Peanut plant growth state intelligent monitoring method and system, and electronic equipment

Abstract

The invention relates to the technical field of peanut plant growth state intelligent monitoring, in particular to a peanut plant growth state intelligent monitoring method and system and electronic equipment. The method comprises the following steps: respectively acquiring visible light and near-infrared images, laser radar point cloud data, near-infrared spectra and microscopic images of peanut plants by using a multispectral camera, a laser radar, a near-infrared camera and a microscopic imager carried by a field robot; carrying out de-atomization and color correction on the image by adopting a self-adaptive illumination compensation algorithm; a laser radar point cloud is fused to construct a plant three-dimensional point cloud model; segmenting an overlapped leaf area based on a graph neural network, and extracting morphological characteristics of a single plant; the nitrogen content and the moisture content of the leaves are analyzed through near infrared spectroscopy, and early scabs are recognized in combination with microscopic imaging; and inputting the morphological characteristics and the physiological parameters into a growth state evaluation model, and outputting growth indexes and disease and pest early warning. According to the invention, automatic and refined monitoring and growth state comprehensive evaluation of peanut plant morphology and physiological parameters are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for peanut plant growth status, specifically to an intelligent monitoring method and system for peanut plant growth status, and electronic equipment. Background Technology

[0002] As an important oilseed and cash crop, accurate monitoring of peanut growth is crucial for guiding field management and improving yield and quality. Traditional monitoring methods mainly rely on manual observation and experience-based judgment, which have drawbacks such as low efficiency, strong subjectivity, and insufficient coverage, making it difficult to meet the needs of modern, large-scale peanut cultivation.

[0003] In recent years, with the development of agricultural information technology, some crop monitoring methods based on machine vision and sensors have been gradually applied. For example, visible light cameras are used to acquire crop phenotypic information, or spectral equipment is used to detect leaf nutritional status. However, in the specific application scenario of peanut plants, many technical bottlenecks still exist. First, peanut plants have dense and often overlapping leaves, making it difficult for ordinary image segmentation methods to accurately distinguish individual leaves and extract reliable morphological features (such as leaf area and plant height), resulting in low individual plant recognition rates. Second, the field environment has complex lighting changes (such as strong light, cloudy days, and fog), making ordinary imaging algorithms susceptible to interference, affecting image quality and the authenticity of color reproduction, thereby reducing the accuracy of subsequent feature extraction. Furthermore, the identification of plant physiological states (especially early pests and diseases) often relies on a single method (such as relying solely on spectral data or ordinary images), lacking effective fusion of multimodal data, making it difficult to achieve early and accurate disease warnings. In addition, existing technologies mostly focus on static parameter monitoring, lacking the ability to perform time-series modeling and comprehensive evaluation of dynamic changes in plant growth. Peanut varieties exhibit significant differences in growth characteristics and morphology, but existing systems generally lack adaptive calibration mechanisms for variety characteristics, limiting the universality and accuracy of monitoring models. Therefore, there is an urgent need to develop an intelligent monitoring method that integrates multi-source information, is resistant to environmental interference, can accurately identify individual plants, and achieve a comprehensive assessment of growth status.

[0004] Therefore, existing technologies still need further development. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide an intelligent monitoring method, system, and electronic device for monitoring the growth status of peanut plants, so as to solve the problems existing in the prior art.

[0006] To achieve the above-mentioned technical objectives, according to a first aspect of the present invention, the present invention provides an intelligent monitoring method for the growth status of peanut plants, comprising:

[0007] S1. Visible light and near-infrared images of peanut plants are acquired using a multispectral camera mounted on a field robot; the images are then dehazed and color corrected using an adaptive illumination compensation algorithm.

[0008] S2. Acquire lidar point cloud data of peanut plants using lidar on a field robot, and construct a three-dimensional point cloud model of the plants by fusing the lidar point cloud data.

[0009] S3. Based on graph neural network segmentation of overlapping leaf regions, extract morphological features of individual plants;

[0010] S4. Near-infrared spectra and microscopic images of peanut plants are obtained by using a near-infrared camera and microscopic imager mounted on a field robot. The nitrogen and water content of leaves are analyzed by near-infrared spectroscopy, and early lesions are identified by combining microscopic imaging.

[0011] S5. Input morphological characteristics and physiological parameters into the growth status assessment model, and output growth index and pest and disease warning.

