. Stress treatment: 35℃ during the day and 25℃ at night for 7 days, and the changes in tuber curvature entropy were detected (the mutant type increased by an average of 0.12, and the wild type increased by 0.05, p < 0.01).

[0102] The degradation warning and control module includes:

[0103] The multimodal risk index MRI calculation engine is configured to fuse the latent disease probability, environmental stress index, and genetic vulnerability score, and dynamically generate risk levels through the fuzzy cognitive map FCM. It should be further explained that the latent disease probability is a 0-1 probability value output by the spectral feature fusion unit, the environmental stress index is a standardized score of 0-1 output by BiC-TCN, and the genetic vulnerability score is based on the QTL locus genotype, such as the StWRKY45 mutant score of 0.8.

[0104] The adaptive control strategy generator, based on the deep deterministic policy gradient (DDPG) algorithm, optimizes greenhouse equipment control parameters according to real-time MRI values ​​and dynamically adjusts the lighting duration and temperature control threshold during the tuber bulking period.

[0105] It should be further explained that the operation logic of the degradation warning module includes the following steps:

[0106] S51.MRI Computation Engine:

[0107] S511. Environmental stress index = Σ(event contribution × duration weight). For example, if the temperature is >30°C for three consecutive days, the contribution is 0.6 and the weight is 1.2.

[0108] S512. Gene vulnerability score = 1-Π(1-QTL risk value). For example, if the risk value of StCIPK23 is 0.3 and the risk value of StWRKY45 is 0.5, then the score = 1-(0.7×0.5) = 0.65.

[0109] S513. The Fuzzy Cognitive Map (FCM) rule base contains 27 IF-THEN rules, such as:

[0110] IF disease probability is high AND environmental stress is medium AND genetic vulnerability is high THEN risk level = red.

[0111] S52.DDPG regulation strategy:

[0112] S521. State space: MRI value (0-1), tuber growth stage (1-5), historical regulation record;

[0113] S522. Action space: fill light on duration (discrete to 0 / 6 / 12 / 18 hours), temperature setting value (discrete to 18 / 20 / 22 / 24°C);

[0114] S523. Reward function: R = 0.5×Δ starch content + 0.3×Δ number of healthy tubers - 0.2× energy consumption cost.

[0115] 06Preferably, the MRI calculation engine performs:

[0116] Environmental pressure index calculation, mapping the contribution of environmental events output by BiC-TCN into a 0-1 standardized score;

[0117] Gene vulnerability score, based on the genotype of QTL loci mapped by 3D-PGAS, calculates the variety-specific disease resistance attenuation coefficient;

[0118] Risk level decision: when the probability of latent disease is >0.6, the environmental stress index is >0.5, and the genetic vulnerability score is >0.7, a high-risk warning is triggered and targeted regulation is initiated.

[0119] The microfluidic metabolomics sampling unit includes:

[0120] The in situ exudate capture chip integrates a nanoporous membrane and a microfluidic structure. The nanoporous membrane with a pore size of 200 nm is covered on the lower epidermis of potato leaves, and the exudate is collected at a flow rate of 5 μL / min using a negative pressure pump.

[0121] The SERS enhancement substrate uses a silicon wafer modified with gold-titanium dioxide core-shell nanoparticles with a particle size of 50 nm as the substrate, with a laser wavelength of 785 nm, a power of 20 mW, and an integration time of 10 seconds;

[0122] The metabolism-spectral spatiotemporal alignment module synchronizes multi-light imaging and SERS detection through GPS timestamps, with a spatial coordinate matching error of <1mm. The SERS detection results are matched with the multispectral imaging data according to the acquisition timestamp and spatial coordinates to construct a metabolite-spectral response association database.

[0123] A monitoring and analysis method for potato breeding comprises the following steps:

[0124] S1. Synchronously acquire environmental parameters, multispectral images, and metabolite concentration data through a multimodal data acquisition module;

[0125] S2. Use the spatiotemporal causal modeling module to construct the BiC-TCN network and 3D-PGAS model to analyze the dynamic association between environment, gene and phenotype;

[0126] S3. Generate an MRI risk index based on the degradation warning control module and trigger an adaptive environment control instruction;

[0127] S4. The model prediction accuracy was verified through double-blind closed-loop validation and gene editing reverse experiments.

