Seed detection method and device based on photoacoustic spectrometry technology

Abstract

The invention relates to the technical field of seed detection, in particular to a seed detection method and device based on a photoacoustic spectrometry technology, and the method comprises the following steps: determining a classification prediction model of soybean seed vigor and vitality, and obtaining soybean seed photoacoustic data; establishing a DBN-based seed vigor classification model; establishing a seed viability prediction model based on SVR; and inputting a seed vigor classification model and a viability prediction model according to the determined soybean seed photoacoustic data, and outputting a soybean seed vigor rating and viability prediction result. According to the method, the photoacoustic spectrum technology is introduced into the field of seed viability and viability detection, the photoacoustic data and the germination rate of the soybean seeds are analyzed, the photoacoustic spectrum technology is used for detecting photoacoustic information in the seeds, nondestructive detection and recognition of the viability and the viability of the soybean seeds are effectively achieved, and damage to the seeds is avoided; meanwhile, the seeds can be quickly and accurately screened, the labor cost for breeding excellent soybean varieties is reduced, and the soybean planting efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of seed detection, and in particular to a seed detection method and device based on photoacoustic spectroscopy technology. Background Art

[0002] As an important crop, soybean has huge demand in the fields of food, oil extraction and animal husbandry. Screening high-quality soybean seeds is of vital importance to improving seed quality and ensuring field productivity.

[0003] In the field of seed testing, traditional testing methods at home and abroad mainly include biochemical treatment and germination test evaluation methods. However, these traditional methods have obvious limitations. Most of them are destructive testing, which means that they will cause certain damage to the seeds during the testing process. In addition, the accuracy of the testing is relatively low, which is difficult to meet the needs of modern seed testing.

[0004] In recent years, with the continuous development of spectral technology, near-infrared spectroscopy, hyperspectral spectroscopy, Raman spectroscopy, terahertz spectroscopy and other spectral technologies have gradually been applied to non-destructive testing of seed quality. These spectral technologies have overcome the defects of traditional methods to a certain extent and realized non-destructive testing of seeds. However, in comparison, photoacoustic spectroscopy technology has more outstanding advantages. It not only has higher detection sensitivity, but also can detect substances in various forms, providing a more effective technical means for seed detection. Based on this, the present invention proposes a seed detection method and device based on photoacoustic spectroscopy technology, which aims to solve the problems existing in the existing seed detection methods and provide strong support for the screening and quality improvement of soybean seeds. Summary of the invention

[0005] The purpose of the present invention is to overcome the above problems or at least partially solve the above problems, and to propose a seed detection method and device based on photoacoustic spectroscopy technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a seed detection method based on photoacoustic spectroscopy technology, characterized in that it comprises the following steps:

[0007] S1. Determine the classification prediction model for soybean seed vigor and vitality, specifically including:

[0008] S1.1. Obtaining photoacoustic data of soybean seeds, including:

[0009] S1.11. Establish a photoacoustic spectroscopy detection system;

[0010] S1.12, perform full-band photoacoustic spectroscopy scanning of a single soybean at 7 frequencies;

[0011] S1.13. Determine the germination rate, germination index and vitality index of soybean seeds based on germination experiments;

[0012] S1.14, preprocessing the photoacoustic data based on ensemble empirical mode decomposition (EEMD) to obtain a denoised photoacoustic spectrum image;

[0013] S1.2. Establish a seed vitality classification model based on DBN, including:

[0014] S1.21. Extract characteristic wavelengths of photoacoustic data based on principal component analysis (PCA);

[0015] S1.22. A soybean vitality classification and recognition model was constructed based on a deep belief network (DBN), with the input being the characteristic wavelength and germination rate at the same modulation frequency;

[0016] S1.3. Establish a seed viability prediction model based on SVR, including:

[0017] S1.31, combined with SVR to obtain the optimal soybean germination rate prediction model, the input is the spectral variable after dimension reduction, and the output is the germination rate;

[0018] S1.32. Based on the photoacoustic data information after dimensionality reduction processing, the optimal parameters of SVR are dynamically selected using a genetic algorithm (GA) to obtain an optimized mixed soybean seed prediction model;

[0019] S2. Based on the soybean seed photoacoustic data determined in S1.1, the seed vigor classification model and vitality prediction model of S1.2 and S1.3 are input, and the soybean seed vitality rating and vitality prediction results are output.

