Crop disease and insect pest intelligent monitoring system based on multispectral imaging and use method thereof

Abstract

The invention relates to the technical field of agricultural information, in particular to an intelligent crop disease and pest monitoring system based on multispectral imaging and a use method thereof, and the system comprises a multispectral data acquisition and preprocessing module, a visual intelligent analysis module, an intelligent diagnosis module and a decision support module. According to the method, a high-contribution-degree spectral band is dynamically screened through a cross-band attention mechanism, and a 3D convolutional network and mutual information constrained feature decoupling reconstruction algorithm is combined, so that adaptive fusion and noise separation of multispectral features are realized to generate sparse feature representation focusing disease and pest sensitive information, and based on an Inception-v3 and LightGBM cascaded multi-model fusion algorithm, the disease and pest sensitive information is obtained. And in combination with Monte Carlo Dropout confidence evaluation, Grad-CAM interpretability analysis and TCN time convolutional network time sequence prediction, an intelligent diagnosis system with reliability quantification, decision visualization and trend pre-judgment capabilities is constructed, and accurate identification of pest and disease damage types and severity and dynamic prediction of diffusion trends in the next five days are realized.

Description

Technical Field

[0001] The present invention relates to the field of agricultural information technology, and in particular to a crop disease and insect pest intelligent monitoring system based on multispectral imaging and a method for using the same. Background Art

[0002] Crop diseases and insect pests refer to abnormal conditions in which crops suffer damage to their physiological functions, tissue structure, and external morphology during their growth and development due to invasion by harmful organisms (such as fungi, bacteria, viruses, pests, weeds, nematodes, etc.) or the influence of adverse environmental factors, thereby affecting crop yield and quality and even leading to plant death. In modern agricultural production, early and accurate monitoring of crop diseases and insect pests is the core link to ensure yield and quality. With the development of spectral detection technology, multispectral imaging has become an important means of disease and insect pest monitoring because it can capture the reflectivity differences of crops in different bands.

[0003] However, traditional multispectral pest and disease monitoring technology has exposed significant shortcomings in practical applications: first, band screening relies on manual experience and is difficult to adapt to the spectral sensitivity characteristics of different pests and diseases, resulting in low feature extraction efficiency; second, the coupling problem between pest and disease characteristics and background noise in multispectral data is prominent, and traditional algorithms cannot achieve effective separation, which seriously affects the recognition accuracy; third, the diagnostic model lacks visual decision-making basis, which makes it difficult to meet the agricultural production's demand for interpretable monitoring results; fourth, it can only identify the current state and cannot predict the development trend of diseases based on time series analysis, resulting in delayed prevention and control measures.

[0004] Based on this, the present invention provides an intelligent crop disease and insect pest monitoring system based on multispectral imaging and a method of using the same to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent crop disease and pest monitoring system based on multispectral imaging and a method of using the same. The present invention uses a cross-band attention mechanism to dynamically screen high-contribution spectral bands and combines a 3D convolutional network with a mutual information constrained feature decoupling and reconstruction algorithm to achieve adaptive fusion and noise separation of multispectral features to generate a sparse feature representation focused on sensitive information of pests and diseases. Based on a multi-model fusion algorithm of Inception-v3 and LightGBM cascade, combined with Monte Carlo Dropout confidence evaluation, Grad-CAM interpretability analysis and TCN temporal convolutional network time series prediction, an intelligent diagnostic system with reliability quantification, decision visualization and trend prediction capabilities is constructed to achieve accurate identification of pest and disease types and severity and dynamic prediction of diffusion trends in the next 5 days.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides an intelligent crop disease and insect pest monitoring system based on multispectral imaging, including a multispectral data acquisition and preprocessing module, a visual intelligent analysis module, an intelligent diagnosis module, and a decision support module, wherein:

[0008] The multispectral data acquisition and preprocessing module is used to acquire multispectral images of crops through a multispectral imaging device and preprocess the acquired original images;

[0009] The visual intelligent analysis module is used to dynamically screen band contributions through a cross-band adaptive feature fusion network, and use a feature decoupling and reconstruction algorithm to separate pest-sensitive features from background noise to generate a sparse feature representation;

[0010] The intelligent diagnosis module: Based on a multi-model fusion algorithm with confidence assessment and interpretability, it accurately identifies the type and severity of pests and diseases, and predicts the development trajectory of the disease by combining time series analysis;

[0011] The decision support module is used to combine the diagnosis results with the agronomic knowledge base to generate an executable plan including the application dosage, operation time, and recommended pesticides, and to visualize the spatial distribution of the disease through a GIS map.