[0128] It should be further explained that this method achieves potato species degradation monitoring through four stages: synchronous multimodal data collection, spatiotemporal causal modeling and analysis, degradation early warning and regulation, and verification experiments. The specific steps include the following:

[0129] S01. During the data collection phase, a distributed environmental sensor array was deployed to obtain real-time greenhouse environmental parameters: the temperature and humidity sensor used an SHT35 chip with a measurement range of -40 to 125°C and an accuracy of ±0.2%, transmitting data every 10 minutes via LoRa wireless technology. The light sensor used a BH1750FVI chip with a measurement range of 0 to 65,535 lx and dynamic tracking of photosynthetically active radiation. The CO2 concentration sensor used a SenseAir K30 with a range of 0 to 10,000 ppm and a resolution of ±50 ppm, monitoring respiration intensity in real time. All data was time-stamped and synchronized using a GPS timing module, with a time deviation of <1 second. The dual-band excitation multispectral imaging device uses a 405nm laser to induce leaf fluorescence and capture the distribution of flavonoids. At the same time, the near-infrared (900-1700nm, germanium-based filter) and short-wave infrared (1700-2500nm, zinc sulfide filter) channels are switched through the spectroscopic filter wheel. InGaAs and HgCdTe detectors are used to collect dual-band images at a resolution of 0.1mm / pixel. After fusion and denoising by the generative adversarial network (GAN), a latent disease probability map is generated. The microfluidic metabolomics sampling unit collects leaf exudate in situ at a flow rate of 5μL / min through a nanoporous membrane with a pore size of 200nm for 30 minutes; a gold-titanium dioxide core-shell nanosubstrate with a 5nm gold layer and a 10nm titanium dioxide layer is used to enhance the Raman signal and detect salicylic acid (SA, 1580cm -1 ) and jasmonic acid (JA, 1650 cm -1 ) concentration, and achieves spatiotemporal alignment with multispectral data through high-precision robotic arm positioning, with spatial error <1mm and time deviation <50ms.

[0130] In the spatiotemporal causal modeling phase, a bidirectional causal temporal convolutional network (BiC-TCN) was constructed. 72 hours of standardized environmental data (temperature, humidity, and CO2) were input with a sliding step of 1 hour. Six layers of dilated causal convolution (with dilation factors of 1 to 32 and 64 filters per layer) were used to analyze the long-range dependencies between environmental parameters and gene expression. Counterfactual interventions were used to quantify the contribution of abnormal events, such as the 42% increase in HSP70 expression caused by continuous high temperatures. Gene expression data were quantified by qPCR for HSP70 and PR1 mRNA levels and normalized after log2 transformation.

[0131] At the same time, the tuber 3D point cloud model was reconstructed based on structured light scanning (binocular camera + DLP4500 projector), the curvature entropy was calculated (Gaussian curvature distribution entropy value > 0.35 is a sign of stress resistance), and the maximum growth rate V was extracted by fitting the Logistic growth curve. max Through genome-wide association analysis (GWAS) combined with SolCAP chip SNP data, stress resistance QTL loci such as StCIPK23 (Chr5: 32,458,721) and StWRKY45 (Chr8: 15,237,904) were located.

[0132] Among them, curvature entropy calculation: calculate the Gaussian curvature of the triangle mesh vertices on the point cloud surface, and the entropy value formula is H = -∑(p i logp i ) Screening tubers with curvature entropy > 0.35 as stress resistance candidates;

[0133] Growth rate modeling: Fitting the Logistic curve V(t) = K / 1 + e based on time series point cloud data -r(t-t0) , extract the maximum growth rate V max With the inflection point date t0.

[0134] Genome-wide association studies (GWAS) include:

[0135] Phenotypic data: Normalized curvature entropy and V max .

[0136] Genotype data: SNP markers were obtained using SolCAP chips, and sites with MAF < 0.05 were filtered.

[0137] Mixed linear model (MLM): y = Xβ + Zu + ε, where X is the SNP genotype and Z is the block random effect. The FarmCPU algorithm was used for optimization, and the significance threshold was P < 1 × 10 -5 .

[0138] Positioning results: StCIPK23 (Chr5: 32,458,721) and StWRKY45 (Chr8: 15,237,904) were identified as key stress resistance QTL loci.