[0012] Specifically, the adaptive illumination compensation algorithm includes:

[0013] The brightness component is decomposed using the HSV color space, and guided filtering is used to eliminate glare.

[0014] Transmittance optimization based on atmospheric scattering model restores true-color images in cloudy scenes.

[0015] Specifically, the construction of a 3D point cloud model of the plant by fusing lidar point cloud data includes:

[0016] Ground point filtering and plant clustering are performed on lidar point clouds;

[0017] The pixel coordinates of the multispectral camera are registered with the spatial coordinates of the point cloud through an affine transformation.

[0018] Specifically, the graph neural network segmentation includes:

[0019] Construct a topological structure diagram of the plant, where nodes are the center points of leaves and edges represent the spatial relationships between adjacent leaves;

[0020] A graph attention mechanism is used to learn the boundary weights of the overlapping areas of the blades.

[0021] Specifically, the method of identifying early lesions using microscopic imaging includes:

[0022] Superpixel segmentation was performed on the microscopic images to extract the HSV abnormal regions of leaf lesions;

[0023] An alert is triggered when the area of ​​the lesion is greater than the first preset threshold and the shape factor is less than the second preset threshold.

[0024] Specifically, the growth state evaluation model is a two-layer LSTM network:

[0025] The first layer learns the growth pattern over time, and the second layer integrates environmental sensor data on temperature and humidity.

[0026] Specifically, the output of the dual-layer LSTM network includes three dimensions: plant height growth rate, leaf area index, and root development level.

[0027] Specifically, the edge computing module is used to perform:

[0028] Replace the convolutional layers of the YOLO-Peanut model with depthwise separable convolutions;

[0029] Deploy the quantized model weights on the FPGA chip.

[0030] Specifically, the method also includes establishing a peanut variety knowledge base:

[0031] Store the leaf morphology baseline parameters and growth curves of different varieties;

[0032] When the detection parameters deviate from the preset reference value range, variety adaptive calibration is initiated.

[0033] According to a second aspect of the present invention, an intelligent monitoring system for the growth status of peanut plants is provided, comprising:

[0034] The multi-source sensing module includes an environmental perception unit that integrates a multispectral camera, a near-infrared camera, a lidar, and a microscopic imager mounted on the field robot. It is used to acquire visible light and near-infrared images of peanut plants via the multispectral camera mounted on the field robot; to acquire lidar point cloud data of peanut plants via the lidar mounted on the field robot; and to acquire near-infrared spectra and microscopic images of peanut plants via the near-infrared camera and microscopic imager mounted on the field robot.

[0035] The image processing module is used to perform dehazing and color correction on images using an adaptive illumination compensation algorithm;

[0036] The control module is used to construct a 3D point cloud model of the plant by fusing lidar point cloud data; to segment overlapping leaf regions based on graph neural networks and extract morphological features of individual plants; to analyze leaf nitrogen and water content through near-infrared spectroscopy and identify early lesions by combining microscopic imaging; to input morphological features and physiological parameters into the growth status assessment model and output growth index and pest and disease warnings; and to update the robot's travel path in real time based on the edge computing module.

[0037] According to a third aspect of the present invention, an electronic device is provided, comprising: a memory; and a processor, wherein the memory stores computer-readable instructions, which, when executed by the processor, implement the above-described intelligent monitoring method for the growth status of peanut plants.

[0038] Beneficial effects:

[0039] The intelligent monitoring method and system for peanut plant growth status provided by this invention has the following significant advantages compared with existing technologies:

[0040] 1. Improve perception accuracy and anti-interference capability: By combining a multispectral camera with an adaptive illumination compensation algorithm (including guided filtering to remove light spots and transmittance optimization based on an atmospheric scattering model), the effects of complex illumination (strong light, cloudy days, fog) are effectively overcome, significantly improving the clarity and color reproduction accuracy of visible light and near-infrared images, providing a high-quality data foundation for subsequent feature extraction.

[0041] 2. Achieve accurate single-plant identification and morphological feature extraction: By integrating LiDAR point cloud to construct a 3D model and combining it with a leaf segmentation method based on graph neural network (utilizing plant topology and graph attention mechanism), it is possible to accurately segment overlapping leaf areas and identify single plants, effectively solving the problem of individual confusion, thereby extracting more accurate and reliable single-plant morphological features (such as plant height and crown width).