[0012] The multispectral data acquisition and preprocessing module includes a spectrum acquisition unit, a radiation correction unit, a geometric registration unit, and a noise suppression unit, wherein:

[0013] The spectrum acquisition unit is used to synchronously collect image data of crops in multiple bands through a multispectral camera array, covering the pest-sensitive spectral range of 400-1000nm;

[0014] The radiation correction unit is used to eliminate sensor response errors based on standard reflector and dark current calibration;

[0015] The geometric registration unit is used to perform sub-pixel alignment of multi-band images using the SIFT feature matching algorithm, with a registration error of <0.3 pixels;

[0016] The noise suppression unit is used to remove Gaussian noise and salt and pepper noise in the image by combining non-local mean filtering with wavelet transform.

[0017] The multispectral camera array is configured to synchronously collect image data of at least five spectral bands, and the at least five spectral bands include but are not limited to 450nm blue light, 550nm green light, 650nm red light, 760nm near infrared, and 850nm short wave infrared.

[0018] The visual intelligent analysis module includes a band screening unit, a feature fusion unit, a decoupling and reconstruction unit, and a sparse representation unit, wherein:

[0019] The band screening unit is used to calculate the weight of each spectrum through a cross-band attention mechanism, and dynamically screen the bands with high contribution to pest and disease identification;

[0020] The feature fusion unit is used to process the spatial-spectral joint features using a 3D convolutional network to generate a fusion feature map containing multi-dimensional information;

[0021] The decoupling and reconstruction unit: separates pest and disease characteristics from background noise based on an encoder-decoder architecture and represents them independently through mutual information constraints;

[0022] The sparse representation unit is used to compress the feature dimension through L1 regularization to generate a sparse feature vector focusing on sensitive information of pests and diseases.

[0023] The band screening unit calculates the weight of each spectrum through a cross-band attention mechanism, and dynamically screens the bands that contribute most to pest and disease identification. The specific operations are as follows:

[0024] A1: Process the reflectance curve of each band through the 1D convolution layer to generate the band feature vector f i ∈R d ;

[0025] A2: Calculate band weight α i , the specific formula is as follows:

[0026]

[0027] Where, f i is the eigenvector of the i-th band, is the mean eigenvector of all bands, W q 、W k is the learnable parameter matrix;

[0028] A3: retain weight α i Bands >0.15 constitute the optimized band combination.

[0029] The decoupling and reconstruction unit separates pest and disease features from background noise based on an encoder-decoder architecture and represents them independently through mutual information constraints. The specific operations are as follows:

[0030] B1: Encoder feature separation:

[0031] ① Using a dual-branch encoder network E=[E s ,E r ], where E s Extract spatial features z s , E r Extract spectral features z r ;

[0032] ② Input multispectral image Mapped to the latent space through convolutional layers and fully connected layers: z s =E s (X),z r =E r (X), where D s 、D r is the feature dimension;

[0033] B2: Mutual information constraint:

[0034] ① By minimizing the latent feature z p With background noise z b Decoupling the mutual information: Where, the mutual information I(·) is estimated approximately by a neural network;

[0035] ②Introduce adversarial training mechanism: Construct discriminator D to distinguish joint distribution p(z s ,z r ) and the marginal distribution product p(z s )p(z r ), the loss function is:

[0036] B3: Decoder reconstruction:

[0037] ① Design condition generator G(z s ,z r ), based on the conditional adversarial network structure, through z s and z r Reconstructing feature maps

[0038] ②Use multi-scale reconstruction loss function, including:

[0039] a. Pixel-level L1 loss:

[0040] b. Perceptual loss: Where VGG is a pre-trained visual feature extraction network;

[0041] B4: Decoupling evaluation and optimization:

[0042] ①Calculate the decoupling index When D < θ, parameter update is triggered;

[0043] ②The joint optimization objective function is: L = L L1 +λ1L per +λ2L adv -λ3I(zs ;z r ), where λ1=100, λ2=1, and λ3=0.1 are weight coefficients.

[0044] The intelligent diagnosis module includes a multi-model cascade unit, a confidence assessment unit, an explainability analysis unit, and a time series prediction unit, wherein:

[0045] The multi-model cascade unit is used to cascade Inception-v3 and LightGBM to perform end-to-end pest and disease recognition;

[0046] The confidence evaluation unit calculates the uncertainty of the prediction result based on Monte Carlo Dropout and outputs a 0-1 confidence score;

[0047] The explainability analysis unit generates a heat map through Grad-CAM to visualize the key decision areas;

[0048] The time series prediction unit is used to analyze historical data using the TCN time convolution network to predict the rate of change of disease spread area in the next 5 days.