[0139] S03. During the degradation warning and regulation phase, the Multimodal Risk Index (MRI) engine integrates the latent disease probability, the environmental stress index (a standardized score output by BiC-TCN), and the genetic vulnerability score (calculated based on QTL genotypes) to generate a risk level through 27 IF-THEN rules of the Fuzzy Cognitive Map (FCM) (e.g., a red warning is triggered when the disease probability is >0.6 and the environmental stress is >0.5). The adaptive control strategy uses a deep deterministic policy gradient (DDPG) algorithm. The state space includes MRI values, tuber growth periods (1-5), and historical control records. The action space sets the lighting duration (0 / 6 / 12 / 18 hours) and temperature setting values ​​(18 / 20 / 22 / 24°C). The reward function integrates the improvement of starch content (weight 0.5), the increase in the number of healthy tubers (0.3), and the energy cost (-0.2). When the MRI risk is red, night lighting (18:00-24:00) and 20°C constant temperature control are implemented. The DDPG policy network convergence condition is defined as the average reward fluctuation amplitude is less than 5% over 10 consecutive training cycles. The initial exploration noise is an OU process, where θ = 0.15, σ = 0.2, and the noise amplitude is attenuated by 10% every 1000 steps. It further includes:

[0140] S031. Multimodal Risk Index MRI Calculation:

[0141] Input data:

[0142] Latent disease probability: derived from spectral feature fusion unit (0-1).

[0143] Environmental stress index: a normalized score (0-1) output by BiC-TCN.

[0144] Gene vulnerability score: calculated based on the genotype of the QTL locus, such as the StWRKY45 mutant score is 0.8.

[0145] Fuzzy cognitive map FCM decision:

[0146] The rule base contains 27 IF-THEN rules, for example: IF disease probability > 0.6 AND environmental pressure > 0.5 AND genetic vulnerability > 0.7 THEN risk level = red.

[0147] The MRI risk level (low / intermediate / high) was output, and the trigger threshold was determined by ROC curve analysis, with a sensitivity of 82% and a specificity of 76%.

[0148] S032. Adaptive control instruction generation:

[0149] DDPG algorithm parameters:

[0150] State space: MRI value (0-1), tuber growth period (1-5), historical regulation records.

[0151] Action space: fill light duration (0 / 6 / 12 / 18 hours), temperature setting value (18 / 20 / 22 / 24℃).

[0152] Reward function: R = 0.5 × Δ starch content + 0.3 × Δ number of healthy tubers - 0.2 × energy cost.

[0153] Control implementation: When the MRI risk level is red, start night lighting (18:00-24:00) and set the temperature to 20℃ for 3 days.

[0154] S04. During the validation phase, double-blind closed-loop experiments and gene editing reverse validation were used to ensure the reliability of the model: the system was deployed in temperate, subtropical, and plateau regions, with the experimental group (adaptive regulation) containing 50 planting units compared to the control group (fixed strategy). Data from three consecutive growing seasons showed that the virus detection rate in the experimental group decreased by 32% (p = 0.003), the yield increased by 19%, and the starch content increased by 15%. The StWRKY45 gene was knocked out using CRISPR-Cas9. The increase in tuber curvature entropy of the knockout plant under high temperature stress was 0.12, which was significantly higher than that of the wild-type plant (0.05), with p < 0.01, verifying the accuracy of the 3D-PGAS model. Layer-wise correlation propagation (LRP) visualization showed that high temperature stress accounted for 58% of MRI decisions, which is consistent with agronomic experience, achieving closed-loop management of the entire chain from data collection to regulatory decision-making, and warning of seed potato degradation risks 14-21 days earlier than traditional methods.

[0155] Through the simultaneous collection of multimodal data, causal modeling to analyze environment-gene-phenotype associations, dynamic risk warning and regulation, and strict double-blind verification, the core problems of insufficient latent disease detection and lack of environment-gene temporal associations in traditional technologies have been systematically solved. This solution realizes a full-chain closed loop from data collection to regulatory decision-making, and can warn of degradation risks 14-21 days earlier than traditional methods, providing a data-driven precision solution for potato stress-resistant breeding.

