Peanut phenotype intelligent analysis method and system based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of peanut phenotype intelligent analysis scheme design based on an unmanned aerial vehicle, in particular to a peanut phenotype intelligent analysis method and system based on the unmanned aerial vehicle. The method comprises the following steps: acquiring a multi-modal image through a visible light, multispectral and thermal infrared sensor carried by an unmanned aerial vehicle, separating a plant area through illumination normalization and soil mask treatment, aligning data in different growth stages by using a time sequence registration algorithm, and extracting morphological, spectral and thermal characteristics to construct a fusion matrix; inputting the matrix into a dynamic growth model fusing LSTM and a crop physiological model, and outputting the number of pods, biomass and disease resistance parameters; and generating fertilization and irrigation decision suggestions in combination with a preset threshold. The system comprises a multi-modal data acquisition module, an image processing module, a feature fusion module, a model calculation module and a decision generation module. According to the method, multi-source data collaborative analysis, time sequence growth precise modeling and field management dynamic optimization are realized, and the peanut phenotype analysis efficiency and decision scientificity are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of designing a peanut phenotypic intelligent analysis scheme based on drones, and in particular to a peanut phenotypic intelligent analysis method and system based on drones. Background Art

[0002] Traditional peanut phenotyping relies primarily on manual field observations and laboratory measurements, which are subject to inefficiency, subjectivity, and destructive sampling. While remote sensing technology has been gradually applied to crop monitoring in recent years, existing methods are mostly based on single sensors (such as visible light or spectral imaging), making it difficult to fully capture the multidimensional phenotypic characteristics of peanut plants, including morphology, physiology, and stress responses. For example, visible light images can capture morphological parameters such as plant height and canopy structure, but cannot quantify chlorophyll content or transpiration status. While thermal infrared data can reflect canopy temperature anomalies, it lacks the ability to model dynamic correlations with growth stages. Furthermore, existing image processing algorithms face challenges in complex field scenarios, such as illumination interference, soil background noise, and plant occlusion, resulting in insufficient feature extraction accuracy. At the data analysis level, most studies use static models or single machine learning methods, failing to effectively integrate time-series growth data with crop physiology, limiting the accuracy of phenotypic parameters (such as pod number and disease resistance). Some attempts have been made to incorporate drone platforms, but the collaborative analysis capabilities of multimodal data are insufficient, and the systems lack real-time performance, making them unable to provide immediate decision support for field management. Therefore, there is an urgent need for an intelligent analysis method that integrates multi-source sensing, adaptive data processing and dynamic growth modeling to solve the problems of fragmented phenotypic analysis and delayed decision-making in existing technologies, and to promote the development of precision peanut breeding and smart agriculture.

[0003] Therefore, the existing technology needs to be further developed. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above technical deficiencies and provide a peanut phenotypic intelligent analysis method and system based on drones to solve the problems existing in the existing technology.

[0005] To achieve the above technical objectives, according to a first aspect of the present invention, a method for intelligent peanut phenotyping based on drones is provided, comprising:

[0006] S1. Collect visible light, multispectral, and thermal infrared images of peanut plants using a multimodal sensor mounted on a drone.

[0007] S2. Performing illumination normalization on the collected images and separating the plants from the background based on a soil masking algorithm;

[0008] S3. Align plant images at different growth stages using a time series registration algorithm.

[0009] S4. Extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix;

[0010] S5. Input the fusion matrix into the pre-trained dynamic growth model to output pod number, biomass, and disease resistance phenotypic parameters;

[0011] S6. Generate peanut growth status assessment reports and field management decision recommendations based on phenotypic parameters.

[0012] Specifically, the multimodal sensor includes:

[0013] Visible light cameras, multispectral sensors, and thermal infrared sensors.

[0014] Specifically, the illumination normalization process adopts a radiation correction method based on a reference white plate, and eliminates illumination differences by linearly mapping the reflectivity of the calibration plate and the grayscale value of the image.

[0015] Specifically, the separation of plants from background uses an improved U-Net network, and the training data includes labeled images of different soil types, weed coverage, and plant density.

[0016] Specifically, the time series registration algorithm extracts key points of the main stem of the plant and combines SIFT feature matching with affine transformation to perform cross-stage image alignment.