[0049] The time series prediction unit uses the TCN temporal convolutional network to analyze historical data and predict the rate of change of disease spread area in the next 5 days. The specific operations are as follows:

[0050] C1: Collect pest and disease identification results, meteorological data, and soil moisture information from the past 14 days to construct a time series input vector;

[0051] C2: Input to a TCN network containing multiple dilated causal convolutional layers, where the dilation coefficient of each layer increases exponentially to expand the receptive field;

[0052] C3: The network output layer generates a vector of length 5, which represents the daily change rate of the disease spread area in the next 5 days;

[0053] C4: Train the TCN network by minimizing the following function:

[0054]

[0055] Where y t is the true diffusion rate, is the predicted value, γ is the regularization coefficient, and W is the network parameter;

[0056] C5: The prediction results are used to assist in formulating precise prevention and control operation plans.

[0057] The decision support module includes an agronomic knowledge matching unit, a dosage optimization unit, a spatiotemporal decision generation unit, and a multi-objective optimization unit, wherein:

[0058] The agronomic knowledge matching unit is used to retrieve pest and disease control thresholds and standard plans from the knowledge base according to crop varieties and growth cycles;

[0059] The dosage optimization unit dynamically calculates the dosage based on the density and severity of the lesions, balancing the control effect and environmental costs;

[0060] The spatiotemporal decision-making unit is used to map the diagnosis results to a GIS map and generate a pesticide application time and area planning scheme with a spatial heat map;

[0061] The multi-objective optimization unit is used to generate a trade-off decision plan between prevention and control efficiency, cost input and ecological impact through the Pareto optimal algorithm.

[0062] The present invention also proposes a method for intelligent monitoring of crop diseases and insect pests based on multispectral imaging, comprising the following steps:

[0063] S1: A multispectral camera array collects crop images in the 400-1000 nm band, and then processes them through radiometric correction, geometric registration, and noise suppression to generate a high-quality multispectral dataset.

[0064] S2: Utilizes a cross-band attention mechanism to dynamically select high-contribution spectral bands, fuses spatial-spectral features through a 3D convolutional network, and generates pest-sensitive sparse feature representations through decoupled reconstruction and L1 regularization;

[0065] S3: Identify the type and severity of pests and diseases based on the Inception-v3 and LightGBM cascade models, use Monte Carlo Dropout to evaluate prediction confidence, and use the TCN temporal convolutional network to predict the disease spread trend in the next 5 days;

[0066] S4: Based on the diagnosis results and agronomic knowledge base, the pesticide dosage and operation time are dynamically calculated, and a control plan with a spatial heat map is generated through a GIS map. The Pareto optimal algorithm is used to balance control efficiency, cost and ecological impact.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] This paper dynamically screens high-contribution spectral bands through a cross-band attention mechanism and combines a 3D convolutional network with a mutual information-constrained feature decoupling and reconstruction algorithm to achieve adaptive fusion and noise separation of multispectral features to generate a sparse feature representation focused on sensitive information of pests and diseases. Based on the multi-model fusion algorithm of Inception-v3 and LightGBM cascade, combined with Monte Carlo Dropout confidence evaluation, Grad-CAM interpretability analysis and TCN temporal convolutional network time series prediction, an intelligent diagnostic system with reliability quantification, decision visualization and trend prediction capabilities is constructed to achieve accurate identification of pest and disease types and severity and dynamic prediction of diffusion trends in the next 5 days. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a system diagram of the intelligent crop disease and insect pest monitoring system based on multispectral imaging of the present invention.

[0070] Figure 2 This is a flow chart of the intelligent crop disease and insect pest monitoring method based on multispectral imaging of the present invention.

[0071] Figure 3 Schematic diagram of hardware deployment in the intelligent crop disease and insect pest monitoring system based on multispectral imaging and its use method of the present invention.

[0072] Description of Figure Numbers:

[0073] 100. Multispectral data acquisition and preprocessing module; 101. Spectral acquisition unit; 102. Radiation correction unit; 103. Geometric registration unit; 104. Noise suppression unit; 200. Visual intelligent analysis module; 201. Band screening unit; 202. Feature fusion unit; 203. Decoupling and reconstruction unit; 204. Sparse representation unit; 300. Intelligent diagnosis module; 301. Multi-model cascade unit; 302. Confidence assessment unit; 303. Interpretability analysis unit; 304. Time series prediction unit; 400. Decision support module; 401. Agronomic knowledge matching unit; 402. Dose optimization unit; 403. Spatiotemporal decision generation unit; 404. Multi-objective optimization unit. DETAILED DESCRIPTION

[0074] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 any creative efforts shall fall within the scope of protection of the present invention.