[0156] Step S2 uses a bidirectional causal temporal convolutional network (BiC-TCN), three-dimensional phenotypic analysis, and counterfactual intervention inference to achieve in-depth analysis of the dynamic relationship between environment, gene, and phenotype, and further includes:

[0157] S21. A bidirectional causal temporal convolutional network (BiC-TCN) was used to extract the temporal causal relationship between environmental parameters and gene expression and to calculate the probability of upregulation of HSP70 gene expression by continuous high temperature. It should be further explained that, in the specific implementation process, S21: BiC-TCN network construction and quantification of high temperature effects include the following steps:

[0158] S211. Data preprocessing:

[0159] Environmental parameter standardization: Environmental data (temperature, humidity, CO2 concentration) were collected for 72 consecutive hours and Z-score normalized according to the time window (each hour is a time step) to eliminate dimensional differences.

[0160] Quantification of gene expression: The mRNA expression levels of heat shock protein HSP70 and pathogenesis-related protein PR1 were measured by qPCR. The data were log2 transformed and normalized to the range of [0,1].

[0161] S212.BiC-TCN network architecture:

[0162] Input layer: Receives a 72-dimensional sequence of environmental parameters (temperature, humidity, CO2), with 3 features corresponding to each time step.

[0163] Causal convolution layer: 6 layers of dilated causal convolution are used, with dilation factors of 1, 2, 4, 8, 16, and 32, respectively. Each layer contains 64 filters and the activation function is ReLU to ensure that the model captures long-range dependencies.

[0164] Output layer: The fully connected layer maps the hidden layer output to the gene expression level (1 output node each for HSP70 and PR1). The loss function uses Huber loss, the optimizer is Adam, the learning rate is 0.001, and the batch size is 32.

[0165] S213. Temporal Causal Relationship Extraction:

[0166] High temperature event identification: define a high temperature event as a maximum daily temperature ≥ 30°C for three consecutive days, and mark its time interval.

[0167] Gene expression response analysis: Actual environmental data were input into BiC-TCN to predict HSP70 expression levels. The sliding window method was used to calculate the expression difference between high-temperature and non-high-temperature periods. For example, the mean HSP70 value during the high-temperature period was 1.8 times higher than the baseline.

[0168] Calculation of up-regulation probability: Monte Carlo simulation was used to randomly perturb the temperature sequence (±2°C) and calculate the probability that the HSP70 expression level exceeded the threshold. For example, when the temperature was >30°C, the up-regulation probability was 78%.

[0169] S22. Extract tuber curvature entropy and volume growth rate based on structured light 3D scanning data, and locate stress resistance QTL loci through genome-wide association analysis. It should be further explained that, in the specific implementation process, S22: 3D phenotypic analysis and stress resistance QTL mapping is subdivided into the following steps:

[0170] S221. 3D phenotyping data collection:

[0171] Structured light scanning system: Using a binocular camera, Basler acA2440, with a resolution of 2448×2048 and a DLP4500 projector, phase-coded fringes are generated and a three-dimensional point cloud model of the tuber is reconstructed through a phase unwrapping algorithm with an accuracy of ±0.05mm.

[0172] Curvature entropy calculation:

[0173] Gaussian curvature extraction: Calculate the Gaussian curvature of each vertex of the triangular mesh on the point cloud surface. The formula is where θ i is the interior angle of the neighborhood triangle of the vertex, and A is the neighborhood area.

[0174] Entropy quantification: divide the curvature value into 10 intervals, statistically analyze the probability distribution and calculate the entropy value Tubers with a curvature entropy greater than 0.35 were selected as candidates for high stress resistance. Growth rate modeling: Based on the tuber volume data for 7 consecutive days, a logistic curve was fitted. Extracting maximum growth rate And the turning point date t0.

[0175] S222. Genome-wide association study (GWAS):

[0176] Genotype data acquisition: 200 potato samples were subjected to SNP typing using the SolCAP array. Sites with a minor allele frequency (MAF) < 0.05 and a missingness rate > 10% were filtered out, retaining 5 sites and 342 high-quality SNPs.

[0177] Mixed linear model MLM:

[0178] Model formula: y = Xβ + Zu + ε, where X is the SNP genotype (0 / 1 / 2 coding), Z is the block random effect, and u follows a normal distribution K is the kinship matrix.