[0017] Specifically, the dynamic growth model is a hybrid architecture that integrates an LSTM neural network and a crop physiological model, and its input includes the current feature matrix and environmental parameters of historical growth stages.

[0018] Specifically, the disease resistance phenotypic parameter is determined by combining the abnormal canopy temperature area in the thermal infrared image with the chlorophyll content change in the multispectral image.

[0019] Specifically, the dynamic growth model is trained using a self-supervised learning framework, using unlabeled images to generate pseudo labels through comparative learning, and fine-tuning the model in combination with expert labeled data.

[0020] Specifically, the field management decision recommendations include fertilizer application amount, irrigation time and disease prevention and control plan, and the decision rules are dynamically generated based on the difference between the phenotypic parameters and the preset threshold.

[0021] According to a second aspect of the present invention, a peanut phenotyping intelligent analysis system based on a drone is provided, comprising:

[0022] An acquisition module is used to collect visible light images, multispectral images, and thermal infrared images of peanut plants using a multimodal sensor carried by a drone;

[0023] The control module performs illumination normalization on the collected images and separates plants from the background based on a soil masking algorithm. It is used to align plant images at different growth stages using a time series registration algorithm. It is used to extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix. It is used to input the fusion matrix into a pre-trained dynamic growth model and output phenotypic parameters such as pod number, biomass, and disease resistance. It is used to generate a peanut growth status assessment report and field management decision recommendations based on the phenotypic parameters.

[0024] Beneficial effects:

[0025] 1. Collaborative perception and efficient analysis of multimodal data: Through the coordinated collection of visible light, multispectral, and thermal infrared sensors, plant morphology, spectral reflectance, and canopy temperature information are simultaneously acquired, overcoming the limitations of a single data source. This provides multi-dimensional input for phenotypic parameter inversion and significantly improves the detection accuracy of complex traits such as disease resistance and biomass.

[0026] 2. Robust image processing and feature fusion: Radiation correction based on a reference whiteboard and an improved U-Net segmentation algorithm effectively eliminates illumination fluctuations and soil background interference, enabling accurate extraction of plant regions. The time series registration algorithm uses stem key point matching and affine transformation to address image misalignment caused by plant deformation and displacement during growth, ensuring consistency of time series data.

[0027] 3. Dynamic growth model drives precise prediction: This approach integrates LSTM time series modeling with crop physiological equations to dynamically correlate environmental parameters, historical growth status, and current multimodal characteristics. This overcomes the generalization bottleneck of traditional static models and enables continuous cross-stage prediction of parameters such as pod number and biomass, providing a reliable basis for breeding screening and yield estimation.

[0028] 4. Self-supervised learning reduces labeling dependency: A contrastive learning framework is used to generate pseudo-labels, combined with a small amount of expert annotation to fine-tune the model, significantly reducing the cost of training data labeling while improving the model's adaptability to complex field scenarios.

[0029] 5. Intelligent decision-making and resource optimization: Based on the dynamic matching of phenotypic parameters and preset thresholds, fertilization, irrigation, and disease control plans are automatically generated, avoiding the blindness of traditional experience-based decision-making, reducing water and fertilizer waste and the risk of excessive pesticide application, and promoting green and sustainable agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 1 is a flow chart of a peanut phenotypic intelligent analysis method based on drones provided in a specific embodiment of the present invention;

[0031] Figure 2 Schematic diagram of the system composition of the drone-based peanut phenotyping intelligent analysis system provided in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is clearly and completely described below in conjunction with the drawings of the present invention. Based on the embodiments in this application, other similar embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of this application. In addition, the directional words mentioned in the following embodiments, such as "up", "down", "left", "right", etc., are only reference to the directions of the drawings. Therefore, the directional words used are used to illustrate rather than limit the invention.

[0033] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0034] See also Figure 1 The present invention provides a peanut phenotypic intelligent analysis method based on drones, comprising:

[0035] S1. Use the multimodal sensor onboard a drone to collect visible light, multispectral, and thermal infrared images of peanut plants.

[0036] Specifically, the multimodal sensor includes:

[0037] Visible light cameras, multispectral sensors, and thermal infrared sensors.