[0042] 3. Enhance the ability to diagnose physiological conditions and pests and diseases in the early stages: By combining near-infrared spectroscopy analysis (to obtain nitrogen and water content in leaves) and high-resolution microscopic imaging (to identify abnormal morphology of leaf areas through superpixel segmentation), a multimodal physiological index joint detection is formed, which significantly improves the accuracy and timeliness of identifying nutrient stress and early lesions in peanut plants (especially for small, initial symptoms), which is conducive to early warning and precise intervention.

[0043] 4. Achieve dynamic comprehensive assessment and early warning of growth status: The growth status assessment model constructed through a two-layer LSTM network learns the inherent temporal patterns of peanut growth on the one hand, and integrates real-time environmental data (temperature and humidity) on the other hand. It can dynamically output multi-dimensional growth indices (such as plant height growth rate, leaf area index, and root development level) and early warning of pest and disease risks, so as to achieve comprehensive and intelligent diagnosis of the overall health status of the plant.

[0044] 5. Improve system intelligence and universality: The adaptive calibration mechanism based on the peanut variety knowledge base automatically triggers adjustments when the detection parameters deviate significantly from the preset benchmark range for a specific variety, thereby improving the model's adaptability and accuracy in monitoring different peanut varieties and enhancing the system's robustness and promotional value.

[0045] 6. Improve operational efficiency and automation: The entire method is integrated into a field robot platform, enabling automated, all-weather, and high-coverage monitoring of peanut plant growth, significantly reducing reliance on manual labor and labor intensity, and increasing the efficiency and frequency of monitoring operations. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the intelligent monitoring method for peanut plant growth status provided in a specific embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of the system composition of the intelligent monitoring system for peanut plant growth status provided in a specific embodiment of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0049] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0050] Please see Figure 1 This invention provides an intelligent monitoring method for the growth status of peanut plants, comprising:

[0051] S1. Visible light and near-infrared images of peanut plants are acquired using a multispectral camera mounted on a field robot; the images are then dehazed and color corrected using an adaptive illumination compensation algorithm.

[0052] Specifically, the adaptive illumination compensation algorithm includes:

[0053] The brightness component is decomposed using the HSV color space, and guided filtering is used to eliminate glare.

[0054] Transmittance optimization based on atmospheric scattering model restores true-color images in cloudy scenes.

[0055] It should be noted here that before step S1, the following is included:

[0056] The first preset threshold, the second preset threshold, and the preset reference value range are preset in the control module.

[0057] It is understood that the first preset threshold, the second preset threshold, and the preset benchmark value range can be specifically set according to the actual needs of the user, as long as they are applicable to the intelligent monitoring method for peanut plant growth status proposed in this invention.

[0058] Preferably, in the preferred embodiment of the present invention, the first preset threshold is preferably 0.5%, the second preset threshold is preferably 0.3%, and the preset reference value range is preferably ±15% of the reference value. It is understood that the above parameter settings were obtained by the present invention technicians through a large number of tests, which can better realize the intelligent monitoring method for peanut plant growth status described in the present invention, and further improve the usability and reliability of the present invention.

[0059] It should be further explained that, regarding step S1, the specific design of this invention includes:

[0060] 1. Hardware configuration:

[0061] Visible light camera: Sony IMX585 sensor (resolution 4056×3040, frame rate 30fps);

[0062] Near-infrared camera: FLIRBFS-PGE-50S5C (spectral range 900-1700nm);

[0063] Acquisition height: 50±5cm from the plant canopy. This means that at a height of 50±5cm, the area covered by a single frame image is approximately 0.3m². 2 Achieving the best balance between ) and blade detail resolution (0.2mm / pixel);

[0064] Light compensation mechanism: Automatically adjusts the exposure time to the 80-120ms range (to avoid overexposure in strong light);

[0065] Furthermore, this invention designs an exposure control formula:

[0066]

[0067] in:

[0068] T e Exposure time (ms);

[0069] k: Scene reflectivity coefficient (0.18 for peanut fields);

[0070] EV target Target exposure values ​​(set to 11 / 13 / 15 for sunny / cloudy / rainy weather respectively);

[0071] S: ISO sensitivity (fixed at 100);

[0072] A: Aperture value (fixed f / 5.6).

[0073] 2. Design an HSV space decomposition scheme:

[0074] ① Convert RGB image to HSV space:

[0075] IV = 0.299R + 0.587G + 0.114B (CIE 1931 standard)

[0076] Among them, R, G, and B are the luminance components.