[0020] In a preferred embodiment, the photoacoustic spectrum scanning of a single soybean at 7 frequencies in step S1.12 includes: performing photoacoustic spectrum scanning of 360 soybeans at 7 frequencies including 100 Hz, 200 Hz, 300 Hz, 400 Hz, 500 Hz, 800 Hz and 1000 Hz, a total of 2520 sets of data, and the collected spectral band is 400-4000 cm -1 .

[0021] In a preferred embodiment, the step S1.13 of determining the germination rate, germination index, and vitality index of soybean seeds based on the germination experiment includes: conducting a 7-day germination experiment on soybean seeds according to the ISTA standard, including soaking seeds, placing in beds, culturing in boxes, observing and recording, and settling results. The germination rate calculation formula is:

[0022]

[0023] Among them, f is the germination rate, n is the number of seeds with buds larger than 1 cm after seven days, and N is the number of seeds to be germinated;

[0024] The calculation formula of germination index is:

[0025]

[0026] Among them, Gt is the number of seeds germinated per day, and Dt is the corresponding number of germination days;

[0027] The calculation formula of vitality index is:

[0028] Vi=Gi×S

[0029] Where S is the root length of the seedlings on the 7th day after germination.

[0030] In a preferred embodiment, the preprocessing of the photoacoustic data based on ensemble empirical mode decomposition (EEMD) in step S1.14 includes: repeating a given number of experiments, averaging all decomposition results, obtaining denoised data, performing EEMD processing on the data, and adding white noise ω(t) having uniformly distributed statistical characteristics within the frequency range to the signal; the EEMD process includes: setting the standard deviation of the added Gaussian white noise to 0.1 and adding it to the original signal; performing EMD on the signal with added white noise to obtain an IMF component and a residual component; repeating the above process according to a given number of experiments, and calculating the overall average value of all experiments.

[0031] In a preferred embodiment, during the EEMD process, the formula for displaying all test results is:

[0032]

[0033] Where x(t) is the original photoacoustic signal, ω(t) is the white noise added in the first experiment, and IMF ij (t) is the first IMF component of the first trial, α(t) is the residual component of the first trial;

[0034] The formula for calculating the overall mean of all trials is:

[0035]

[0036] The formula for the final decomposition result of the original spectral data signal by EEMD is:

[0037]

[0038] In a preferred embodiment, the step S1.22 constructs a soybean vitality classification and recognition model based on a deep belief network (DBN), including: classifying and modeling the vitality of a single variety of soybeans to determine the classification model under the optimal modulation frequency, and classifying mixed soybeans based on the classification model under the optimal scanning frequency, wherein 70% of the total photoacoustic data is used as a training set and 30% is used as a test set in the classification model, and the optimal modulation frequency is 200Hz; the photoacoustic data is divided into non-vitality seeds, low-vitality seeds, and high-vitality seeds for labeling according to the vitality recognition model, and each vitality category corresponds to a specific vector as the DBN network output, and a corresponding category label is set; the relevant parameters of the DBN model are determined, including but not limited to the number of hidden layer nodes, the number of training set training times, the learning rate of each hidden layer, and the number of BPNN network fine-tuning times.

[0039] In a preferred embodiment, the step S1.31 in which the optimal soybean germination rate prediction model is obtained in combination with SVR includes: extracting the characteristic wavelength and germination rate at each modulation frequency using PCA features as input, randomly selecting a training set and a prediction set in the data matrix at a ratio of 3:1, and using the germination rate as the model output to obtain the optimal soybean germination rate prediction model and find the optimal modulation frequency; determining that the penalty coefficient c of the better soybean SVR vitality prediction model is 0.5 and the kernel function radius g is 10.