[0038] It should be further explained that, regarding step S1, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0039] 1. Sensor configuration:

[0040] The drone is equipped with a visible light camera (preferably with a resolution of 20 million pixels), a multispectral sensor (preferably with wavelengths of 710-730nm for red edge and 800-900nm for near infrared), and a thermal infrared sensor (preferably with a temperature resolution of 0.05°C).

[0041] 2. Reasons for selection:

[0042] The red edge band is sensitive to chlorophyll, the near infrared can distinguish vegetation from non-vegetation, and the 0.05°C resolution can detect tiny temperature changes (the present invention preferably detects tiny temperature changes in the early stages of the disease).

[0043] Furthermore, regarding step S1, in a preferred embodiment of the present invention, the solution designed by the present invention further includes:

[0044] 1. UAV flight parameters:

[0045] Flight altitude: preferably set to 3 meters above the ground, and vertical resolution is preferably 0.5 mm / pixel.

[0046] Basis: The average height of peanut plants is 15-40cm, and a height of 3 meters can cover the details of individual plants and balance the distribution of the group.

[0047] Flight time: preferably between 10:00-11:00 am every day (sun altitude angle 45°-60°) to avoid interference from strong light reflection at noon.

[0048] Overlap rate: The heading overlap rate is preferably 80%, and the lateral overlap rate is preferably 70% to ensure seamless multispectral image stitching.

[0049] 2. Multispectral sensor band optimization:

[0050] Band selection:

[0051] Red edge 1 (710 nm): Chlorophyll absorption peak, the formula of the present invention is preferably as follows:

[0052]

[0053] Among them, REIP is the red edge inflection point position, which is used to monitor blade aging, 700 is the reference wavelength (the starting reference value for red edge inflection point calculation), 40 is the empirical scaling factor, and R 670 is the vegetation reflectance at a wavelength of 670nm, R 700 is the vegetation reflectance at a wavelength of 700nm, R 780 is the vegetation reflectance at a wavelength of 780nm.

[0054] Near infrared (850nm): Calculate NDVI using the following formula:

[0055]

[0056] The threshold is preferably set at 20.6 for healthy plants.

[0057] Among them, NIR is the reflectance of the near-infrared band, Red is the reflectance of the infrared band, and NDVI is the calculated value of the normalized vegetation index, which is used to compress the reflectance difference into the range of [-1, 1] and provide a standardized and comparable vegetation status assessment indicator.

[0058] Spectral resolution: Red edge band bandwidth ≤ 5nm, ensuring chlorophyll content inversion error < 5%.

[0059] 3. Thermal infrared calibration:

[0060] Blackbody radiation source: The drone is equipped with a miniature blackbody (temperature range 20-50°C, accuracy ±0.1°C) to collect the ambient radiation baseline value every 5 minutes.

[0061] Temperature inversion formula:

[0062]

[0063] Where c1 = 1.19104 × 10 -16 W\cdotpm 2 / sr,

[0064] c²=1.43877×10 -2 m\cdotpK, L is the radiation intensity received by the sensor.

[0065] S2. Perform illumination normalization on the collected images and separate the plants from the background based on the soil masking algorithm.

[0066] Specifically, the illumination normalization process adopts a radiation correction method based on a reference white plate, and eliminates illumination differences by linearly mapping the reflectivity of the calibration plate and the grayscale value of the image.

[0067] Specifically, the separation of plants from background uses an improved U-Net network, and the training data includes labeled images of different soil types, weed coverage, and plant density.

[0068] It should be further explained that, regarding step S2, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0069] 1. Lighting normalization algorithm:

[0070] Use a reference white plate (with known reflectivity) for radiation correction, formula:

[0071]

[0072] Among them, R calibrated is the reflectivity after calibration, I raw is the grayscale value of the original image, I dark is the dark current value I ref is the grayscale value of the reference white board, R ref is the reflectivity of the reference white board.

[0073] 2. Background separation:

[0074] An improved U-Net network (input size 512×512, output binary mask) is used, and the training data contains 4 soil types and 3 weed densities (labeled with pixel-level masks).

[0075] Furthermore, the improved U-Net network designed by the present invention includes:

[0076] 1. Encoder-Decoder Structure:

[0077] Encoder: 4-layer MobileNetV3 (width factor 0.75), output feature map size is 1 / 16 of the input.