[0075] Example 1:

[0076] like Figure 1 and Figure 3 As shown, this embodiment provides an intelligent crop disease and insect pest monitoring system based on multispectral imaging, including a multispectral data acquisition and preprocessing module 100, a visual intelligent analysis module 200, an intelligent diagnosis module 300, and a decision support module 400, wherein: the multispectral data acquisition and preprocessing module 100 is used to collect multispectral images of crops through a multispectral imaging device and preprocess the collected original images; the visual intelligent analysis module 200 is used to dynamically screen band contributions through a cross-band adaptive feature fusion network, and use a feature decoupling reconstruction algorithm to separate pest and disease sensitive features from background noise to generate a sparse feature representation; the intelligent diagnosis module 300 is based on a multi-model fusion algorithm with confidence assessment and interpretability to accurately identify the type and severity of pests and diseases, and combine time series analysis to predict the development trajectory of the disease; the decision support module 400 is used to combine the diagnosis results with the agronomic knowledge base to generate an executable plan including pesticide dosage, operation time, and recommended pesticides, and visualize the spatial distribution of the disease through a GIS map.

[0077] Among them, it should be noted that the multispectral data acquisition and preprocessing module 100 obtains and preprocesses crop multispectral images, the visual intelligent analysis module 200 dynamically screens sensitive bands and decouples pest and disease characteristics, and the intelligent diagnosis module 300 integrates multiple models to achieve accurate identification and trend prediction of pests and diseases. Finally, the decision support module 400 combines agronomic knowledge to generate a visual pesticide application plan.

[0078] In this embodiment, it should also be noted that the multispectral data acquisition and preprocessing module 100 includes a spectral acquisition unit 101, a radiation correction unit 102, a geometric registration unit 103, and a noise suppression unit 104. The spectral acquisition unit 101 is configured to synchronously acquire crop image data in multiple bands using a multispectral camera array, covering the pest-sensitive spectral range of 400-1000 nm. The multispectral camera array is configured to synchronously acquire image data in at least five spectral bands, including but not limited to 450 nm blue light, 550 nm green light, 650 nm red light, 760 nm near-infrared, and 850 nm short-wave infrared. The radiation correction unit 102 eliminates sensor response errors based on standard reflectors and dark current calibration. The geometric registration unit 103 uses the SIFT feature matching algorithm to perform sub-pixel alignment of the multi-band images, with a registration error of less than 0.3 pixels. The noise suppression unit 104 uses a combination of non-local mean filtering and wavelet transform to remove Gaussian noise and salt-and-pepper noise from the image.

[0079] It should be noted that the spectral acquisition unit 101 synchronously acquires crop images of at least five characteristic bands (including 450nm / 550nm / 650nm / 760nm / 850nm) in the range of 400-1000nm, and the radiation correction unit 102 eliminates sensor errors. The geometric registration unit 103 realizes sub-pixel multi-band alignment, and finally the noise suppression unit 104 fuses the non-local mean and wavelet transform to remove noise, completing the entire preprocessing process from raw data to standardized spectral feature matrix.

[0080] Furthermore, it should be noted that the DJI Zenmuse P1 multispectral camera is used to simultaneously collect five characteristic bands in the range of 400-1000nm (450nm blue light, 550nm green light, 650nm red light, 760nm near infrared, and 850nm short-wave infrared). The ground sampling distance reaches 2.5cm, which meets the needs of crop leaf-level disease identification. The radiation correction unit 102 deploys SpectraVue standard reflectors in the field and collects calibration data every 2 hours. Correct the sensor response, where I(λ) is the original pixel value, S(λ) is the sensor gain, and Ref(λ) is the standard reflectance. The noise suppression unit 104 first uses non-local means filtering (search window 11×11, similarity window 5×5) to remove Gaussian noise, and then suppresses salt and pepper noise through wavelet transform (db4 wavelet, 3-layer decomposition), thereby improving the image signal-to-noise ratio to above 35 dB.