[0040] In a preferred embodiment, the step S1.32 of dynamically selecting the optimal parameters of SVR using a genetic algorithm (GA) includes: using a GA algorithm to dynamically select the optimal parameters of SVR to improve the accuracy of the load prediction model, determining that the maximum iteration number of the prediction model is 100 and the population size is 20.

[0041] The present invention also provides a seed detection device based on photoacoustic spectroscopy technology, which is applied to the above-mentioned seed detection method based on photoacoustic spectroscopy technology, and the device includes: an infrared light source, a monochromator, a plane mirror, a chopper, a driver, a quartz cover, a photoacoustic cell, a microphone, a phase-locked amplifier and a computer; wherein the light source is provided by an infrared light source, the monochromator is used to limit the wavelength of light waves, the plane mirror is used to reflect light, the chopper is used to limit the frequency of light waves, the driver is used to provide the frequency of transmitted light waves, the quartz cover is used to provide a closed space, the photoacoustic cell is used to perform photoacoustic processing on soybean seeds, the microphone obtains the photoacoustic data of soybean seeds and transmits it to the computer through the phase-locked amplifier, the computer controls the monochromator, controls the frequency of light waves that can pass through the chopper through the driver, and the information of the driver is synchronously transmitted to the phase-locked amplifier.

[0042] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the steps of the above-mentioned seed detection method based on photoacoustic spectroscopy technology.

[0043] Compared with the prior art, the present invention has the following beneficial effects: the photoacoustic spectroscopy technology is introduced into the field of seed vitality and vigor detection, and the photoacoustic spectroscopy technology is used to detect the internal photoacoustic information of the seeds by analyzing the photoacoustic data and germination rate of soybean seeds, thereby effectively realizing the non-destructive detection and identification of the vitality and vigor of soybean seeds, and avoiding the damage to seeds by traditional detection methods; at the same time, the method can quickly and accurately screen seeds, reduce the labor cost of breeding excellent soybean varieties, and improve the planting efficiency of soybeans;

[0044] In addition, the large amount of data accumulated during the detection process enriched the seed vitality characteristic spectrum database, providing research ideas for the application of optical detection technology in seed vitality detection; and the ensemble empirical mode decomposition (EEMD) was used to denoise the bean seed photoacoustic data, reducing the influence of factors such as changes in the sound environment, improving the accuracy of subsequent classification prediction models, and making the detection results more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a structural schematic diagram of a seed detection device based on photoacoustic spectroscopy technology of the present invention;

[0046] Figure 2 It is a flow chart of the seed detection method based on photoacoustic spectroscopy technology of the present invention;

[0047] Figure 3 This is a workflow diagram for dynamically selecting the optimal parameters of SVR using the GA algorithm in the present invention.

[0048] In the figure: 1. Infrared light source; 2. Monochromator; 3. Plane mirror; 4. Chopper; 5. Driver; 6. Quartz cover; 7. Photoacoustic cell; 8. Soybean seed; 9. Microphone; 10. Lock-in amplifier; 11. Computer. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1-3 The present invention provides a technical solution: a seed detection method based on photoacoustic spectroscopy technology, comprising the following steps:

[0051] S1. Determine the classification prediction model for soybean seed vigor and vitality, specifically including:

[0052] S1.1. Obtaining photoacoustic data of soybean seeds, including:

[0053] S1.11. Establish a photoacoustic spectroscopy detection system, which specifically includes an infrared light source 1, a monochromator 2, a plane mirror 3, a chopper 4, a driver 5, a quartz cover 6, a photoacoustic cell 7, a microphone 9, a phase-locked amplifier 10 and a computer 11; wherein the infrared light source 1 provides the light source, the monochromator 2 is used to limit the wavelength of the light wave, the plane mirror 3 is used to reflect the light, the chopper 4 is used to limit the frequency of the light wave, the driver 5 is used to provide the transmitted light wave frequency, the quartz cover 6 is used to provide a closed space, the photoacoustic cell 7 is used to perform photoacoustic treatment on the soybean seeds 8, the microphone 9 obtains the photoacoustic data of the soybean seeds 9 and transmits it to the computer 11 through the phase-locked amplifier 10, the computer 11 controls the monochromator 2, controls the frequency of the light wave that can pass through the chopper 4 through the driver 5, and the information of the driver 5 is synchronously transmitted to the phase-locked amplifier 10.

[0054] S1.12, perform full-band photoacoustic spectroscopy scanning of a single soybean at 7 frequencies;

[0055] In the specific implementation, 360 soybeans were scanned by photoacoustic spectroscopy at seven frequencies including 100 Hz, 200 Hz, 300 Hz, 400 Hz, 500 Hz, 800 Hz and 1000 Hz, with a total of 2520 sets of data, and the spectral band collected was 400-4000 cm -1 The specific operation steps are as follows: put beans of different types and aging times into the photoacoustic cell, and purge the photoacoustic cell with helium continuously for 10 seconds (10mL / min); set the scanning resolution to 8cm -1 , the mirror speed is 0.32cm / s, and the carbon black background is used as a control. Under the irradiation of infrared modulated light, the soybean seeds in the photoacoustic cell absorb light energy, the internal energy of the seeds increases to produce thermal expansion, and the heat is transferred to the helium in the sample cell (helium does not absorb light). The gas volume expands and contracts to produce pressure waves and acoustic signals, which are the measured soybean seed photoacoustic signals.

[0056] The preparatory operation for the experiment was as follows: prepare soybean seed samples. All seeds were stored in a 4°C refrigerator before the experiment. Before collecting soybean photoacoustic signals, the photoacoustic spectrometer needed to be preheated for 30 minutes, and the rear chamber of the instrument was purged with nitrogen through the intelligent humidity control system to reduce the interference of water vapor and other factors in the detection, ensuring the accuracy and validity of the experimental measurement data.

[0057] S1.13. Determine the germination rate, germination index and vitality index of soybean seeds based on the germination experiment. Specifically, a 7-day germination test of soybean seeds is conducted in accordance with the regulations of the International Seed Testing Association (ISTA). The test includes:

[0058] Soaking seeds; putting the bean seeds measured by photoacoustic spectroscopy into germination baskets according to their numbers and soaking them at 25°C for 24 hours;

[0059] Bed placement: Place the seeds for germination in a germination basket and cover it with moist germination paper. To ensure the consistency of the photoacoustic spectroscopy data and the germination test data, place the seeds of the sample on the moist germination paper in the order of their numbers, leaving a certain amount of growth space to reduce fungal infection;

[0060] Incubation in a box: Place the germination basket in an artificial climate incubator (model: PQX-250L), and keep the temperature at 25°C ± 2°C and the humidity at about 80% according to the most suitable temperature and humidity required for the growth of soybean seeds;

[0061] Observation records: During the germination experiment, the germination of soybean seeds under different natural aging conditions was recorded. The growth results of the seeds were recorded twice a day (8:00 and 19:00). According to the ISTA standard, seeds with a 1cm embryo length were considered to be in the germination state. The number of days was calculated based on the first 24 hours after soaking and placing in the bed as the first day;

[0062] Result calculation: Record the germination status of soybean seeds with different natural aging times.

[0063] The germination rate calculation formula is:

[0064]

[0065] Among them, f is the germination rate, n is the number of seeds with buds larger than 1 cm after seven days, and N is the number of seeds to be germinated;

[0066] The calculation formula of germination index is:

[0067]

[0068] Among them, Gt is the number of seeds germinated per day, and Dt is the corresponding number of germination days;

[0069] The calculation formula of vitality index is:

[0070] Vi=Gi×S

[0071] Where S is the root length of the seedlings on the 7th day after germination.