[0078] Decoder: Bilinear upsampling + skip connection (add SE attention module), formula Where σ is the Sigmoid function, W is the adaptive weight matrix, Represents channel-by-channel multiplication, Fin is the input multi-dimensional feature map, which contains spatial and channel information.

[0079] 2. Training strategy:

[0080] The initial learning rate is 0.001, cosine annealing scheduling is used (cycle 50 rounds), and the minimum learning rate is 0.00001.

[0081] The batch size is 16, the input image is normalized to [-1, 1], and adversarial training (introducing the PatchGAN discriminator) improves the accuracy of segmentation boundaries.

[0082] 3. Loss function:

[0083] Cross entropy loss + Dice loss (weight ratio 1:1), optimizer is Adam (learning rate 0.0002, iteration 200 rounds).

[0084] S3. Use time series registration algorithm to align plant images at different growth stages.

[0085] Specifically, the time series registration algorithm extracts key points of the main stem of the plant and combines SIFT feature matching with affine transformation to perform cross-stage image alignment.

[0086] It should be further explained that, regarding step S3, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0087] 1. Key point matching:

[0088] Extract the SIFT feature points of the main stem of the plant (threshold: the number of feature points is greater than or equal to 50, and the mismatched points are eliminated by the RANSAC algorithm (the proportion of inliers is greater than or equal to 80%).

[0089] Furthermore, the present invention also designs SIFT feature point extraction optimization, and the method further includes:

[0090] The number of Gaussian pyramid layers is preferably 4, the scale factor is preferably 1.5, and the contrast threshold is preferably 0.04 (to filter out weak edge responses).

[0091] Main stem key point screening: preferably only feature points with a curvature radius greater than 10 pixels are retained (excluding leaf interference).

[0092] 2. Affine transformation matrix calculation:

[0093] formula:

[0094]

[0095] Among them, a, b, c, d are rotation and scaling parameters, t x ,t y is the translation parameter, and the optimal transformation is fitted by the least squares method.

[0096] S4. Extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix.

[0097] It should be further explained that, regarding step S4, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0098] Construct a feature matrix, including:

[0099] Morphological characteristics (plant height, canopy area), spectral characteristics (NDVI, red-edge chlorophyll index), and thermal characteristics (average canopy temperature) were concatenated into a matrix by time step with a dimension of T × 10 (T is the number of time points).

[0100] Furthermore, the method includes:

[0101] 1. Feature Standardization:

[0102] Morphological characteristics (plant height, crown width) were normalized by Z-Score, and spectral characteristics (NDVI, REIP) were normalized to [0, 1].

[0103] 2. Fusion matrix construction:

[0104] Time step T = 10 (covering the key growth period), matrix dimension

[0105] 10×10, missing values ​​are interpolated using time linear interpolation.

[0106] S5. Input the fusion matrix into the pre-trained dynamic growth model and output the pod number, biomass, and disease resistance phenotypic parameters.

[0107] Specifically, the dynamic growth model is a hybrid architecture that integrates an LSTM neural network and a crop physiological model, and its input includes the current feature matrix and environmental parameters of historical growth stages.

[0108] Specifically, the disease resistance phenotypic parameter is determined by combining the abnormal canopy temperature area in the thermal infrared image with the chlorophyll content change in the multispectral image.

[0109] Specifically, the dynamic growth model is trained using a self-supervised learning framework, using unlabeled images to generate pseudo labels through comparative learning, and fine-tuning the model in combination with expert labeled data.

[0110] It should be further explained that, regarding step S5, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0111] 1. Model Architecture:

[0112] LSTM layer (64 hidden units) + crop physiological model (Farquhar photosynthesis equation), the output layer is fully connected (3 nodes: pod number, biomass, disease resistance score).

[0113] 2. Training parameters: input historical environmental data (temperature, humidity, light), batch size 32, initial learning rate 0.001 (decayed by 10% every 50 rounds), and number of training rounds 300.

[0114] 3. Reason for selection: LSTM captures temporal dependencies, and the Farquhar equation introduces constraints on the photosynthesis mechanism, improving model generalization.

[0115] Furthermore, the method includes:

[0116] 1. LSTM layer design:

[0117] ① The hidden state is 64-dimensional, the time step expansion length is 10, and the forget gate bias is initialized to 1.0 (to alleviate gradient disappearance).