[0077] ②Guided filtering to remove light spots:

[0078] Mathematical derivation of guided filtering:

[0079]

[0080] Solve by minimizing the cost function:

[0081]

[0082] in:

[0083] a k and b k The linear coefficient parameter is for a local window ω centered at pixel k. k The calculated value is a. k Indicates the slope, used to adjust brightness changes, b k This indicates the intercept, used to adjust the brightness reference value;

[0084] q i : Filtered output;

[0085] p i Original brightness value;

[0086] ω k A window centered at pixel k;

[0087] ∈: Regularization coefficient (set to 0.01 to suppress noise amplification);

[0088] E(a k ,b k ): Solving for parameter a k and b k The cost function is given. The optimal coefficients are obtained by minimizing this function.

[0089] 3. Design a defogging scheme using an atmospheric scattering model:

[0090] I(x)=J(x)t(x)+A(1-t(x))

[0091] In the formula: I is the observed image, J is the actual image, A is the atmospheric light intensity, and t is the transmittance. Design transmittance optimization formula:

[0092]

[0093] Among them, β = 0.7 (the optimal value has been verified to avoid artifacts caused by excessive defogging on cloudy days);

[0094] Specifically, the preferred values ​​for the atmospheric scattering model parameters are shown in Table 1:

[0095] Table 1 Optimal parameters for the atmospheric scattering model

[0096] parameter Range of values Preferred value Selection Criteria (Shandong Experimental Field) β 0.4-0.9 0.7 Fog line curvature change rate ≤ 5% A 200-240 225 The sky area accounts for more than 30%.

[0097] S2. Acquire lidar point cloud data of peanut plants using lidar on a field robot, and construct a three-dimensional point cloud model of the plants by fusing the lidar point cloud data.

[0098] Specifically, the construction of a 3D point cloud model of the plant by fusing lidar point cloud data includes:

[0099] Ground point filtering and plant clustering are performed on lidar point clouds;

[0100] The pixel coordinates of the multispectral camera are registered with the spatial coordinates of the point cloud through an affine transformation.

[0101] It should be further explained that, regarding step S2, the specific design of this invention includes:

[0102] 1. Ground point filtering:

[0103] The RANSAC algorithm is used to fit the ground plane equation:

[0104] ax + by + cz + d = 0

[0105] in:

[0106] x, y, z are the three-dimensional coordinate components of any point in three-dimensional space;

[0107] a, b, and c together define the normal vector of the plane. The RANSAC algorithm randomly selects sample points from the point cloud and fits a planar model using these points. The core objective of the fitting process is to calculate the best normal vector (a, b, c) that best represents the plane orientation of these sample points.

[0108] d is a constant or offset. Together with the normal vector, it determines the specific location of the plane in space.

[0109] Threshold setting: Points with a distance >10mm from the plane are considered plant points;

[0110] Derivation of the ground filtering equation:

[0111]

[0112] in:

[0113] (x p ,y p ,z p ): Point cloud coordinates.

[0114] (1) Planarity coefficient solution: RANSAC was used for 10,000 iterations, with an interior point threshold d < 10 mm; (2) Validation: Test data for different types of fields are shown in Table 2.

[0115] Table 2 Ground Filtering Accuracy Verification

[0116] Land surface type leveling dry land ridge cropping field Muddy after rain average error 2.3 5.1 7.8 maximum deviation 8.1 13.4 25.3

[0117] 2. Coordinate transformation formula extension:

[0118] ① Design the affine transformation matrix:

[0119]

[0120] in:

[0121] [u,v]: Image pixel coordinates;

[0122] [X,Y,Z]: Spatial point cloud coordinates;

[0123] r ij Rotation matrix elements (calibration accuracy ±0.1°);

[0124] s x ,s y Pixel size ratio (CMOS specification 5.5μm);

[0125] t x ,t y Translation component (8cm baseline between lidar and camera).

[0126] ② Iterative optimization of s using the ICP algorithm x r ij Parameters (error convergence threshold 0.5 pixels). S3. Segment overlapping leaf regions based on graph neural network and extract morphological features of individual plants.

[0127] Specifically, the graph neural network segmentation includes:

[0128] Construct a topological structure diagram of the plant, where nodes are the center points of leaves and edges represent the spatial relationships between adjacent leaves;

[0129] A graph attention mechanism is used to learn the boundary weights of the overlapping areas of the blades.