[0072] S1.14. Preprocess the photoacoustic data based on ensemble empirical mode decomposition (EEMD) to obtain denoised photoacoustic spectrum images. Specifically, by repeating the experiment a given number of times and averaging all decomposition results, the denoised data is obtained. The data is processed by EEMD, and white noise ω(t) with uniform distribution statistical characteristics within the frequency range is added to the signal, so that the signal assisted by white noise not only has a unified decomposition scale, but also can smooth abnormal events such as environmental sound and temperature interference.

[0073] The brief process of EEMD is as follows:

[0074] (1) Set the standard deviation of the added Gaussian white noise ω(t) to 0.1 and add it to the original signal;

[0075] (2) Perform EMD on the signal with white noise added to obtain an IMF component and a residual component;

[0076] (3) The given number of trials is 30, and the above steps are repeated. In each trial, the number of IMFs K is a constant. The following formula shows the results of all trials:

[0077]

[0078] Where x(t) is the original photoacoustic signal, ω(t) is the white noise added in the first experiment, and IMF ij (t) is the first IMF component of the first trial, and α(t) is the residual component of the first trial.

[0079] Calculate the overall mean of all trials as follows:

[0080]

[0081] The final decomposition result of EEMD on the original spectral data signal is as follows:

[0082]

[0083] S1.2. Establish a seed vitality classification model based on DBN, including:

[0084] S1.21. Extract characteristic wavelengths of photoacoustic data based on principal component analysis (PCA). Specifically, based on the original spectrum, the experimental characteristic wavelengths are obtained with the help of PCA algorithm. Taking Gutian soybean as an example, when the modulation frequency is 200 Hz, λ1, λ2, λ3, λ4, and λ5 are extracted as characteristic wavelengths of soybean.

[0085] Among them, the PCA (Principal Components Analysis) algorithm is a data dimensionality reduction technology that replaces a large number of original interrelated variables with a smaller number of unrelated principal components, while retaining as much related information as possible in the original variable data set. Taking the two-dimensional photoacoustic data of this experiment as an example, assume that one of the photoacoustic data matrices has n samples as column data, and each sample is represented by a P-dimensional variable, that is, x1, x2, ..., x p The original data set can be expressed as X = (X1, X2, ..., X p )=(x ij ) n×p, and PCA is to reselect the coordinate axis of the new coordinate origin for the part of the original two-dimensional data that you want to save, and project the discrete points onto the new coordinate system, that is, to reduce the original two-dimensional data to one dimension, and require that the new coordinates selected at this time minimize the loss of the original information. The original data matrix (x ij ) n×p The direction with the largest variance is taken as the principal component. In order to make the new coordinate contain a large amount of original data, the contribution rate and cumulative contribution rate of the principal component are used to express it. According to the cumulative contribution rate, m principal components are extracted. Generally, the 1st, 2nd, ..., m (m < p) principal components are taken, corresponding to the eigenvalues ​​λ1, λ2, ..., λ m , eigenvalue λ m The cumulative contribution rate of and the contribution rate of the i-th principal component are expressed as:

[0086]

[0087] S1.22. A soybean vitality classification and recognition model was constructed based on a deep belief network (DBN). The input was the characteristic wavelength and germination rate under the same modulation frequency. Specifically, a classification model was performed on the vitality of a single variety of soybeans to determine the classification model under the optimal modulation frequency, and the classification model under the optimal scanning frequency was used to classify mixed soybeans. In the classification model, 70% of the total photoacoustic data was used as a training set and 30% as a test set. The optimal modulation frequency was 200 Hz.