[0118] ② Input gating formula:

[0119] i t =σ(W xi x t +W hi h t-1 +b i ), output historical environmental data (temperature, humidity, PAR) and current feature matrix.

[0120] Among them, i t is the activation state of the input gate, ranging from 0 to 1, which is used to determine the proportion of the current input information to be retained (0 = completely discarded, 1 = completely retained); σ is the Sigmoid activation function, which is used to compress the linear transformation result to [0,1] to implement the gating mechanism. xi is the weight matrix input to the input gate, used to map the current input feature (preferably temperature and humidity in the present invention) to the gated space; W hi is the weight matrix from hidden state to input gate, used to map the historical hidden state (preferably the plant growth stage characteristics in the present invention) to the gated space; t The input features of the current time step include environmental parameters (temperature °C, humidity%, PAR μmol / m 2 / s) and multimodal features extracted by drones (preferably NDVI in the present invention); ht-1 b is the hidden state at the previous moment, used to encode the historical growth state of the peanut plant (preferably biomass accumulation and leaf area change in the present invention); i It is the bias term of the input gate, which is used to adjust the activation threshold to prevent invalid learning caused by too small input signals.

[0121] 2. Farquhar equation integration:

[0122] Photosynthetic rate A=min{W c ′,W j ,W s},in:

[0123] (Rubisco restrictions)

[0124] (Light energy conversion limit)

[0125] Among them, V cmax is the maximum carboxylation rate of Rubisco enzyme, which is used to reflect the Rubisco enzyme activity in peanut leaves and is dynamically predicted by LSTM (input: leaf age, nitrogen content, temperature): C i is the intercellular CO2 concentration, which is indirectly estimated by inverting stomatal conductance from UAV thermal infrared data; Γ is the CO2 concentration threshold in the absence of photorespiration, preferably 38 μmol / mol; K is the Michaelis-Menten constant of Rubisco enzyme, which is temperature-dependent and has a value of ≈404 μmol / mol for peanuts at 25°C; J is the potential electron transfer rate (related to light intensity, dynamically adjusted by LSTM output, and input is PAR (photosynthetically active radiation) and chlorophyll fluorescence data); 4.5C i +10.5Γ is the coupling term between light energy and Rubisco activity, which is used to quantify the balance between light reaction and dark reaction. The larger the denominator value, the lower the light energy conversion efficiency.

[0126] Parameter V cmax , J is dynamically adjusted by LSTM output, and the error with the measured photosynthetic rate is <15%.

[0127] It should be further explained that regarding the indirect estimation of intercellular CO2 concentration by inverting stomatal conductance through UAV thermal infrared data, the scheme designed by the present invention includes:

[0128] 1. Synchronous collection of environmental parameters:

[0129] Air temperature (Tair), relative humidity (RH), wind speed (u), and solar radiation (Rn) are obtained through a meteorological module carried by the UAV or a ground meteorological station.

[0130] The atmospheric CO2 concentration (Ca) is taken as an approximate value of 410 ppm by default, or measured in real time by a portable CO2 meter.

[0131] 2. Canopy temperature extraction and energy balance modeling:

[0132] Canopy temperature inversion: Use a thermal infrared image segmentation algorithm (preferably a vegetation mask based on NDVI in the present invention) to extract the temperature of the pure vegetation area (Tcanopy) and exclude soil background interference.

[0133] Energy balance equation (Penman-Monteith modified version):

[0134]

[0135] Explanation of symbols:

[0136] λE: Latent heat flux (W / m 2 ), which is positively correlated with the transpiration rate;

[0137] R_n: Net radiation (W / m 2 );

[0138] G: soil heat flux (can be ignored or estimated by empirical formula);

[0139] ρ_a: air density (1.2kg / m 3 );

[0140] c_p: specific heat capacity of air (1004 J / kg·K);

[0141] r_a: aerodynamic drag (s / m), calculated from wind speed and canopy roughness;

[0142] r_s: stomatal resistance (s / m), i.e. stomatal conductance g_s = 1 / r_s;

[0143] Δ: slope of saturated water vapor pressure-temperature curve (kPa / ℃);

[0144] γ: Psychrometric constant (0.066 kPa / ℃).

