Rice blast early diagnosis method based on hyperspectral imaging

Abstract

The invention discloses a hyperspectral imaging-based rice blast early diagnosis method, and relates to the technical field of agricultural disease monitoring, and the specific steps of the method are as follows: S100, obtaining spectral data of rice in a region, S200, constructing a healthy tissue end member spectrum and predicting a lesion tissue end member spectrum, S300, constructing a mixed pixel model, and combining constraint conditions to determine the early diagnosis of rice blast. According to the method, through the mixed pixel model of the time sequence dynamic end members, the fine change of the spectrum can be accurately captured in the early stage of rice blast infection, through continuous time sequence hyperspectral imaging, the real-time spectrum of the rice blast can be accurately captured, and the real-time spectrum of the rice blast can be obtained. And learning dynamic evolution of a lesion end member spectrum by combining an LSTM network, regarding a mixed pixel at each time point as a mixture of a healthy tissue end member spectrum and a gradually evolved lesion tissue end member spectrum, and inverting the state and abundance of the lesion tissue end member spectrum at each time point so as to quantitatively monitor the early development progress of infection.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural disease monitoring, and in particular to an early diagnosis method for rice blast based on hyperspectral imaging. Background Art

[0002] Rice blast, as a disease in rice production, can lead to a significant reduction in rice yield. Traditional disease detection mainly relies on manual observation or laboratory analysis, which has problems such as detection lag, strong subjectivity, and low efficiency, and cannot meet the needs of precise management of modern agriculture. Hyperspectral imaging technology can capture the subtle spectral characteristics of crop physiological and biochemical changes. By analyzing the reflectivity changes of rice leaves in different bands, it can achieve early detection of rice blast infection.

[0003] At present, there is a long incubation period from spore germination to symptom manifestation in rice blast, and the spectral characteristics of the diseased tissue evolve dynamically with the infection process. Traditional static spectral analysis methods are difficult to accurately track the development process of the disease. In terms of endmember spectrum construction, fixed endmember spectral models are mostly used, ignoring the dynamic change characteristics of the spectrum of diseased tissue over time. During the rice blast infection process, physiological changes such as chlorophyll degradation, cell wall destruction, and water loss inside the leaves will cause the spectral characteristics to continue to evolve. The fixed endmember model cannot accurately reflect this process, resulting in a decrease in the accuracy of mixed pixel decomposition, affecting the reliability of diagnosis.

[0004] Therefore, there is an urgent need for a diagnostic method that can dynamically track the spectral evolution characteristics of rice blast and effectively process time-series hyperspectral data to achieve accurate monitoring of the entire cycle of rice blast from the incubation period to the manifestation period. Summary of the Invention

[0005] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide an early diagnosis method for rice blast based on hyperspectral imaging. It can accurately capture subtle spectral changes in the early stage of rice blast infection through a mixed pixel model of time-series dynamic endmembers. Through continuous time-series hyperspectral imaging, combined with the LSTM network to learn the dynamic evolution of the lesion endmember spectrum, the mixed pixel at each time point is regarded as a mixture of the healthy tissue endmember spectrum and the gradually evolving lesion tissue endmember spectrum, and the state of the lesion tissue endmember spectrum at each time point and its abundance in the pixel are inverted, thereby non-invasively and quantitatively monitoring the early development progress of the infection.

[0006] In order to solve the above technical problems, the present invention provides the following technical solution: an early diagnosis method for rice blast based on hyperspectral imaging, wherein the specific steps of the method are as follows:

[0007] S100, use a hyperspectral imager to perform continuous time-series imaging of the target rice leaf area to obtain the spectral data Y of the rice in the area t ;

[0008] S200, construct healthy tissue endmember spectrum E h , and predict the endmember spectrum E of the diseased tissue evolving over time d (t);

[0009] S300, based on the spectrum Y of each pixel in the hyperspectral image collected at each time point t in S100 t , construct a mixed pixel model, combine the constraints, and calculate the abundance f of the diseased tissue at each time point d (t) and the actual endmember spectrum E′ of the diseased tissue d (t);

[0010] S400, based on f d (t) and E′ d The spectral evolution parameters of (t) generate a quantitative index of infection progress.