[0081] In this embodiment, it should also be noted that the visual intelligent analysis module 200 includes a band screening unit 201, a feature fusion unit 202, a decoupling and reconstruction unit 203, and a sparse representation unit 204, wherein: the band screening unit 201 is used to calculate the weights of each spectrum through a cross-band attention mechanism and dynamically screen the bands with high contribution to pest and disease identification; the specific operations are as follows: A1: Process the reflectance curve of each band through a 1D convolution layer to generate a band feature vector f i ∈R d ; A2: Calculate band weight α i , the specific formula is as follows:

[0082]

[0083] Where, f i is the eigenvector of the i-th band, is the mean eigenvector of the band, W q 、W k is the learnable parameter matrix; A3: retained weight α iThe bands with a value greater than 0.15 constitute an optimized band combination. Feature fusion unit 202: is used to process spatial-spectral joint features using a 3D convolutional network to generate a fusion feature map containing multi-dimensional information; decoupling reconstruction unit 203: separates pest and disease features from background noise based on an encoder-decoder architecture and independently characterizes them through mutual information constraints; the specific operations are as follows: B1: Encoder feature separation: ① Using a dual-branch encoder network E = [E s ,E r ], where E s Extract spatial features z s , E r Extract spectral features z r ;② Input multispectral image Mapped to the latent space through convolutional layers and fully connected layers: z s =E s (X),z r =E r (X), where D s 、D r is the feature dimension; B2: mutual information constraint: ① by minimizing the potential feature z p With background noise z b Decoupling the mutual information: In the formula, the mutual information I(·) is estimated approximately by a neural network; ②Introducing the adversarial training mechanism: Constructing a discriminator D to distinguish the joint distribution p(z s ,z r ) and the marginal distribution product p(z s )p(z r ), the loss function is: B3: Decoder reconstruction: ① Design the condition generator G(z s ,z r ), based on the conditional adversarial network structure, through z s and z r Reconstructing feature maps ② Use multi-scale reconstruction loss function, including: a. Pixel-level L1 loss: b. Perceptual loss: Where VGG is a pre-trained visual feature extraction network; B4: Decoupling evaluation and optimization: ① Calculate the decoupling index When D < θ, the parameter update is triggered; ② The joint optimization objective function is: L = L L1 +λ1L per +λ2L adv -λ3I(z s ;z r), where λ1=100, λ2=1, and λ3=0.1 are weight coefficients. Sparse representation unit 204 is used to compress feature dimensions through L1 regularization to generate a sparse feature vector focusing on pest and disease sensitive information.

[0084] It should be noted that the band screening unit 201 dynamically selects key spectral bands based on the cross-band attention mechanism, the feature fusion unit 202 integrates spatial-spectral features using 3D convolution, and the decoupling reconstruction unit 203 uses the encoder-decoder architecture and mutual information constraint to separate pest and disease features from background noise. Finally, the sparse representation unit 204 generates a highly discriminative sparse feature vector through L1 regularization.

[0085] Furthermore, it should be noted that the band feature vector f i ∈R d In the example, d=64. The dual-branch encoder network of the decoupling and reconstruction unit 203 adopts the ResNet18 architecture. The spatial branch extracts the texture features of the edge of the lesion (such as the yellow rust-like texture of stripe rust), and the spectral branch captures the characteristics of the decrease in near-infrared reflectivity caused by chlorophyll attenuation.

[0086] In this embodiment, it should also be noted that the intelligent diagnosis module 300 includes a multi-model cascade unit 301, a confidence assessment unit 302, an interpretability analysis unit 303, and a time series prediction unit 304, wherein: the multi-model cascade unit 301: is used to cascade Inception-v3 and LightGBM for end-to-end pest and disease identification; the confidence assessment unit 302: calculates the uncertainty of the prediction result based on Monte Carlo Dropout, and outputs a 0-1 confidence score; the interpretability analysis unit 303: generates a heat map through Grad-CAM to visualize the key decision areas; the time series prediction unit 304: is used to use the TCN time convolutional network to analyze historical data and predict the rate of change of the disease spread area in the next 5 days. The specific operations are as follows: C1: Collect pest and disease identification results, meteorological data, and soil moisture information from the past 14 days to construct a time series input vector; C2: Input this vector into a TCN network containing multiple dilated causal convolutional layers, where the dilation coefficient of each layer increases exponentially to expand the receptive field; C3: The network output layer generates a vector of length 5, representing the daily change rate of the disease spread area in the next 5 days; C4: Train the TCN network by minimizing the following function:

[0087]

[0088] Where y t is the true diffusion rate, is the predicted value, γ is the regularization coefficient, and W is the network parameter; C5: The prediction results are used to assist in formulating accurate prevention and control operation plans.

[0089] It should be noted that the multi-model cascade unit 301 integrates Inception-v3 and LightGBM to achieve accurate identification of pests and diseases, the confidence assessment unit 302 quantifies the prediction reliability, the interpretability analysis unit 303 generates a heat map to reveal the decision basis, and uses the TCN network of the time series prediction unit 304 to predict the disease spread trend in the next 5 days.