[0130] It should be further explained that, regarding step S4, the specific design of this invention includes:

[0131] 1. Topology graph construction:

[0132] Node characteristics:

[0133] Spatial coordinates: (x, y, z) → relative position percentage (x w / w,y h / h,z d / d);

[0134] Definition of normal vector:

[0135]

[0136] Where Φ is the point cloud surface function;

[0137] Edge connection conditions: the center distance between the two blades is <25mm and the angle between their normal vectors is <35° (the optimal value is based on the creeping characteristics of peanuts).

[0138] 2. Graph attention mechanism:

[0139]

[0140] In the formula:

[0141] α ij The attention weight between nodes i and j is used to quantify the correlation between the features of the two nodes and to score the importance of the overlapping area of ​​the leaves.

[0142] h i h j : This is the node feature vector, used to describe the attributes of a single leaf. The peanut leaf feature vector is used to represent: center point coordinates (x, y, z), normal vector, area, and curvature.

[0143] W is the weight matrix with a dimension of 128×128, used to study the morphology and spatial relationship of the leaf.

[0144] 'a' is a trainable parameter vector used to calculate feature similarity weights. The vector has a dimension of 256 and determines which features are more important for overlap detection.

[0145] || indicates vector concatenation, used to merge the features of two nodes. After concatenation, [Wh i ||Wh k The dimension of ] is 256:

[0146] a T [Wh i ||Whj The symbol ] represents the inner product operation, used to calculate the feature association strength between two nodes. The higher the output value, the greater the possibility of leaf overlap.

[0147] LeakyReLU is an activation function used to address the problem of lost negative features. Specifically:

[0148]

[0149] exp(·) is an exponential operation used to highlight important connections and convert similarity scores into positive real numbers;

[0150] N i Let be the set of nodes, and let be the set of neighbors of node i. Leaves that satisfy the connection conditions are: center distance < 25mm and the angle between normal vectors < 35°.

[0151] ∑ k : is the normalized denominator, and is the sum of node similarities, ensuring

[0152] Boundary weight learning: When the edge weight of the overlapping region is >0.8, it is determined to be a cross leaf. The decision rule is based on the following table:

[0153] Table 3. Basis for Decision-Making Rules

[0154]

[0155] The calculation process of the above scheme is illustrated below with a specific example:

[0156] 1. Node feature vectors:

[0157] h i = [0.12, 0.33, 0.05, 0.92, -0.21, 0.31, 15.7, 0.08] (center coordinates (0.12, 0.33, 0.05), normal vector (0.92, -0.21, 0.31), area 15.7 square centimeters, curvature 0.08)

[0158] 2. Calculation steps:

[0159] ① Feature projection: Wh i ∈R 128 (Extracting spatial and morphological features);

[0160] ② Vector concatenation: [Wh i ||Wh j (Merge 256-dimensional features of two blades);

[0161] ③Similarity rating: a T [·] = 2.17 (positive correlation);

[0162] ④ Activate output: LeakyReLU(2.17) = 2.17;

[0163] ⑤ Weight calculation: α is obtained after softmax normalization. ij =0.86;

[0164] ⑥ Decision output: 0.86>0.8, indicating that the intersecting blades need to be separated.

[0165] Understandably, this formula adaptively captures the complex spatial relationships of peanut plants through learnable weights (W and a):

[0166] 1. Topology modeling:

[0167] The node connection conditions (center distance < 25 mm and included angle < 35°) are precisely matched to the creeping growth characteristics of peanuts;

[0168] 2. Feature Enhancement: Normal Vector Feature Calculation Effectively quantify blade tilt angle;

[0169] 3. Dynamic learning:

[0170] During training, the loss function L = 0.7L. edge +0.3L node Strengthen boundary weight learning;

[0171] in:

[0172] 1)L edge The boundary segmentation error term refers to the boundary recognition loss in the overlapping area of ​​leaves in the peanut scenario. The loss function is preferably binary cross-entropy (BCE) or DiceLoss.

[0173] 2)L node The node prediction error term mainly consists of regression loss on attributes such as the blade center point and normal vector. The preferred loss function is mean squared error (MSE) or cosine similarity loss.

[0174] 3) A weighting ratio of 0.7 / 0.3: the optimal ratio verified through field trials.

[0175] Prioritize boundary loss (70%): because overlapping leaf segmentation is the biggest challenge in the monitoring system;

[0176] Node loss is minor (30%): the basic features are relatively easy to learn.