[0011] Furthermore, the S100 uses a hyperspectral imager with a spectral range of 400-2500nm to continuously image the target rice leaf area at a time interval of Δt. The time interval Δt is dynamically and adaptively adjusted according to the pathogenesis of rice blast and the characteristics of the rice variety, including:

[0012] Initial stage: During the blast spore germination period, Δt = 6 hours was set to capture the spectral changes during the spore germination period;

[0013] Mid-stage: During the appressorium formation period, Δt = 12 hours, tracking the spectral characteristics of the appressorium formation period;

[0014] Late stage: During the lesion expansion period, Δt = 24 hours, monitor the dynamics of the lesion expansion period;

[0015] After the collected hyperspectral image data is stored in time series, it is preprocessed to remove the high-frequency noise generated by the equipment and the environment to form time series data. For each time point t, the corresponding spectral data is expressed as y t , the y t A vector containing reflectance information of multiple bands, reflecting the spectral characteristics of the leaf at that moment.

[0016] Furthermore, the healthy tissue end member spectrum E in S200 h The construction process is as follows: before rice blast infection, the area is imaged hyperspectrally for n ≥ 3 times, and the average spectrum of the healthy pixel set is calculated as the end member spectrum of the healthy tissue. Among them, i=1,2,…,n,E h represents the endmember spectrum of healthy tissue, n is the number of imaging times, S i represents the spectral data obtained by the i-th imaging;

[0017] Endmember spectrum of diseased tissue Ed The construction process of (t) is as follows: select k historical moments, k ≥ 3, of the target rice leaf area at different time points. [t-k:t-1] and environmental parameters P of temperature and humidity env , the Y [t-k:t-1] Contains the spectrum information from time tk to time t-1, and the environmental parameter P env With historical spectral data Y [t-k：t-1] As the input of the LSTM network, combined with the end member spectrum E of the diseased tissue at the previous moment d (t-1), through E d (t) = E d (t-1)+LSTM(Y [t-k:t-1] ,P env ; θ) to predict the endmember spectrum E of the diseased tissue at the current time t d (t), continuously update E with new input data d (t), realizing the dynamic construction of the time evolution of the endmember spectrum of the diseased tissue.

[0018] Furthermore, the functional relationship of the pixel spectrum model in S300 is: t =f h (t)·E h +f d (t)·E d (t)+∈, where Y t is the pixel spectral vector at time point t, which contains the reflectance value of the pixel in each band, f h (t) and f d (t) are the abundance of healthy tissue and diseased tissue at time point t, respectively, and f h (t) = 1 - f d (t), E h is the constructed healthy tissue endmember spectral vector, E d (t) is the endmember spectrum vector of the diseased tissue predicted based on the LSTM network, and ∈ is the residual vector, which represents the model error.

[0019] Furthermore, the constraints in S300 include physical constraints and timing constraints, wherein:

[0020] Physical constraints include abundance non-negativity and abundance sum being 1, where the abundance non-negativity is: h (t)≥0,f d (t)≥0, the abundance sum is 1:f h (t)+f d (t) = 1, indicating that the pixel consists of only two end-members: healthy and diseased;

[0021] Temporal constraints include lesion abundance smoothness and spectral evolution continuity, where lesion abundance smoothness is |f d (t)-f d (t-1)|≤δ·Δt, where δ is the maximum change rate threshold, ranging from 0.05% to 0.2%, and Δt is the time interval of acquisition. The continuity of the spectral evolution is Among them, γ is the spectral change rate threshold, and its value range is 0.001-0.01.

[0022] Furthermore, the S301, initialization parameters: obtaining the initial predicted value of the lesion end member spectrum at the current moment from S200, using the lesion abundance value at the previous moment as the initial value at the current moment;

[0023] S302, updating the lesion abundance: constructing a linear equation system based on the current lesion endmember spectrum and the healthy endmember spectrum, solving the equation system to obtain a preliminary lesion abundance value, and applying the lesion abundance smoothness conditions of physical constraints and temporal constraints;

[0024] S303, updating the lesion endmember spectrum: based on the current lesion abundance value, constructing a linear equation system, solving the equation system to obtain a preliminary lesion endmember spectrum, and applying a time sequence constraint spectrum evolution continuity condition;

[0025] S304: Convergence determination: Calculate the abundance change between the current iteration and the previous iteration, calculate the spectrum change between the current iteration and the previous iteration, and perform threshold determination.