[0145] 3. Solve for the pore conductance (g_s):

[0146] Solve the equation by iterative method or numerical optimization (preferably Newton-Raphson method in this invention), input T_canopy and environmental parameters, and output gs (unit: mol / m 2 / s).

[0147] 4.C i calculate:

[0148] The stomatal conductance g sCombined with Fick's law:

[0149]

[0150] Substitute into the Farquhar model equation and solve Ci by numerical method (preferably bisection method in the present invention).

[0151] S6. Generate peanut growth status assessment reports and field management decision recommendations based on phenotypic parameters.

[0152] Specifically, the field management decision recommendations include fertilizer application amount, irrigation time and disease prevention and control plan, and the decision rules are dynamically generated based on the difference between the phenotypic parameters and the preset threshold.

[0153] It should be further explained that, regarding step S6, in a preferred embodiment of the present invention, the solution designed by the present invention includes:

[0154] Threshold setting: Biomass threshold: 50g / m 2 (If it is lower than this, it will trigger the recommendation of topdressing);

[0155] Canopy temperature anomaly threshold: 1.5 degrees Celsius or higher than the surrounding area (determined as disease risk).

[0156] Basis: Based on the optimal biomass range of peanut growth and early temperature response data of disease in field experiments.

[0157] It should be noted that the present invention also designs a disease resistance score calculation scheme, including:

[0158] 1. Canopy temperature anomaly detection:

[0159] Sliding window method (window size 3×3 pixels) to calculate the local temperature standard deviation Among them, T i is the temperature value of the i-th pixel in the sliding window, μ is the average temperature of all pixels in the window, N is the total number of pixels in the window, and σ is the standard deviation of the local temperature.

[0160] 2. If | T i -μ|>2σ is marked as abnormal.

[0161] Calculation of chlorophyll change rate:

[0162]

[0163] Among them, Chl current is the current chlorophyll content, Chl baseline is the baseline value of chlorophyll content in healthy plants, the baseline value is the average chlorophyll content in healthy plants (calibrated by destructive sampling), and ΔChl is the chlorophyll change rate.

[0164] Furthermore, the present invention has been verified in its implementation mode, and the verification data includes:

[0165] Segmentation accuracy:

[0166] The IoU (plant area) reaches 92.3% (test set: 500 annotated images).

[0167] Phenotype prediction error:

[0168] Pod number: RMSE = 8.2 per plant (compared with manual counting);

[0169] Biomass: R 2 =0.91 (linear regression fit);

[0170] Disease detection: F1-score = 0.87 (classification of early blight and healthy plants).

[0171] The working process of the present invention is described below by specific examples:

[0172] 1. Flight plan:

[0173] Time: Choose noon (10:00-14:00), when the stomatal opening is the largest and the environmental parameters are stable; Weather: Clear and cloudless, wind speed <3m / s, to avoid leaves getting wet after rainfall and affecting the temperature.

[0174] 2. Data processing flow:

[0175] ① Thermal infrared data calibration and canopy temperature extraction;

[0176] ② Environmental parameter interpolation (time-space matching);

[0177] ③Energy balance calculation g_s;

[0178] ④Solve the simultaneous equations Ci.

[0179] Sensitivity analysis:

[0180] The sensitivity ranking of input parameters (preferably Tcanopy, RH, and u in the present invention) to Ci is: Tcanopy>RH>u;

[0181] Optimization direction: Improve the accuracy of canopy temperature measurement and prioritize the calibration of humidity sensors.

[0182] Peanut drought stress monitoring:

[0183] When the inverted Ci drops to 200 ppm (normal value ≈ 250 ppm), combined with a drop in NDVI, the system triggers an irrigation advisory;

[0184] Fertilization decision: Increased Ci (>300ppm) and low g_s indicate that excess nitrogen has caused stomatal closure, and it is recommended to reduce nitrogen fertilizer.

[0185] It should be further explained that regarding Ci inversion and NDVI analysis, the solutions designed by the present invention include:

[0186] 1. This inversion process:

[0187] Inversion of Stomatal Conductance g from Thermal Infrared Data s (See the scheme above for methods);

[0188] Combine the Farquhar model and Fick's law to solve Ci, formula:

[0189]

[0190] Where A = min(W c ,W j ),C a =410ppm.

