Seed vigor detection device and method based on polarization spectrum technology

Abstract

The invention discloses a seed vigor detection device based on a polarization spectrum technology, which comprises a large-stroke electric control mobile platform, a black paperboard, a seed sample, a halogen light source, a polarizer, a lens, a hyperspectral sorter, an analyzer and a computer, and all the parts are arranged in a camera obscura; the light emitted by the halogen light source is reflected by the polarizer and the seed sample, reaches the lens and is captured by the hyperspectral sorter; the invention also provides a detection method which comprises the following steps: firstly, collecting spectral data; carrying out image analysis; preprocessing the polarization spectrum data, performing competitive adaptive reweighting algorithm dimension reduction, and selecting a characteristic wavelength; finally, establishing a seed vigor detection model, fusing ideal modeling methods by adopting Blending ensemble learning, and carrying out hybrid modeling on the grain seeds; the polarization hyperspectral technology is introduced into the field of seed vigor detection, the seed vigor detection precision is improved, and rapid and nondestructive detection of the soybean seed vigor is achieved.

Description

Technical Field

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

[0002] Seed vigor is an important indicator that takes into account many characteristics, including seed germination rate, field emergence rate, and lodging resistance of germinated plants. Its level is directly related to field emergence. It is usually manifested as high-quality seeds with high vigor having a high germination rate and being able to grow seedlings that are more resistant to lodging, thereby increasing crop yields. On the contrary, if the seed vigor is low, it will lead to a low germination rate. If the seeds are inactive, they may not even germinate, which will directly affect the yield and harvest, and significantly increase farmers' production costs. Therefore, accurate detection of seed vigor before sowing can avoid problems such as non-germination and poor stress resistance caused by low seed vigor, ensure the high vitality of the seeds to be sown, thereby increasing seed yields and harvests, and better helping agricultural producers.

[0003] Traditional seed vigor detection methods include tetrazolium chloride staining, cold soaking test, pH, potassium content, and electrical conductivity (EC) determination. These methods mainly use the biochemical characteristics of seeds, such as membrane permeability and changes in the content of substances inside seeds. Although they can achieve a certain degree of detection accuracy, they have problems such as long detection cycle, high professional requirements, large subjective judgment influence, and great damage to seeds. Therefore, scholars have studied and discussed how to conduct rapid and non-destructive detection of seed vigor, and the emergence and rapid development of spectral technology can just solve this problem. In order to solve the problems existing in traditional seed vigor detection methods and realize rapid and non-destructive detection of seed vigor, machine vision, near-infrared spectroscopy, polarization spectroscopy technology, hyperspectral technology, polarization hyperspectral technology and other methods have been applied to seed vigor detection research. These methods mainly detect the physical characteristics of seeds outside or inside, such as seed shape, size, shell color, embryo shape, and internal cracks / holes.

[0004] Compared with traditional detection methods, machine vision technology can better predict and evaluate the vitality of crop seeds, ensuring high prediction accuracy while achieving non-destructive testing. However, contamination on the seed surface and uneven lighting in the experiment will affect the test results. At the same time, when using machine vision technology to detect seed vitality, it is mainly based on the external physical characteristics of the seeds, and it is impossible to obtain information about the internal substances of the seeds. Unlike machine vision, near-infrared spectroscopy technology can detect the composition and content of internal substances in seeds, and the seed information obtained is more complete, so qualitative and quantitative analysis of seed vitality can be carried out at the same time, but this method also has the limitation of being unable to obtain spatial information of seeds. When using near-infrared spectroscopy technology to detect seed vitality, generally only the spectral information of seeds can be collected, while when using hyperspectral imaging technology to detect seed vitality, more abundant characteristic information of seeds can be obtained at one time, not only two-dimensional spectral information can be obtained, but also three-dimensional cubic diagrams can be obtained.