[0026] S305 , outputting decomposition results: taking the lesion abundance obtained in the last iteration as the final result, and taking the lesion end-member spectrum obtained in the last iteration as the final result.

[0027] Furthermore, the S400 is based on the abundance of diseased tissue f at each time point. d (t) and the actual endmember spectrum E′ of the diseased tissue d (t), generates a quantitative index of infection progress Here, α and β are weight coefficients and α=0.6 and β=0.4.

[0028] Furthermore, the S400 classifies the infection progress level according to the calculated infection progress quantitative index I(t):

[0029] Low infection level: 0≤I(t)<0.3, indicating that rice blast infection is in the early stage, with a small lesion area and slow development;

[0030] Medium infection level: 0.3≤I(t)<0.7, indicating that rice blast infection is in the development stage, with the lesion area gradually expanding and the development speed accelerating;

[0031] High infection level: I(t) ≥ 0.7, indicating that rice blast infection is at a serious stage, with a large lesion area and rapid development;

[0032] The infection progress quantification index I(t) and the corresponding infection progress grade are output as the diagnosis results, providing a quantitative basis for rice blast prevention and control decisions.

[0033] Compared with existing technologies, this method for early diagnosis of rice blast based on hyperspectral imaging has the following beneficial effects:

[0034] 1. The present invention uses a mixed pixel model of time-series dynamic endmembers to accurately capture subtle spectral changes in the early stages of rice blast infection. Through continuous time-series hyperspectral imaging and combined with the LSTM network to learn the dynamic evolution of the lesion endmember spectrum, the mixed pixel at each time point is regarded as a mixture of the healthy tissue endmember spectrum and the gradually evolving lesion tissue endmember spectrum. The state of the lesion tissue endmember spectrum at each time point and its abundance in the pixel are inverted, thereby non-invasively and quantitatively monitoring the early development progress of the infection.

[0035] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0037] Figure 1 This is an operational flow chart of the early diagnosis method for rice blast based on hyperspectral imaging;

[0038] Figure 2 This is a flowchart of the early diagnosis method of rice blast based on hyperspectral imaging. DETAILED DESCRIPTION

[0039] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0040] Example 1

[0041] This embodiment provides a method for early diagnosis of rice blast based on hyperspectral imaging, such as Figure 2 As shown in the figure, by constructing a mixed pixel model of time-series dynamic endmembers, combining hyperspectral imaging technology with LSTM neural network, full-cycle spectral monitoring of rice blast from spore germination to lesion expansion is achieved. This method collects hyperspectral data of rice leaves by dynamically adjusting the time interval, uses static modeling of healthy tissue endmember spectra and dynamic prediction of diseased tissue endmember spectra, and combines physical constraints and time-series constraints to perform mixed pixel decomposition. Finally, a quantitative index of infection progress is generated to achieve disease grade classification, providing a quantitative basis for early and accurate diagnosis and prevention of rice blast.