[0088] DBN (Deep confidence network) is a deep neural network obtained by randomly initializing samples and training them layer by layer. Its basic structure consists of several layers of restricted Boltzmann machines (RBM) and a layer of back propagation neural network (BPNN). DBN completes two tasks in the process of data training and classification: First, each layer of RBM is trained separately and unsupervised to map the features of the samples to be trained to different feature spaces; Second, after completing the multi-layer training of RBM, the output features are used as the input layer of the BPNN network to supervise the input and output of the training set. The specific operation steps include:

[0089] The photoacoustic data are divided into three categories for annotation according to the vitality recognition model: no vitality seeds (001), low vitality seeds (010), and high vitality seeds (100). These three vectors are set as DBN network outputs, and the corresponding labels are 1, 2, and 3;

[0090] Determine the optimal parameters of the DBN model, the number of hidden layer nodes is set to 20, the number of training sets is trained 300 times, the learning rate of each hidden layer is set to 0.05, and the number of BPNN network fine-tuning is set to 300;

[0091] The DBN classification model begins to stabilize when the number of training iterations reaches about 250. Furthermore, the accuracy, correlation coefficient Rc, and root mean square error SEP are used to evaluate the accuracy and performance of the model.

[0092] In this experiment, the test set accuracy is the average value obtained after running the model 10 times, with the first two decimal places retained. The test set accuracy of the PCA-DBN model can reach 95.58%.

[0093] S1.3. Establish a seed viability prediction model based on SVR, including:

[0094] S1.31. Combine SVR to obtain the optimal soybean germination rate prediction model. The input is the spectral variable after dimensionality reduction, and the output is the germination rate. Specifically, the characteristic wavelength and germination rate under each modulation frequency extracted by PCA features are used as input, and the training set and prediction set are randomly selected in the data matrix at a ratio of 3:1. The germination rate is used as the model output to obtain the optimal soybean germination rate prediction model and find the optimal modulation frequency.

[0095] The SVR (Support Vector Regression) model represents the training sample set as a vector in a spatial mapping so that samples from different categories are divided by as wide a clear gap as possible. Then new samples from the cross-validation or test set are mapped to the same space. Depending on which side of the gap between categories they belong to, they are predicted to belong to one category or another. When solving prediction problems, the regression prediction problem of high-dimensional features can be solved, especially when the feature dimension is greater than the number of samples. The three feature extraction methods used in this experiment reduced the 715-dimensional data to between 10 and 60 dimensions. SVR uses part of the support vectors to make hyperplane decisions to reduce the use of photoacoustic data. And it uses a large number of kernel functions to solve various nonlinear regression problems. Especially when the photoacoustic sample size is not a large amount of data, the prediction accuracy is high and the generalization ability is strong. Therefore, the penalty coefficient c of the better soybean SVR vitality prediction model is determined to be 0.5 and the kernel function radius g is 10.

[0096] S1.32. Based on the photoacoustic data information after dimensionality reduction, the genetic algorithm (GA) is used to dynamically select the optimal parameters of SVR to obtain the optimized mixed soybean seed prediction model. Specifically, GA (Genetic Algorithm) is a method of searching for the optimal solution by simulating the natural evolution process. The execution process of the GA algorithm is based on the fitness of the group search and optimization process, which plays a linking role between the problem and the algorithm in this process. Using the GA algorithm to dynamically select the optimal parameters of SVR can improve the accuracy of the load prediction model. The workflow is as follows: Figure 2. It was determined that the maximum iteration number of the prediction model was 100, the population size was 20, and the correlation coefficients of the mixed bean training set and the prediction set were improved.

[0097] S2. Based on the soybean seed photoacoustic data determined in S1.1, the seed vigor classification model and vitality prediction model of S1.2 and S1.3 are input, and the soybean seed vitality rating and vitality prediction results are output.

[0098] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to perform specific steps of a seed detection method based on photoacoustic spectroscopy technology.