[0177] It should be further noted that the hyperparameter optimization of the graph attention network (GAT) in the above scheme is shown in Table 4:

[0178] Table 4 shows the hyperparameter optimization of the Attention Network (GAT).

[0179] parameter Range of values Preferred value Test basis (accuracy%) Number of attention heads 4-16 8 89.2 vs 92.3 Hidden layer dimensions 64-256 128 85.7 vs 92.1 Learning rate 0.0001-0.01 0.001 loss curve smoothness Edge weight threshold 0.6-0.9 0.8 Maximize F1 score

[0180] It should be further explained that, in the above scheme, regarding the loss function L:

[0181] L = 0.7·BCE(w pred ,w gt )+0.3·||θ pred -θ gt ||2

[0182] In the formula:

[0183] w: edge weight (0-1);

[0184] θ: Angle between normal vectors (in radians);

[0185] Specifically, the functions of each component in the above formula are shown in Table 5:

[0186] Table 5 shows the role of each component in the loss function.

[0187]

[0188]

[0189] S4. Near-infrared spectra and microscopic images of peanut plants are obtained by using a near-infrared camera and microscopic imager mounted on a field robot. The nitrogen and water content of the leaves are analyzed by near-infrared spectroscopy, and early lesions are identified by combining microscopic imaging.

[0190] Specifically, the method of identifying early lesions using microscopic imaging includes:

[0191] Superpixel segmentation was performed on the microscopic images to extract the HSV abnormal regions of leaf lesions;

[0192] An alert is triggered when the area of ​​the lesion is greater than the first preset threshold and the shape factor is less than the second preset threshold.

[0193] It should be further explained that, regarding step S4, the specific design of this invention includes:

[0194] 1. Near-infrared nitrogen content model

[0195]

[0196] in:

[0197] R 890 The reflectance is at a wavelength of 890 nm, which is the chlorophyll absorption valley.

[0198] R 680 The reflectance is at a wavelength of 680 nm, which is the chlorophyll absorption peak.

[0199] The coefficients were calibrated by measuring 300 samples using the Kjeldahl nitrogen determination method.

[0200] 2. Early lesion identification:

[0201] Superpixel segmentation: The SLIC algorithm generates 300 regions (optimal balance between detail and computational cost).

[0202] Criteria for diagnosing lesions:

[0203]

[0204] Specifically, the method for calculating the shape factor is as follows:

[0205]

[0206] Table 6 shows the meaning of each component in the above formula and its application in peanut plant monitoring.

[0207] Table 6. Meaning of each component of the shape factor formula and its application in peanut plant monitoring.

[0208]

[0209] A shape threshold of 0.3 can distinguish between round insect holes and irregular lesions.

[0210] Understandably, the biological basis for the critical value of 0.3 is: the roundness of insect holes is 0.6±0.1, and the lesion is 0.2±0.1 (derived from the analysis of 2,350 samples).

[0211] S5. Input morphological characteristics and physiological parameters into the growth status assessment model, and output growth index and pest and disease warning.

[0212] Specifically, the growth state evaluation model is a two-layer LSTM network:

[0213] The first layer learns the growth pattern over time, and the second layer integrates environmental sensor data on temperature and humidity.

[0214] The output includes three dimensions: plant height growth rate, leaf area index, and root development level.

[0215] It should be further explained that, regarding step S5, the specific design of this invention includes:

[0216] 1. LSTM network structure:

[0217] Input layer: temporal morphological parameters (plant height, leaf area, etc.) + environmental parameters (temperature and humidity);

[0218] First layer LSTM: 128 neurons (capturing weekly growth trends);

[0219] The second layer of LSTM: 64 neurons (integrating daily environmental fluctuations);

[0220] 2. Output layer design:

[0221] Root development level:

[0222]

[0223] Where ΔH is the plant height and Δt is the time interval.

[0224] The threshold is based on the peanut growth stenography standard NY / T1354-2018.

[0225] Specifically, this invention also relates to data augmentation methods during model training:

[0226] Time warp: Randomly scaled timing (amplitude ±20%);

[0227] Noise injection: Gaussian noise σ=0.05·max(X).

[0228] Specifically, this invention also relates to an early stopping method during model training:

[0229]

[0230] in, Let be the validation set loss value at the current time (step t); This represents the validation set loss value 10 steps ago (step t-10).