[0042] First, the time series acquisition and preprocessing stage of hyperspectral data (S100) is entered. Continuous spectral data of the rice leaf area is obtained through a hyperspectral imager to provide basic information for subsequent analysis. Based on the spectral resolution advantage of hyperspectral imaging technology, it is possible to capture subtle changes in the reflectance of leaves in different bands. These changes are directly related to the internal physiological and biochemical state of the leaves (chlorophyll content, degree of cell structure damage). During the infection process of rice blast, the metabolic activity of the pathogen will cause the leaf tissue to be gradually damaged, and the response of this damage at the spectral level is earlier than the lesions visible to the naked eye. Specifically, a hyperspectral imager with a spectral range of 400-2500nm is selected. This band range covers the absorption peaks of plant photosynthesis-related pigments (chlorophyll, carotenoids) and the characteristic bands of water absorption, which can sensitively reflect changes in the physiological state of leaves. The dynamic adjustment mechanism of the time interval Δt is based on the infection law of rice blast: during the spore germination period (initial stage), the pathogen has not yet caused obvious tissue damage, but physiological metabolic activities have begun to affect the leaves. Spectra, at this time, need to be collected at a high frequency with an interval of 6 hours to ensure that the early weak spectral disturbances are captured. The role is to improve the recognition rate of early disease signals through high-density time sampling, and avoid the loss of spectral features in the incubation period due to too long sampling intervals; entering the appressorium formation period (middle stage), the pathogen begins to penetrate the leaf epidermis, tissue damage gradually appears, and the rate of change of spectral features accelerates, so the interval is extended to 12 hours, while ensuring data continuity and optimizing acquisition efficiency, the effect is to balance data integrity and monitoring cost; in the lesion expansion period (late stage), the disease development enters a rapid period, the spectral feature changes tend to be stable but the spatial heterogeneity increases, and a 24-hour interval can meet the monitoring needs, reduce redundant data collection, and focus on the dynamic monitoring of the spatial expansion of the disease; the collected raw data contains equipment noise and environmental interference (ambient light intensity changes, leaf jitter), so the high-frequency noise is removed by preprocessing, the preprocessing includes radiation calibration (converting pixel values ​​into actual reflectance) and noise removal. The processed time series data is represented by Y t , the Y tIt is a multidimensional vector, each dimension corresponds to the reflectance value of a band. The temporal change trajectory of the vector reflects the evolution of the leaf spectral characteristics with the development of the disease, laying a data foundation for subsequent analysis.

[0043] Then, the endmember spectrum construction and dynamic prediction stage (S200) is entered, which includes the static construction of the endmember spectrum of healthy tissue and the dynamic prediction of the endmember spectrum of diseased tissue. The endmember represents the spectral characteristics of pure substances, and the abundance represents the proportion of the endmember in the pixel. For the endmember spectrum of healthy tissue E h Construction: Before rice blast infection (i.e., healthy control group), the target area is subjected to n (n ≥ 3) repeated hyperspectral imaging, and the spectral data S of each pixel is extracted. i , the end member is constructed by averaging the spectrum, and the formula is: By eliminating random noise through statistical averaging, a stable healthy spectrum benchmark is established, and this end member is used as the benchmark for subsequent mixed pixel decomposition; for the end member spectrum E of the diseased tissue d Dynamic prediction of (t): This embodiment takes into account the characteristics of the dynamic evolution of the spectrum of the diseased tissue over time and uses LSTM (long short-term memory network) to perform time series modeling. LSTM can capture the dependencies in long sequences. Specifically, select the spectral data Y of k (k ≥ 3) moments in history [t-k:t-1] And the corresponding temperature and humidity environmental parameters P env As input, Y [t-k:t-1] Contains the spectral information from time tk to time t-1, reflecting the historical evolution trajectory of the spectrum of the diseased tissue, P env As external driving factors, such as temperature increase, which accelerates the reproduction of pathogens and humidity increase, which promotes spore germination, the two jointly affect the rate of disease development. By integrating multi-dimensional time series data to improve the accuracy of lesion spectrum prediction, LSTM can be used to predict the disease spectrum through E d (t) = E d (t-1)+LSTM(Y [t-k:t-1] ,P env ; θ), where θ is the bias term of the LSTM network. This formula indicates that the spectrum of the diseased end member at the current moment is composed of the spectrum at the previous moment and the spectrum increment predicted by LSTM, which reflects the continuity of the spectrum evolution. By continuously inputting new data, the model can adaptively update E d (t), thereby dynamically tracking the spectral changes of diseased tissues and more accurately reflecting the dynamic evolution of the spectrum caused by pathogen infection.