[0090] Furthermore, it should be noted that a cascaded architecture of Inception-v3 (input size 299×299, output 2048-dimensional features) and LightGBM (tree depth = 7, number of leaves = 63) is used. The time series prediction unit 304 uses a 4-layer TCN network (expansion coefficients [1, 2, 4, 8], convolution kernel k = 3).

[0091] In this embodiment, it should also be noted that the decision support module 400 includes an agronomic knowledge matching unit 401, a dosage optimization unit 402, a spatiotemporal decision generation unit 403, and a multi-objective optimization unit 404, wherein: the agronomic knowledge matching unit 401 is used to retrieve pest and disease control thresholds and standard plans from the knowledge base according to crop varieties and growth cycles; the dosage optimization unit 402 dynamically calculates the pesticide dosage based on the density and severity of the lesions to balance the control effect and environmental costs; the spatiotemporal decision generation unit 403 is used to map the diagnostic results to a GIS map to generate a pesticide application time and regional planning plan with a spatial heat map; the multi-objective optimization unit 404 is used to generate a trade-off decision plan between control efficiency, cost input and ecological impact through the Pareto optimal algorithm.

[0092] It should be noted that the agronomic knowledge matching unit 401 calls the crop control knowledge base, the dosage optimization unit 402 calculates the precise pesticide application plan, and the spatiotemporal decision generation unit 403 realizes spatial visualization of the disease on the GIS platform. Finally, the Pareto optimal algorithm of the multi-objective optimization unit 404 is used to balance the control efficiency, economic cost and ecological impact.

[0093] Furthermore, it should be noted that the formula for calculating the dosage of the pesticide is: Where: D is the amount of pesticide applied (ml / m 2 ), S d is the density of lesions (pieces / cm 2 ), S v is the severity score (1-5), and α and β are crop growth stage parameters.

[0094] Example 2:

[0095] like Figure 2 and Figure 3As shown, in this embodiment, the intelligent monitoring method for crop diseases and insect pests based on multispectral imaging specifically includes the following steps:

[0096] S1: Multispectral data acquisition and standardization preprocessing

[0097] 1. Spectral acquisition:

[0098] Use the DJI Zenmuse P1 multispectral camera (configured with five bands: 450nm blue light, 550nm green light, 650nm red light, 760nm near infrared, and 850nm shortwave infrared);

[0099] Flight altitude 20m, ground resolution 2.5cm, frame rate 30fps;

[0100] Simultaneously acquire crop canopy reflectance data in the 400-1000nm range;

[0101] 2. Data preprocessing:

[0102] Radiation correction: Based on the SpectraVue standard reflector, according to the formula Calibration (I(λ) is the original pixel value, S(λ) is the sensor gain, and Ref(λ) is the standard reflectance);

[0103] Geometric registration: using the improved SIFT algorithm (feature point matching error < 0.3 pixels) to achieve multi-band sub-pixel alignment;

[0104] Noise suppression: First, non-local means filtering (11×11 search window) is performed, and then db4 wavelet 3-layer decomposition is implemented for denoising;

[0105] S2: Intelligent cross-modal feature analysis

[0106] 1. Dynamic band screening:

[0107] Extract the feature vector f of each band through 1D convolution i ∈R d ;

[0108] According to the specific formula Q=W q f i , Calculate weight (d = 64);

[0109] Retention weight α i Bands >0.15 constitute an optimized combination;

[0110] 2. Feature decoupling and reconstruction:

[0111] Coding phase:

[0112] Spatial branch: ResNet18 extracts texture features (output 256 dimensions);

[0113] Spectral branch: 3-layer 1D CNN extracts spectral features (output 128 dimensions);

[0114] Decoupling constraints: Minimize mutual information I(z p ;z b )≤0.1bit;

[0115] Reconstructed output: L1 loss (weight 0.7) + perceptual loss (VGG16 features, weight 0.3) are jointly optimized;

[0116] S3: Multi-model collaborative diagnosis and prediction

[0117] 1. Identification of pests and diseases:

[0118] Inception-v3 extracts 2048-dimensional image features, reduces them to 128 dimensions through PCA, and then inputs them into LightGBM (maximum depth 7);

[0119] Output the type and severity of pests and diseases (1-5 rating);

[0120] 2. Time Series Forecasting:

[0121] Construct a 14-day historical data time series (including 8 features such as disease index, temperature, and humidity);

[0122] The TCN network configuration is:

[0123] 4 residual blocks (expansion coefficients [1, 2, 4, 8]);

[0124] Each block contains: dilated causal convolution (kernel size =3), LayerNorm layer, ReLU activation;

[0125] Output layer: linear full connection (output dimension 5);

[0126] Output the predicted value of the disease area change rate in the next 5 days