[0231] Understandably, the actual execution logic of the formula is to check the improvement every 10 epochs of training (to save computational resources);

[0232] For example:

[0233] Historical losses

[0234] Current losses

[0235] relative change

[0236] Since Δ = -0.001 is not greater than -0.001, continue training;

[0237] If Δ≥-0.0009 in the next round, stop immediately.

[0238] Specifically, the method also includes establishing a peanut variety knowledge base:

[0239] Store the leaf morphology baseline parameters and growth curves of different varieties;

[0240] When the detection parameters deviate from the preset reference value range, variety adaptive calibration is initiated.

[0241] It should be further explained that the solution designed in this invention for establishing a peanut variety knowledge base includes:

[0242] 1. Design the variety benchmark parameter system as shown in Table 7.

[0243] Table 7. Variety Benchmark Parameter System

[0244]

[0245] Note: The above design covers 12 main cultivated varieties, with ≥200 plants for each variety (joint field measurement by Shandong Academy of Agricultural Sciences).

[0246] 2. Design an adaptive calibration trigger mechanism:

[0247] ① Design a deviation metric algorithm:

[0248]

[0249] in:

[0250] A d The detected leaf area value;

[0251] A b This serves as the baseline value for the leaf area of ​​the knowledge base.

[0252] σ A The standard deviation of the benchmark value;

[0253] G d (t) represents the detected growth index value;

[0254] G b (t) represents the predicted value of the growth curve.

[0255] Triggering condition: Detection of δ > Δ for 5 consecutive frames th (Δ th =Preset threshold) Dynamic threshold settings are shown in Table 8:

[0256] Table 8 Dynamic Threshold Setting

[0257] Parameter type Static threshold Dynamic correction factor Leaf morphology 15％ Reduce to 12% when the temperature is above 35℃. growth rate 20％ Reduce the moisture level to 15% when soil moisture is below 60%.

[0258] 3. Design of morphological parameter calibration scheme:

[0259] ① Adjustment of Hough circle detection radius:

[0260]

[0261] in:

[0262] k: Scaling factor (empirical value 0.15);

[0263] A d The detected leaf area value;

[0264] A b This serves as the baseline value for the leaf area of ​​the knowledge base.

[0265] R b The original detection radius;

[0266] R new The adjusted detection radius;

[0267] Function: To expand the detection radius when the leaf area is large.

[0268] ②Design superpixel cluster number correction:

[0269]

[0270] in:

[0271] R b The original detection radius;

[0272] R new The adjusted detection radius;

[0273] N seg This is the corrected number of superpixel clusters;

[0274] The baseline value is 300; reduce the number of segments when the leaf area is too large.

[0275] ③ Growth model refitting:

[0276] Design online optimization parameters for growth potential:

[0277]

[0278] in:

[0279] Table 9 shows the meaning of each component in the online optimization formula for growth potential parameters and examples of field applications:

[0280] Table 9. Meaning of each component in the online optimization formula for growth potential parameters and examples of field applications.

[0281]

[0282]

[0283] Constraints:

[0284] K∈[0.8K b 1.2K b ](Kb (The original baseline value);

[0285] The root mean square error (RMSE) is less than 1.5 cm.

[0286] It should be further explained that, regarding step S6, the specific design of this invention includes:

[0287] Please see Figure 2 The present invention provides another embodiment, which provides an intelligent monitoring system for the growth status of peanut plants. The intelligent monitoring system for the growth status of peanut plants includes:

[0288] The multi-source sensing module 100 includes an environmental perception unit that integrates a multispectral camera, a near-infrared camera, a lidar, and a microscopic imager mounted on the field robot. It is used to acquire visible light and near-infrared images of peanut plants via the multispectral camera mounted on the field robot; to acquire lidar point cloud data of peanut plants via the lidar mounted on the field robot; and to acquire near-infrared spectra and microscopic images of peanut plants via the near-infrared camera and microscopic imager mounted on the field robot.

[0289] Image processing module 200 is used to perform dehazing and color correction on images using an adaptive illumination compensation algorithm;

[0290] The control module 300 is used to construct a three-dimensional point cloud model of the plant by fusing lidar point cloud data; to segment overlapping leaf regions based on graph neural networks and extract morphological features of individual plants; to analyze leaf nitrogen and water content through near-infrared spectroscopy and identify early lesions by combining microscopic imaging; to input morphological features and physiological parameters into the growth status assessment model and output growth index and pest and disease warnings; and to update the robot's travel path in real time based on the edge computing module.