[0099] The soybean seed germination rate in the present invention is a germination test for healthy soybean seeds. Photoacoustic spectroscopy is used to detect the quality of soybean seeds. Photoacoustic spectroscopy is a method that can detect the photoacoustic information inside the seeds. The use of photoacoustic spectroscopy to accurately detect the quality of crops or seeds is of great significance for agricultural production. It can effectively realize non-destructive detection and identification of the vitality and vigor of soybean seeds, reduce the labor cost of breeding excellent soybean varieties, improve the planting efficiency of soybeans, and enrich the seed vitality characteristic spectrum database, providing research ideas for the application of optical detection technology in seed vitality detection.

[0100] Therefore, the present invention introduces photoacoustic spectroscopy technology into the field of seed vitality and vigor detection. Based on the photoacoustic data and germination rate of soybean seeds, the ensemble empirical mode decomposition (EEMD) is used to denoise the photoacoustic data of soybean seeds, thereby reducing the impact of changes in the sound environment or noise and other information, and improving the accuracy of subsequent classification and prediction models, providing a faster and non-destructive detection technology for the field of soybean seed vitality and vigor detection.

[0101] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A seed detection method based on photoacoustic spectroscopy, characterized in that: The following steps are involved: S1. Determine the classification prediction model for soybean seed vigor and vitality, specifically including: S1.

1. Obtaining photoacoustic data of soybean seeds, including: S1.

11. Establish a photoacoustic spectroscopy detection system; S1.12, perform full-band photoacoustic spectroscopy scanning of a single soybean at 7 frequencies; S1.

13. Determine the germination rate, germination index and vitality index of soybean seeds based on germination experiments; S1.14, preprocessing the photoacoustic data based on ensemble empirical mode decomposition (EEMD) to obtain a denoised photoacoustic spectrum image; S1.

2. Establish a seed vitality classification model based on DBN, including: S1.

21. Extract characteristic wavelengths of photoacoustic data based on principal component analysis (PCA); S1.

22. A soybean vitality classification and recognition model was constructed based on a deep belief network (DBN), with the input being the characteristic wavelength and germination rate at the same modulation frequency; S1.

3. Establish a seed viability prediction model based on SVR, including: S1.31, combined with SVR to obtain the optimal soybean germination rate prediction model, the input is the spectral variable after dimension reduction, and the output is the germination rate; S1.

32. Based on the photoacoustic data information after dimensionality reduction processing, the optimal parameters of SVR are dynamically selected using a genetic algorithm (GA) to obtain an optimized mixed soybean seed prediction model; S2. Based on the soybean seed photoacoustic data determined in S1.1, the seed vigor classification model and vitality prediction model of S1.2 and S1.3 are input, and the soybean seed vitality rating and vitality prediction results are output.

2. The seed detection method based on photoacoustic spectroscopy according to claim 1, characterized in that: The step S1.12 includes scanning the single soybean with a full-band photoacoustic spectrum at seven frequencies: scanning 360 soybeans with a photoacoustic spectrum at seven frequencies, including 100 Hz, 200 Hz, 300 Hz, 400 Hz, 500 Hz, 800 Hz and 1000 Hz, for a total of 2520 sets of data, with a spectrum band of 400 to 4000 cm -1 .

3. The seed detection method based on photoacoustic spectroscopy according to claim 2, characterized in that: Determining the germination rate, germination index, and vitality index of soybean seeds based on the germination experiment in step S1.13 includes: conducting a germination experiment on soybean seeds for 7 days according to the ISTA standard, including soaking seeds, placing them in beds, culturing them in boxes, observing and recording, and settling results. The germination rate calculation formula is: Among them, f is the germination rate, n is the number of seeds with buds larger than 1 cm after seven days, and N is the number of seeds to be germinated; The calculation formula of germination index is: Among them, Gt is the number of seeds germinated per day, and Dt is the corresponding number of germination days; The calculation formula of vitality index is: Vi=Gi×S Where S is the root length of the seedlings on the 7th day after germination.