[0044] Next, we enter the mixed pixel decomposition and parameter optimization stage (S300). Based on the mixed pixel model: Y t =f h (t)·E h +f d (t)·E d(t)+∈, solve the abundance f of diseased tissue through constraint conditions d (t) and the actual end member spectrum E′ d (t), the model sets the pixel to consist of only two end-members, healthy and diseased, ∈ is the residual of the model, and the constraints include physical constraints and temporal constraints, where: physical constraints include abundance non-negativity and abundance sum is 1, and the abundance non-negativity: f h (t)≥0 and f d (t) ≥ 0, reflecting the actual existence of healthy and diseased tissues in the pixel. There is no physical entity of "negative proportion", ensuring the physical interpretability of the solution and avoiding the negative abundance that does not conform to biological significance. The sum of the abundance is 1: f h (t)+f d (t) = 1, indicating that the pixel is composed of only two end members, eliminating the interference of other end members (such as soil and weeds), improving the calculation efficiency and stability. The temporal constraints include lesion abundance smoothness and spectral evolution continuity. The lesion abundance smoothness: |f d (t)-f d (t-1)|≤δ·Δt, where δ is the maximum change rate threshold (range 0.05%-0.2%). The introduction of Δt makes the threshold related to the time interval. For example, when Δt=6, the maximum allowed change is δ×6, ensuring the consistency of the constraint conditions in different acquisition stages and suppressing abundance mutations caused by noise or model errors. The spectral evolution continuity is: Where γ is the spectral change rate threshold (0.001-0.01). This condition requires that the temporal change rate of the diseased endmember spectrum is within a reasonable range to avoid abnormal fluctuations caused by model noise, reflecting the gradual change of spectral characteristics with the development of the disease. The specific steps of the mixed pixel decomposition and parameter optimization stage are as follows:

[0045] S301, initialization: setting initial abundance and the initial value of the diseased endmember spectrum

[0046] S302, Fixed E d (t), updated abundance f d (t): Combined constraints Through the pixel spectral model function relationship Solution Where i is the number of iterations, represents the value of the lesion abundance at time t at the i-th iteration, represents the value of the endmember spectrum of the diseased tissue at time t at the i-1th iteration;

[0047] S303, fixed fd (t), update the lesion endmember spectrum E d (t): Combined constraints: Through the pixel spectral model function relationship Solution

[0048] S304, Convergence judgment: When the and The iteration stops when ε1=10 -3 and ε2=10 -2 is the convergence threshold;

[0049] S305, result output: by alternately updating the abundance and endmember spectrum, the final lesion abundance and the final diseased endmember spectrum They represent the abundance of diseased tissue at time t and the actual endmember spectrum, respectively, and are used to quantitatively describe the disease state of the pixel.

[0050] Finally, the infection progress quantification and classification stage (S400) is entered. By integrating the dynamic information of the lesion abundance and the endmember spectrum, the infection progress quantification index I(t) is generated. Taking into account the abundance of the disease and the spectral change rate, the formula is: Among them, α = 0.6 and β = 0.4 are weight coefficients, which reflect the dominant role of abundance in disease diagnosis, while taking into account the dynamic characteristics of spectral evolution. The time derivative of the lesion endmember spectrum reflects the rate of change of the spectral characteristics per unit time, converting the spectral change into a quantifiable dynamic index. The infection progress level is divided according to the calculated infection progress quantitative index I(t):

[0051] Low infection level (0≤I(t)<0.3): corresponds to the spore germination stage, when the pathogen has just invaded the leaf and is only multiplying in local cells, without causing obvious tissue damage;

[0052] Medium infection level (0.3≤I(t)<0.7): With the formation of appressorium and expansion of hyphae, the number of lesions gradually increases. d (t) rises, and the rate of spectrum change accelerates. At this time, the disease enters the development stage, and chemical or biological control measures must be taken in time to curb the spread of the disease;

[0053] High infection level (I(t) ≥ 0.7): Lesions have spread over a large area, and spectral characteristics show significant abnormalities. Emergency prevention and control plans are initiated. Significant abnormalities in spectral characteristics (such as disordered reflectance across the entire spectrum) indicate that leaf function is severely damaged. At this time, an emergency integrated prevention and control plan must be initiated to minimize yield losses.

[0054] The diagnostic results are output in the form of I(t) values ​​and corresponding grades, providing agricultural producers with an intuitive basis for decision-making. Through standardized grade division, disease data from different fields and different years are comparable, supporting the optimization of long-term plant protection strategies.