[0179] The FarmCPU algorithm was used for optimization, with 20 iterations and a significance threshold of P < 1 × 10 after Bonferroni correction. -5 .

[0180] QTL mapping results: StCIPK23 gene located on chromosome 5 was identified at Chr5: 32,458,721, P = 3.2 × 10 -6 , and the StWRKY45 gene on chromosome 8, Chr8: 15,237,904, P = 7.8 × 10 -6 , both of which are related to the curvature entropy and V max Significantly correlated.

[0181] S23. Through counterfactual intervention deduction, quantify the contribution of abnormal environmental events to seed potato degradation and generate an attribution analysis report. It should be further explained that in the specific implementation process, S23: Counterfactual intervention deduction and attribution analysis includes the following steps:

[0182] S231. Counterfactual sequence construction:

[0183] Event removal strategy: Generate counterfactual sequences for abnormal events in the actual environmental data, such as continuous high temperatures or sudden drops in humidity. For example, replace periods of temperature > 30°C in the actual temperature data with simulated data of 25°C, keeping other parameters unchanged.

[0184] Data interpolation: Cubic spline interpolation is used to smooth the mutation boundaries of the counterfactual sequence to avoid artifact interference.

[0185] Among them, the counterfactual sequence generation formula is: based on the structural equation model E t =f(E t-1 ,G t )+e, where E t is the environmental parameter, G t is the gene expression level, and e is the noise term. t ′ is generated by conditional intervention:

[0186] Among them, E sim is the cubic spline interpolation result based on historical data.

[0187] S232. Gene expression prediction and differential calculation:

[0188] Counterfactual prediction: The counterfactual environment sequence is input into the trained BiC-TCN network to predict the expression levels of HSP70 and PR1.

[0189] Contribution quantification: Calculate the difference rate of gene expression between actual and counterfactual conditions, the formula is: Contribution = (actual expression - counterfactual expression) / actual expression × 100%;

[0190] For example, continuous high temperature events make the actual value of HSP70 expression 2.1, and the counterfactual value is 1.5, then the contribution is

[0191] S233. Attribution analysis report generation:

[0192] Sorting of key events: Arrange environmental abnormal events in descending order of contribution, and screen the top 20% high-impact events, including high temperature contribution >25% and humidity fluctuation >15%.

[0193] Biological interpretation: Combined with QTL mapping results, the interaction mechanism between environmental events and stress resistance genes was analyzed. For example, high temperature increases tuber epidermal thickness by activating StWRKY45 expression, thereby increasing curvature entropy.

[0194] Visualization output: Generate a heat map to display the environmental event-gene-phenotype association network, and use a Sankey diagram to quantify the contribution weight of each path.

[0195] In an independent test set (50 sets of environmental-gene data), the prediction error MAE of HSP70 expression was 0.12, which was 37% lower than that of the traditional LSTM model. The correlation coefficient R between the predicted results of the high temperature event on the probability of HSP70 upregulation and the RT-qPCR measured data was 2 =0.83.

[0196] Tubers with a curvature entropy greater than 0.35 exhibited significant disease resistance in field trials, with a 41% reduction in virus infection (p < 0.01). The StWRKY45 knockout strain exhibited a 58% reduction in the increase in curvature entropy compared to the wild-type strain under high temperature (p = 0.004), negatively validating the QTL's functionality.

[0197] After removing the continuous high temperature events from the simulation, the model predicted a 19% increase in tuber starch content, which matched the actual control experiment result of 17% at 89%. The key events listed in the attribution report were 92% consistent with the agronomist's experience, with a Kappa coefficient of 0.85.

[0198] By integrating dilated convolution with counterfactual intervention in BiC-TCN, the limitations of traditional statistical models in long-range dependency analysis are overcome, and the dynamic impact of environmental events on gene expression is accurately quantified. By combining structured light scanning with curvature entropy calculation, new stress resistance phenotypic parameters are defined, achieving a leap from two-dimensional morphology to three-dimensional functional traits. Through gene editing reverse experiments and double-blind field trials, the model prediction results are ensured to have both algorithmic reliability and biological significance, avoiding the risk of unexplainability of purely data-driven models. This solution systematically solves the core problems of "missing environment-gene temporal association", "phenotype-genotype data fragmentation" and inaccurate degradation attribution, providing a full-chain solution for potato stress resistance breeding from data collection to causal intervention.