4. The seed detection method based on photoacoustic spectroscopy according to claim 3, characterized in that: The preprocessing of the photoacoustic data based on ensemble empirical mode decomposition (EEMD) in step S1.14 includes: repeating a given number of experiments, averaging all decomposition results, obtaining denoised data, subjecting the data to EEMD processing, and adding white noise ω(t) having uniformly distributed statistical characteristics within the frequency range to the signal; the EEMD process includes: setting the standard deviation of the added Gaussian white noise to 0.1 and adding it to the original signal; performing EMD on the signal with added white noise to obtain an IMF component and a residual component; and repeating the above process according to a given number of experiments to calculate the overall average value of all experiments.

5. The seed detection method based on photoacoustic spectroscopy according to claim 4, characterized in that: In the EEMD process, the formula showing all test results is: Where x(t) is the original photoacoustic signal, ω(t) is the white noise added in the first experiment, and IMF ij (t) is the first IMF component of the first trial, α(t) is the residual component of the first trial; The formula for calculating the overall mean of all trials is: The formula for the final decomposition result of the original spectral data signal by EEMD is:

6. The seed detection method based on photoacoustic spectroscopy according to claim 5, characterized in that: The step S1.22 constructs a soybean vitality classification and recognition model based on a deep belief network (DBN), including: classifying and modeling the vitality of a single variety of soybeans to determine the classification model under the optimal modulation frequency, and classifying mixed soybeans based on the classification model under the optimal scanning frequency, wherein 70% of the total photoacoustic data is used as a training set and 30% is used as a test set in the classification model, and the optimal modulation frequency is 200Hz; the photoacoustic data is divided into non-vitality seeds, low-vitality seeds, and high-vitality seeds for labeling according to the vitality recognition model, and each vitality category corresponds to a specific vector as the DBN network output, and a corresponding category label is set; and the relevant parameters of the DBN model are determined, including but not limited to the number of hidden layer nodes, the number of training set training times, the learning rate of each hidden layer, and the number of BPNN network fine-tuning times.

7. The seed detection method based on photoacoustic spectroscopy according to claim 6, characterized in that: The step S1.31 combines SVR to obtain the optimal soybean germination rate prediction model, including: extracting the characteristic wavelength and germination rate at each modulation frequency using PCA features as input, randomly selecting a training set and a prediction set in the data matrix at a ratio of 3:1, using the germination rate as the model output, obtaining the optimal soybean germination rate prediction model, and finding the optimal modulation frequency; determining the penalty coefficient c of the better soybean SVR vitality prediction model to be 0.5 and the kernel function radius g to be 10.

8. The seed detection method based on photoacoustic spectroscopy according to claim 7, characterized in that: The step S1.32 of dynamically selecting the optimal parameters of the SVR using a genetic algorithm (GA) includes: dynamically selecting the optimal parameters of the SVR using a GA algorithm to improve the accuracy of the load prediction model, and determining that the maximum iteration number of the prediction model is 100 and the population size is 20.

9. A seed detection device based on photoacoustic spectroscopy technology, characterized in that: The device is applied to the photoacoustic spectroscopy detection system in the seed detection method based on photoacoustic spectroscopy technology as described in any one of claims 1 to 8, and the device includes: An infrared light source (1), used to provide a light source; A monochromator (2) for limiting the wavelength of light; A plane mirror (3) for reflecting light; A chopper (4) for limiting the frequency of the light wave; A driver (5), used for providing a frequency of a transmitted light wave; A quartz cover (6) for providing a closed space; A photoacoustic cell (7) for performing photoacoustic treatment on soybean seeds (8); A microphone (9) for acquiring photoacoustic data of the soybean seeds (8); A lock-in amplifier (10) is used to receive the photoacoustic data acquired by the microphone (9) and transmit it to a computer (11); The computer (11) is used to control the monochromator (2) and the frequency of the light waves that can pass through the chopper (4) through the driver (5), and the information of the driver (5) is synchronously transmitted to the phase-locked amplifier (10).

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is executed by a processor to implement the steps of the seed detection method based on photoacoustic spectroscopy technology as described in any one of claims 1-8.

Images