[0055] In summary, this embodiment organically combines hyperspectral imaging technology with LSTM neural network to construct a dynamic and accurate early diagnosis system for rice blast. The time series data acquisition strategy of S100 ensures spectral coverage of the entire disease process, the end-member modeling method of S200 breaks through the limitations of traditional static models, the constrained optimization process of S300 improves the reliability of mixed pixel decomposition, and the quantitative index of S400 converts complex spectral information into intuitive prevention and control decision-making indicators. Through the method of this embodiment, the efficiency and accuracy of early diagnosis of rice blast can be significantly improved.

[0056] Example 2

[0057] like Figure 1 As shown, this embodiment provides a workflow for early detection and diagnosis of rice blast based on hyperspectral imaging. The specific steps of the workflow are as follows:

[0058] Use a hyperspectral imager with a spectral range of 400-2500nm;

[0059] To set a dynamic time interval:

[0060] Initial stage (spore germination stage): data is collected every 6 hours to capture subtle changes in the early spectrum;

[0061] Mid-term (appressorium formation period): collect data every 12 hours to track the key stages of disease development;

[0062] Late stage (lesion expansion stage): collect samples every 24 hours to monitor the spread of the disease;

[0063] Continuously shoot the target rice leaf area and automatically store time series image data;

[0064] Remove equipment noise and environmental high-frequency interference to generate standardized spectral vector data;

[0065] Before disease infection, the healthy leaf area was imaged ≥3 times, and the average spectrum of healthy pixels was extracted as the healthy endmember spectrum;

[0066] Using the spectral data and temperature and humidity parameters of the past k moments, the LSTM network is used to predict the spectrum of the diseased endmember at the current moment and iteratively updates it with new data.

[0067] Each pixel spectrum is considered as a linear combination of the healthy and diseased endmember spectra. Combined with physical constraints (abundance is non-negative and sum is 1) and temporal constraints (abundance is smooth and spectrum evolution is continuous), a mixed pixel model is constructed.

[0068] Output the abundance of diseased tissue and actual diseased endmember spectrum at the current moment;

[0069] Calculate quantitative index: Generate a quantitative index of infection progress based on the lesion abundance and endmember spectrum change rate, which comprehensively reflects the severity of the disease;

[0070] Classify the infection levels: low infection, medium infection, high infection;

[0071] Generate reports containing quantitative indices and grades to provide a basis for prevention and control decisions.

[0072] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for early diagnosis of rice blast based on hyperspectral imaging, characterized in that: The specific steps of this method are: S100, use a hyperspectral imager to perform continuous time-series imaging of the target rice leaf area to obtain the spectral data Y of the rice in the area t ; S200, construct healthy tissue endmember spectrum E h , and predict the endmember spectrum E of the diseased tissue evolving over time d (t); S300, based on the spectrum Y of each pixel in the hyperspectral image collected at each time point t in S100 t , construct a mixed pixel model, combine the constraints, and calculate the abundance f of the diseased tissue at each time point d (t) and the actual endmember spectrum E′ of the diseased tissue d (t); S400, based on f d (t) and E′ d The spectral evolution parameters of (t) generate a quantitative index of infection progress.

2. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 1, characterized in that: The S100 uses a hyperspectral imager with a spectral range of 400-2500nm to continuously image the target rice leaf area at a time interval of Δt. The time interval Δt is dynamically and adaptively adjusted according to the pathogenesis of rice blast and the characteristics of the rice variety, including: Initial stage: During the blast spore germination period, Δt = 6 hours was set to capture the spectral changes during the spore germination period; Mid-stage: During the appressorium formation period, Δt = 12 hours, tracking the spectral characteristics of the appressorium formation period; Late stage: During the lesion expansion period, Δt = 24 hours, monitor the dynamics of the lesion expansion period; After the collected hyperspectral image data is stored in time series, it is preprocessed to remove the high-frequency noise generated by the equipment and the environment to form time series data. For each time point t, the corresponding spectral data is expressed as y t , the y t A vector containing reflectance information of multiple bands, reflecting the spectral characteristics of the leaf at that moment.

3. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 1, characterized in that: The healthy tissue end member spectrum E in S200 h The construction process is as follows: before rice blast infection, the area is imaged with hyperspectral images n ≥ 3 times, and the average spectrum of the healthy pixel set is calculated as the end member spectrum of the healthy tissue. Among them, i=1,2,…,n,E h represents the endmember spectrum of healthy tissue, n is the number of imaging times, S i represents the spectral data obtained by the i-th imaging; Endmember spectrum of diseased tissue E d The construction process of (t) is as follows: select k historical moments, k ≥ 3, of the target rice leaf area at different time points. [t-k:t-1] and environmental parameters P of temperature and humidity env , the Y [t-k:t-1] Contains the spectrum information from time tk to time t-1, and the environmental parameter P env With historical spectral data Y [t-k:t-1] As the input of the LSTM network, combined with the end member spectrum E of the diseased tissue at the previous moment d (t-1) and the bias term θ of the LSTM network, through E d (t) = E d (t-1)+LSTM(Y [t-k:t-1] ,P env ; θ) to predict the endmember spectrum E of the diseased tissue at the current time t d (t).

4. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 1, characterized in that: The functional relationship of the pixel spectral model in S300 is: t =f h (t)·E h +f d (t)·E d (t)+∈, where Y t is the pixel spectral vector at time point t, which contains the reflectance value of the pixel in each band, f h (t) and f d (t) are the abundance of healthy tissue and diseased tissue at time point t, respectively, and f h (t) = 1 - f d (t), E h is the constructed healthy tissue endmember spectrum vector, E d (t) is the endmember spectrum vector of the diseased tissue predicted based on the LSTM network, and ∈ is the residual vector, which represents the model error.

5. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 4, characterized in that: The constraints in S300 include physical constraints and timing constraints, where: Physical constraints include abundance non-negativity and abundance sum being 1, where the abundance non-negativity is: h (t)≥0,f d (t)≥0, the abundance sum is 1:f h (t)+f d (t) = 1, indicating that the pixel consists of only two end-members: healthy and diseased; Temporal constraints include lesion abundance smoothness and spectral evolution continuity, where lesion abundance smoothness is |f d (t)-f d (t-1)|≤δ·Δt, where δ is the maximum change rate threshold, ranging from 0.05% to 0.2%, and Δt is the time interval of acquisition. The continuity of the spectral evolution is Among them, γ is the spectral change rate threshold, and its value range is 0.001-0.

01.

6. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 5, characterized in that: The specific steps of S300 are: S301, initialization parameters: obtain the initial predicted value of the lesion endmember spectrum at the current moment from S200, and use the lesion abundance value at the previous moment as the initial value at the current moment; S302, updating the lesion abundance: constructing a linear equation system based on the current lesion endmember spectrum and the healthy endmember spectrum, solving the equation system to obtain a preliminary lesion abundance value, and applying the lesion abundance smoothness conditions of physical constraints and temporal constraints; S303, updating the lesion endmember spectrum: based on the current lesion abundance value, constructing a linear equation system, solving the equation system to obtain a preliminary lesion endmember spectrum, and applying a time sequence constraint spectrum evolution continuity condition; S304: Convergence determination: Calculate the abundance change between the current iteration and the previous iteration, calculate the spectrum change between the current iteration and the previous iteration, and perform threshold determination. S305 , outputting decomposition results: taking the lesion abundance obtained in the last iteration as the final result, and taking the lesion end-member spectrum obtained in the last iteration as the final result.

7. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 1, characterized in that: The S400 is based on the abundance of diseased tissue at each time point f d (t) and the actual endmember spectrum E′ of the diseased tissue d (t), generates a quantitative index of infection progress Here, α and β are weight coefficients and α=0.6 and β=0.

4.

8. The method for early diagnosis of rice blast based on hyperspectral imaging according to claim 7, characterized in that: The step S400 classifies the infection progress level according to the calculated infection progress quantitative index I(t): Low infection level: 0≤I(t)<0.3, indicating that rice blast infection is in the early stage; Medium infection level: 0.3≤I(t)<0.7, indicating that rice blast infection is in the development stage; High infection level: I(t) ≥ 0.7, indicating that rice blast infection is at a serious stage; The infection progress quantification index I(t) and the corresponding infection progress grade are output as the diagnosis results, providing a quantitative basis for rice blast prevention and control decisions.

Images