[0199] It should be further explained that, during its specific implementation, this technical solution significantly improved the accuracy and timeliness of potato breeding monitoring through multimodal data fusion and spatiotemporal causal modeling. First, the system integrates dual-band multispectral imaging, microfluidic metabolic detection, and three-dimensional phenotypic analysis, breaking through the limitations of traditional visible light imaging and achieving an improvement in the accuracy of latent disease identification compared to traditional methods. Based on structured light scanning and curvature entropy calculation, it defines three-dimensional phenotypic parameters for stress resistance, reducing the error in tuber morphology quantification. Secondly, through a bidirectional causal temporal convolutional network and counterfactual intervention deduction, it analyzes the dynamic association between environmental fluctuations and gene expression, quantifies the contribution of high temperature events to the upregulation of heat shock protein HSP70 expression, and combines genome-wide association analysis to locate the key stress resistance gene StWRKY45, providing a molecular mechanism explanation for degradation attribution.

[0200] This solution has achieved three major breakthroughs at the application level: First, the multimodal risk index early warning system can identify degradation risks 14-21 days in advance, and combined with the DDPG reinforcement learning algorithm to dynamically control greenhouse parameters, thereby improving starch accumulation efficiency; second, through gene editing reverse verification (such as CRISPR knockout of StWRKY45) and double-blind field trials, the reliability of model predictions was confirmed, and the virus detection rate in the experimental group was reduced and the yield was increased; third, a three-dimensional "phenotype-environment-genotype" database was constructed to provide data-driven decision-making support for the selection and breeding of stress-resistant varieties, reduce the cost of blind planting adjustments, and promote the transformation of potato breeding from experience-based to intelligent and precise.

[0201] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0202] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A monitoring and analysis system for potato breeding, characterized in that: include: a multimodal data acquisition module configured to simultaneously acquire greenhouse environmental parameters, potato plant multispectral imaging data, and leaf metabolite concentration data; a spatiotemporal causal modeling module, connected to the multimodal data acquisition module, for constructing a dynamic association model between environmental fluctuations, gene expression, and phenotypic characteristics, and quantifying the genetic effects of environmental anomalies based on counterfactual interventions; a degradation warning and control module, connected to the spatiotemporal causal modeling module, for generating a multimodal risk index and triggering adaptive environmental control instructions; Data storage and interaction module, used to store multi-source heterogeneous data and model parameters, and provide a cross-platform data interface; Wherein, the multimodal data acquisition module includes: A distributed environmental sensor array integrates temperature, humidity, light intensity, and CO2 concentration sensors, uploading data every ten minutes; A dual-band excitation multispectral imaging device is configured to simultaneously scan the plant canopy in the near-infrared and short-wave infrared bands and extract stomatal density, secondary metabolite distribution, and pathogen characteristic spectra; The microfluidic metabolome sampling unit uses surface-enhanced Raman spectroscopy (SERS) to detect the concentrations of salicylic acid and jasmonic acid in leaf exudates in real time.

2. The potato breeding monitoring and analysis system according to claim 1, wherein: The dual-band excitation multispectral imaging device further comprises: The laser-induced fluorescence excitation module uses a 405nm wavelength laser light source to stimulate leaf autofluorescence and capture the distribution characteristics of flavonoids; a spectroscopic filter wheel configured to switch between near-infrared and short-wave infrared filters in a single scan and acquire a dual-channel spectral image using an InGaAs sensor; The spectral feature fusion unit performs noise suppression and feature alignment on dual-band images based on a generative adversarial network, and outputs a latent disease probability map.

3. The potato breeding monitoring and analysis system according to claim 1, wherein: The spatiotemporal causal modeling module includes: The bidirectional causal temporal convolutional network (BiC-TCN) uses a dilated convolution kernel to extract the long-term dependencies between environmental parameter time series and gene expression data, and calculates the impact weights of environmental events on the expression pathways of heat shock protein HSP70 and pathogenesis-related protein PR1 through counterfactual intervention. The three-dimensional phenotype-genome association analysis unit reconstructs the tuber point cloud model based on structured light three-dimensional scanning, extracts curvature entropy and volume growth rate phenotypic parameters, and performs whole-genome association analysis with SNP marker data to locate QTL sites related to stress resistance.