[0191] 2. NDVI dynamic monitoring:

[0192] The NDVI baseline of healthy peanuts is 0.65-0.75 during the flowering period and 0.55-0.65 during the pod-setting period.

[0193] Threshold setting: An NDVI drop of >5% for three consecutive days and an absolute value of <0.6 triggers an early warning.

[0194] 3. Irrigation decision logic:

[0195] Trigger conditions (satisfied simultaneously):

[0196] Ci abnormal: C i ≤200ppm (a decrease of ≥20% from the normal value);

[0197] NDVI decrease: the current NDVI decreases by ≥8% compared with the average value 7 days ago;

[0198] Soil moisture: Soil moisture at 20 cm depth <50% FC (sandy loam FC≈20% VWC).

[0199] 4. Implementation suggestions:

[0200] Irrigation amount: Supplement to 80% FC, formula: water volume (m 3 / mu) = (0.8 × FC - current humidity) × soil depth (m) × 667

[0201] Timing: Drip irrigation should be carried out early in the morning (5:00-7:00) to reduce evaporation losses.

[0202] 5. Verification indicators:

[0203] Within 3 days after irrigation, Ci recovered to ≥220 ppm and the downward trend of NDVI stabilized.

[0204] It should be further explained that, regarding the nitrogen fertilizer optimization decision based on Ci and g_s, the scheme designed by the present invention includes:

[0205] 5.1 Data Correlation Analysis:

[0206] Physiological responses to nitrogen excess:

[0207] Excess nitrogen leads to ammonium accumulation in leaves, which in turn causes cytotoxicity and subsequently stomatal closure (g s <0.15mol / m 2 / s);

[0208] Photosynthesis is blocked (A decreases), which leads to intercellular CO2 retention (C i >300ppm).

[0209] Key parameter thresholds:

[0210] Ci warning value: C for 5 consecutive days i ≥300ppm;

[0211] g_s threshold: g s <0.18mol / m 2 / s (0.2-0.4 during normal flowering period).

[0212] 5.2 Decision logic and execution:

[0213] Trigger conditions (satisfied simultaneously):

[0214] ①Ci rises: C i ≥300ppm;

[0215] ②g_s inhibition: g s <0.18mol / m 2 / s;

[0216] ③Nitrogen fertilizer history: The amount of nitrogen applied in the past 30 days is ≥8kg / mu (the recommended amount for peanuts is 6-8kg / mu).

[0217] Implementation recommendations:

[0218] Nitrogen reduction ratio: in the next growth stage (preferably the pod-setting stage in the present invention), reduce nitrogen fertilizer by 30%-50%;

[0219] Alternative solution: apply more potassium fertilizer (2 kg / mu) to alleviate ammonium toxicity, and spray 0.2% magnesium sulfate solution (to promote nitrogen metabolism). The potassium fertilizer is preferably K2O.

[0220] Verification method: within 10 days after nitrogen reduction, Ci drops to below 280ppm and g_s rises to >0.2mol / m2 / s;

[0221] Leaf ammonium content (portable meter) <1.5 mg / g.

[0222] See also Figure 2 The present invention provides another embodiment, which provides a peanut phenotypic intelligent analysis system based on a drone, and the peanut phenotypic intelligent analysis system based on a drone includes:

[0223] An acquisition module 100 is configured to collect visible light images, multispectral images, and thermal infrared images of peanut plants using a multimodal sensor carried by a drone;

[0224] Control module 200 performs illumination normalization on the captured images and separates the plants from the background based on a soil masking algorithm; is used to align plant images at different growth stages using a time series registration algorithm; is used to extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix; is used to input the fusion matrix into a pre-trained dynamic growth model and output phenotypic parameters such as pod number, biomass, and disease resistance; and is used to generate a peanut growth status assessment report and field management decision recommendations based on the phenotypic parameters.

[0225] In a preferred embodiment, the present application further provides an electronic device, comprising:

[0226] A memory; and a processor, wherein the memory stores computer-readable instructions, which, when executed by the processor, implement the drone-based intelligent peanut phenotyping analysis method. The computer device can be broadly defined as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device can include non-volatile storage media and internal memory. The non-volatile storage medium can store an operating system, computer programs, etc. in or on the non-volatile storage medium. The internal memory can provide an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect to and communicate with external devices via a network. When executed by the processor, the computer program performs the steps of the method of the present invention.