[0291] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising:

[0292] The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the intelligent monitoring method for peanut plant growth status. The computer device can be broadly categorized 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 may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0293] This 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 methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such 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, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0294] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media 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 drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0295] The 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, provided that such combination does not contain contradictions.

[0296] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent monitoring of peanut plant growth status, characterized in that, The method includes: S1. Visible light and near-infrared images of peanut plants are acquired using a multispectral camera mounted on a field robot; the images are then dehazed and color corrected using an adaptive illumination compensation algorithm. S2. Acquire lidar point cloud data of peanut plants using lidar on a field robot, and construct a three-dimensional point cloud model of the plants by fusing the lidar point cloud data. S3. Based on graph neural network segmentation of overlapping leaf regions, extract morphological features of individual plants; S4. Near-infrared spectra and microscopic images of peanut plants are obtained by using a near-infrared camera and microscopic imager mounted on a field robot. The nitrogen and water content of leaves are analyzed by near-infrared spectroscopy, and early lesions are identified by combining microscopic imaging. S5. Input morphological characteristics and physiological parameters into the growth status assessment model, and output growth index and pest and disease warning.

2. The intelligent monitoring method for peanut plant growth status according to claim 1, characterized in that, The adaptive illumination compensation algorithm specifically includes: The brightness component is decomposed using the HSV color space, and guided filtering is used to eliminate glare. Transmittance optimization based on atmospheric scattering model restores true-color images in cloudy scenes.

3. The intelligent monitoring method for peanut plant growth status according to claim 2, characterized in that, The method of constructing a three-dimensional point cloud model of the plant by fusing lidar point cloud data includes: Ground point filtering and plant clustering are performed on lidar point clouds; The pixel coordinates of the multispectral camera are registered with the spatial coordinates of the point cloud through an affine transformation.

4. The intelligent monitoring method for peanut plant growth status according to claim 1, characterized in that, The graph neural network segmentation includes: Construct a topological structure diagram of the plant, where nodes are the center points of leaves and edges represent the spatial relationships between adjacent leaves; A graph attention mechanism is used to learn the boundary weights of the overlapping areas of the blades.

5. The intelligent monitoring method for peanut plant growth status according to claim 1, characterized in that, The method of combining microscopic imaging to identify early lesions includes: Superpixel segmentation was performed on the microscopic images to extract the HSV abnormal regions of leaf lesions; An alert is triggered when the area of ​​the lesion is greater than the first preset threshold and the shape factor is less than the second preset threshold.

6. The intelligent monitoring method for peanut plant growth status according to claim 1, characterized in that, The growth state evaluation model is a two-layer LSTM network: The first layer learns the growth pattern over time, and the second layer integrates environmental sensor data on temperature and humidity.

7. The intelligent monitoring method for peanut plant growth status according to claim 6, characterized in that, The output of the growth status assessment model includes three dimensions: plant height growth rate, leaf area index, and root development level.

8. The intelligent monitoring method for peanut plant growth status according to any one of claims 1-7, characterized in that, The method also includes establishing a peanut variety knowledge base: Store the leaf morphology baseline parameters and growth curves of different varieties; When the detection parameters deviate from the preset reference value range, variety adaptive calibration is initiated.

9. An intelligent monitoring system for the growth status of peanut plants, characterized in that, include: The multi-source sensing module includes an environmental perception unit that integrates a multispectral camera, a near-infrared camera, a lidar, and a microscopic imager mounted on the field robot. It is used to acquire visible light and near-infrared images of peanut plants via the multispectral camera mounted on the field robot; to acquire lidar point cloud data of peanut plants via the lidar mounted on the field robot; and to acquire near-infrared spectra and microscopic images of peanut plants via the near-infrared camera and microscopic imager mounted on the field robot. The image processing module is used to perform dehazing and color correction on images using an adaptive illumination compensation algorithm; The control module is used to construct a three-dimensional point cloud model of the plant by fusing lidar point cloud data; to segment overlapping leaf regions based on graph neural networks and extract morphological features of individual plants; and to analyze leaf nitrogen and water content through near-infrared spectroscopy and identify early lesions by combining microscopic imaging. It is used to input morphological characteristics and physiological parameters into the growth status assessment model, and output growth index and pest and disease warning.

10. An electronic device, characterized in that, include: Memory; The processor, wherein the memory stores computer-readable instructions, which, when executed by the processor, implement the intelligent monitoring method for peanut plant growth status according to any one of claims 1 to 8.

Images