4. The potato breeding monitoring and analysis system according to claim 3, wherein: The three-dimensional phenotype-genome association analysis unit performs: The curvature entropy of the tuber point cloud was calculated, and the surface morphological complexity was quantified using the Gaussian curvature differential algorithm. Tubers with a curvature entropy greater than 0.35 were selected as candidate samples for stress resistance. Dynamic growth rate modeling, fitting the logistic growth curve based on time series point cloud data, extracting the maximum growth rate Vmax and inflection point date; The stress resistance QTL mapping engine input the curvature entropy and Vmax phenotypic parameters into a mixed linear model, combined with SolCAP chip SNP data, to identify the StCIPK23 gene on chromosome 5 and the StWRKY45 gene on chromosome 8 as key resistance loci.

5. The potato breeding monitoring and analysis system according to claim 1, wherein: The degradation warning and control module includes: MRI computing engine, configured to fuse latent disease probability, environmental stress index and genetic vulnerability score, and dynamically generate risk levels through fuzzy cognitive maps; The adaptive control strategy generator, based on the deep deterministic policy gradient algorithm, optimizes the control parameters of greenhouse equipment according to real-time MRI values ​​and dynamically adjusts the lighting duration and temperature control threshold during the tuber swelling period.

6. The potato breeding monitoring and analysis system according to claim 5, characterized in that: The MRI calculation engine performs: Environmental pressure index calculation, mapping the contribution of environmental events output by BiC-TCN into a 0-1 standardized score; Gene vulnerability score, based on the genotype of QTL loci mapped by 3D-PGAS, calculates the variety-specific disease resistance attenuation coefficient; Risk level decision: when the probability of latent disease is >0.6, the environmental stress index is >0.5, and the genetic vulnerability score is >0.7, a high-risk warning is triggered and targeted regulation is initiated.

7. The potato breeding monitoring and analysis system according to claim 1, wherein: The microfluidic metabolomics sampling unit comprises: The in situ exudate capture chip integrates a nanoporous membrane and a microfluidic structure to collect leaf exudate at a flow rate of 5 μL / min; The SERS enhancement substrate uses a gold-titanium dioxide core-shell nanostructure to improve the detection sensitivity of the characteristic peaks of salicylic acid and jasmonic acid; The metabolism-spectral spatiotemporal alignment module matches the SERS detection results with the multispectral imaging data according to the acquisition timestamp and spatial coordinates to construct a metabolite-spectral response association database.

8. A monitoring and analysis method for potato breeding, characterized in that: The following steps are involved: S1. Synchronously acquire environmental parameters, multispectral images, and metabolite concentration data through a multimodal data acquisition module; S2. Use the spatiotemporal causal modeling module to construct the BiC-TCN network and 3D-PGAS model to analyze the dynamic association between environment, gene and phenotype; S3. Generate an MRI risk index based on the degradation warning control module and trigger an adaptive environment control instruction; S4. The model prediction accuracy was verified through double-blind closed-loop validation and gene editing reverse experiments.

9. The monitoring and analysis method for potato breeding according to claim 8, characterized in that: The step S2 further comprises: S21. A bidirectional causal temporal convolutional network (BiC-TCN) was used to extract the temporal causal relationship between environmental parameters and gene expression, and to calculate the probability of upregulation of HSP70 gene expression by continuous high temperature. S22. Extracting tuber curvature entropy and volume growth rate based on structured light 3D scanning data, and mapping stress resistance QTLs through genome-wide association analysis. S23. Through counterfactual intervention simulation, quantify the contribution of abnormal environmental events to seed potato degradation and generate an attribution analysis report.

10. The monitoring and analysis method for potato breeding according to claim 8, characterized in that: The step S4 comprises: S41. We simulated eight temperature-humidity combinations in an artificial climate chamber and used a Bayesian network to infer the pathways by which environmental parameters influence gene expression, verifying that the causal model error was <8%. S42. Use CRISPR-Cas9 to knock out the StWRKY45 gene and compare the phenotypic parameters of the knockout strain with those of the wild-type strain to verify the biological significance of the 3D-PGAS model. S43. Visualize MRI decision-making evidence through layer-wise relevance propagation to ensure that warning results conform to agronomic empirical rules.

Images