[0227] The present invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the method of an embodiment of the present invention to be performed. In one embodiment, the computer program is distributed on a plurality of computer devices or processors coupled to a network so that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, can be performed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations can be performed by one or more computer devices or processors, and one or more other method steps / operations can be performed by one or more other computer devices or processors. One or more computer devices or processors can perform a single method step / operation, or perform two or more method steps / operations.

[0228] It will be understood by those skilled in the art that the method steps of the present invention can be performed by instructing relevant hardware such as a computer device or a processor through a computer program, and the computer program can be stored in a non-transitory computer-readable storage medium, which causes the steps of the present invention to be performed when the computer program is executed. Depending on the circumstances, any reference to memory, storage, database or other media herein may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state disk, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0229] The various technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification as long as such combination does not conflict.

[0230] The specific embodiments of the present invention described above do not limit the scope of protection of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A peanut phenotypic intelligent analysis method based on drones, characterized in that: The method comprises: S1. Collect visible light, multispectral, and thermal infrared images of peanut plants using a multimodal sensor mounted on a drone. S2. Performing illumination normalization on the collected images and separating the plants from the background based on a soil masking algorithm; S3. Align plant images at different growth stages using a time series registration algorithm. S4. Extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix; S5. Input the fusion matrix into the pre-trained dynamic growth model to output pod number, biomass, and disease resistance phenotypic parameters; S6. Generate peanut growth status assessment reports and field management decision recommendations based on phenotypic parameters.

2. The peanut phenotypic intelligent analysis method based on drone according to claim 1, characterized in that: The multimodal sensor comprises: Visible light cameras, multispectral sensors, and thermal infrared sensors.

3. The peanut phenotypic intelligent analysis method based on drone according to claim 1, characterized in that: The illumination normalization process adopts a radiation correction method based on a reference white plate, and eliminates illumination differences by linearly mapping the reflectivity of the calibration plate and the grayscale value of the image.

4. The peanut phenotypic intelligent analysis method based on drone according to claim 3, characterized in that: The improved U-Net network is used to separate the plants from the background, and the training data includes labeled images of different soil types, weed coverage and plant density.

5. The peanut phenotypic intelligent analysis method based on drone according to claim 1, characterized in that: The time series registration algorithm extracts key points of the main stem of the plant and combines SIFT feature matching with affine transformation to perform cross-stage image alignment.

6. The peanut phenotypic intelligent analysis method based on drone according to claim 1, characterized in that: The dynamic growth model is a hybrid architecture that integrates an LSTM neural network and a crop physiological model, and its input includes the current feature matrix and environmental parameters of historical growth stages.

7. The peanut phenotypic intelligent analysis method based on drone according to claim 6, characterized in that: The disease resistance phenotypic parameters are determined by combining the abnormal canopy temperature area in the thermal infrared image with the chlorophyll content change in the multispectral image.

8. The peanut phenotypic intelligent analysis method based on drone according to claim 7, characterized in that: The dynamic growth model is trained using a self-supervised learning framework, using unlabeled images to generate pseudo labels through contrastive learning, and fine-tuning the model in combination with expert-labeled data.

9. The peanut phenotypic intelligent analysis method based on drone according to claim 1, characterized in that: The field management decision recommendations include fertilizer application amount, irrigation time and disease prevention and control plan, and the decision rules are dynamically generated based on the difference between phenotypic parameters and preset thresholds.

10. A peanut phenotypic intelligent analysis system based on drones, characterized in that: include: An acquisition module is used to collect visible light images, multispectral images, and thermal infrared images of peanut plants using a multimodal sensor carried by a drone; The control module performs illumination normalization on the collected images and separates plants from the background based on a soil masking algorithm. It is used to align plant images at different growth stages using a time series registration algorithm. It is used to extract the morphological, spectral, and thermal characteristics of the plants and construct a multimodal feature fusion matrix. It is used to input the fusion matrix into a pre-trained dynamic growth model and output phenotypic parameters such as pod number, biomass, and disease resistance. It is used to generate a peanut growth status assessment report and field management decision recommendations based on the phenotypic parameters.

Images