[0127] S4: Precision pesticide application decision generation

[0128] 1. Knowledge matching:

[0129] Calling the control threshold library according to crop varieties (such as winter wheat) and growth stages (such as jointing stage);

[0130] Example: Stripe rust control threshold = lesion density 5 / cm 2 ;

[0131] 2. Dose optimization:

[0132] According to the formula Calculate the application rate (α = 1.2, β = 0.8 are winter wheat parameters);

[0133] Constraints: Environmental carrying capacity ≤ 15ml / m 2 ;

[0134] 3.GIS visualization:

[0135] Generate heat maps on the ArcGIS Pro platform:

[0136] Red warning zone (lesion density > threshold 120%);

[0137] Yellow monitoring zone (threshold 80%-120%);

[0138] Green safe zone (<80%).

[0139] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0140] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent crop disease and insect pest monitoring system based on multispectral imaging, characterized by: The system comprises a multispectral data acquisition and preprocessing module (100), a visual intelligent analysis module (200), an intelligent diagnosis module (300), and a decision support module (400), wherein: The multispectral data acquisition and preprocessing module (100) is used to acquire multispectral images of crops through a multispectral imaging device and preprocess the acquired original images; The visual intelligent analysis module (200) is used to dynamically screen band contributions through a cross-band adaptive feature fusion network, and to separate pest-sensitive features from background noise using a feature decoupling and reconstruction algorithm to generate a sparse feature representation; The intelligent diagnosis module (300) accurately identifies the type and severity of pests and diseases based on a multi-model fusion algorithm with confidence assessment and interpretability, and predicts the development trajectory of the disease in combination with time series analysis; The decision support module (400) is used to combine the diagnosis results with the agronomic knowledge base to generate an executable plan including the application dosage, operation time, and recommended pesticides, and to visualize the spatial distribution of the disease through a GIS map.

2. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 1, characterized in that: The multispectral data acquisition and preprocessing module (100) comprises a spectrum acquisition unit (101), a radiation correction unit (102), a geometric registration unit (103), and a noise suppression unit (104), wherein: The spectrum acquisition unit (101) is used to synchronously acquire image data of crops in multiple bands through a multispectral camera array, covering a spectrum sensitive to pests and diseases in the range of 400-1000 nm; The radiation correction unit (102) eliminates sensor response errors based on standard reflector and dark current calibration; The geometric registration unit (103) is used to perform sub-pixel alignment of multi-band images using a SIFT feature matching algorithm, with a registration error of <0.3 pixels; The noise suppression unit (104) is used to remove Gaussian noise and salt and pepper noise in the image by combining non-local mean filtering with wavelet transform.

3. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 2, characterized in that: The multispectral camera array is configured to synchronously collect image data of at least five spectral bands, and the at least five spectral bands include but are not limited to 450nm blue light, 550nm green light, 650nm red light, 760nm near infrared, and 850nm short wave infrared.

4. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 1, characterized in that: The visual intelligent analysis module (200) comprises a band screening unit (201), a feature fusion unit (202), a decoupling and reconstruction unit (203), and a sparse representation unit (204), wherein: The band screening unit (201) is used to calculate the weight of each spectrum through a cross-band attention mechanism, and dynamically screen the bands with high contribution to the identification of pests and diseases; The feature fusion unit (202) is used to process the spatial-spectral joint features using a 3D convolutional network to generate a fusion feature map containing multi-dimensional information; The decoupling and reconstruction unit (203) is configured to separate pest and disease characteristics from background noise based on an encoder-decoder architecture and independently represent them through mutual information constraints; The sparse representation unit (204) is used to compress the feature dimension through L1 regularization to generate a sparse feature vector focusing on sensitive information of pests and diseases.

5. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 4 is characterized in that: The band screening unit (201) calculates the weight of each spectrum through a cross-band attention mechanism, and dynamically screens the bands with high contribution to pest and disease identification. The specific operation is as follows: A1: Process the reflectance curve of each band through the 1D convolution layer to generate the band feature vector f i ∈R d ; A2: Calculate band weight α i , the specific formula is as follows: Where, f i is the eigenvector of the i-th band, is the mean eigenvector of all bands, W q 、W k is the learnable parameter matrix; A3: retain weight α i Bands >0.15 constitute the optimized band combination.

6. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 5, characterized in that: The decoupling and reconstruction unit (203) separates the pest and disease characteristics from the background noise based on the encoder-decoder architecture and independently characterizes them through mutual information constraints. The specific operations are as follows: B1: Encoder feature separation: ① Using a dual-branch encoder network E=[E s ,E r ], where E s Extract spatial features z s , E r Extract spectral features z r ; ② Input multispectral image Mapped to the latent space through convolutional layers and fully connected layers: z s =E s (X),z r =E r (X), where D s 、D r is the feature dimension; B2: Mutual information constraint: ① By minimizing the latent feature z p With background noise z b Decoupling the mutual information: Where, the mutual information I(·) is estimated approximately by a neural network; ②Introduce adversarial training mechanism: Construct discriminator D to distinguish joint distribution p(z s ,z r ) and the marginal distribution product p(z s )p(z r ), the loss function is: B3: Decoder reconstruction: ① Design condition generator G(z s ,z r ), based on the conditional adversarial network structure, through z s and z r Reconstructing feature maps ②Use multi-scale reconstruction loss function, including: a. Pixel-level L1 loss: b. Perceptual loss: Where VGG is a pre-trained visual feature extraction network; B4: Decoupling evaluation and optimization: ①Calculate the decoupling index When D < θ, parameter update is triggered; ②The joint optimization objective function is: L = L L1 +λ1L per +λ2L adv -λ3I(z s ;z r ), where λ1=100, λ2=1, and λ3=0.1 are weight coefficients.

7. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 1, characterized in that: The intelligent diagnosis module (300) includes a multi-model cascade unit (301), a confidence evaluation unit (302), an explainability analysis unit (303), and a time series prediction unit (304), wherein: The multi-model cascade unit (301) is used to cascade Inception-v3 and LightGBM to perform end-to-end pest and disease recognition; The confidence evaluation unit (302) calculates the uncertainty of the prediction result based on Monte Carlo Dropout and outputs a confidence score of 0-1; The explainability analysis unit (303) generates a heat map through Grad-CAM to visualize the key decision areas; The time series prediction unit (304) is used to analyze historical data using a TCN temporal convolutional network to predict the rate of change of the disease spread area in the next five days.

8. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 7, characterized in that: The time series prediction unit (304) uses a TCN temporal convolutional network to analyze historical data and predict the rate of change of the disease spread area in the next 5 days. The specific operation is as follows: C1: Collect pest and disease identification results, meteorological data, and soil moisture information from the past 14 days to construct a time series input vector; C2: Input to a TCN network containing multiple dilated causal convolutional layers, where the dilation coefficient of each layer increases exponentially to expand the receptive field; C3: The network output layer generates a vector of length 5, which represents the daily change rate of the disease spread area in the next 5 days; C4: Train the TCN network by minimizing the following function: Where y t is the true diffusion rate, is the predicted value, γ is the regularization coefficient, and W is the network parameter; C5: The prediction results are used to assist in formulating precise prevention and control operation plans.

9. The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to claim 1, characterized in that: The decision support module (400) includes an agronomic knowledge matching unit (401), a dosage optimization unit (402), a spatiotemporal decision generation unit (403), and a multi-objective optimization unit (404), wherein: The agronomic knowledge matching unit (401) is used to retrieve pest control thresholds and standard solutions from a knowledge base based on crop varieties and growth cycles; The dosage optimization unit (402) dynamically calculates the dosage of the pesticide based on the density and severity of the lesions, balancing the control effect and the environmental cost; The spatiotemporal decision generating unit (403) is used to map the diagnosis results to a GIS map and generate a pesticide application time and area planning scheme with a spatial heat map; The multi-objective optimization unit (404) is used to generate a trade-off decision plan among prevention and control efficiency, cost input and ecological impact through a Pareto optimal algorithm.

10. An intelligent crop disease and insect pest monitoring method based on multispectral imaging, characterized by: The intelligent crop disease and insect pest monitoring system based on multispectral imaging according to any one of claims 1 to 9 comprises the following steps: S1: A multispectral camera array collects crop images in the 400-1000nm band, and then processes them through radiometric correction, geometric registration, and noise suppression to generate a high-quality multispectral dataset. S2: Utilizes a cross-band attention mechanism to dynamically select high-contribution spectral bands, fuses spatial-spectral features through a 3D convolutional network, and generates pest-sensitive sparse feature representations through decoupling reconstruction and L1 regularization; S3: Identify the type and severity of pests and diseases based on the Inception-v3 and LightGBM cascade models, use Monte Carlo Dropout to evaluate prediction confidence, and use the TCN temporal convolutional network to predict the disease spread trend in the next 5 days; S4: Based on the diagnosis results and agronomic knowledge base, the pesticide dosage and operation time are dynamically calculated, and a control plan with a spatial heat map is generated through a GIS map. The Pareto optimal algorithm is used to balance control efficiency, cost and ecological impact.

Images