[0005] Therefore, compared with near-infrared spectroscopy, hyperspectral imaging technology has greater advantages. Therefore, the present application proposes a seed vitality detection device and method based on polarization spectrum technology. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention discloses a seed vitality detection device and method based on polarization spectrum technology to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions: a seed vitality detection device based on polarization spectrum technology, comprising a long-stroke electrically controlled mobile platform, a black cardboard, a seed sample, a halogen light source, a polarizer, a lens, a hyperspectral sorter, an analyzer, a first bracket, a second bracket and a computer, and each component is installed inside a dark box to avoid external interference and ensure its concealment;

[0008] The black cardboard is fixed on the long-stroke electric-controlled mobile platform, the seed samples are placed on the black cardboard in a certain arrangement order, the long-stroke electric-controlled mobile platform is used to adjust the travel distance that the seed samples need to move, and then the long-stroke electric-controlled mobile platform is used to move the samples to complete line scanning;

[0009] The hyperspectral sorter is located at the upper end of the long-stroke electrically-controlled mobile platform, and is fixed by a first bracket, the halogen light source is located at one side of the long-stroke electrically-controlled mobile platform, and is fixed by a second bracket, and the halogen light source is connected to a computer;

[0010] A polarizer is arranged above the seed sample, the lower end of the hyperspectral sorter is connected to a lens, and a polarizer is arranged at the lower end of the halogen light source, so that the light emitted by the halogen light source is ensured to reach the lens and be captured by the hyperspectral sorter after being reflected by the polarizer and the seed sample through the polarizer;

[0011] A light shield is arranged outside the halogen light source, and the light shield covers the outside of the halogen light source. The aluminum light shield can be removed during operation, and the aluminum light shield can be used to shield the halogen light source during non-operation.

[0012] The present invention also provides a detection method of a seed vitality detection device based on polarization spectrum technology, comprising the following steps:

[0013] S1, first collect polarization spectrum data of seed samples;

[0014] S2, then performing image analysis on the collected polarization spectrum data;

[0015] S3, then analyzing the spectral data, including preprocessing the polarization spectral data, and performing dimension reduction on the spectral data using a competitive adaptive reweighting algorithm to select characteristic wavelengths;

[0016] S4. Finally, a seed vitality detection model is established. At the same time, blending ensemble learning is used on the basis of a single model to integrate the ideal modeling methods, apply the best modeling method, and perform mixed modeling on each variety of grain seeds.

[0017] Preferably, step S1 specifically includes the following steps:

[0018] First, the seed samples are artificially aged; a germination experiment is carried out on the seeds; then the seed vitality detection device is adjusted, the halogen light source is turned on for preheating, the polarization angle of the polarizer is adjusted, the lens is focused, and the forward speed of the large-stroke electrically controlled mobile platform is set. Without adding a polarizer, ordinary hyperspectral images of the seed samples are collected using a hyperspectral sorter, and then an analyzer is installed in front of the lens, and a polarizer is installed in front of the halogen light source. Polarized hyperspectral images of each sample are collected when the polarizer installed in front of the lens is rotated at different angles.

[0019] Preferably, step S2 specifically includes the following steps:

[0020] The collected hyperspectral and polarized hyperspectral images are subjected to black and white correction, and then the most important target part in the acquired seed image is automatically extracted by ROI.

[0021] Preferably, step S3 specifically includes the following steps:

[0022] A moving average filter is used to smooth the vector data of each sample hyperspectral and polarization hyperspectral data collected. Then, four preprocessing methods, including first-order derivative, least squares smoothing filter, multivariate scattering correction and normal variable correction, are used to preprocess the seed hyperspectral and polarization hyperspectral data. The polarization component and polarization state are calculated by calculating the original polarization hyperspectral data of the seed sample, and then the spectral data is subjected to CARS dimensionality reduction.

[0023] Preferably, step S4 specifically includes the following steps:

[0024] Five machine learning algorithms including SVR, BPNN, GRNN, PLSR and RF were introduced to model and analyze the collected hyperspectral and polarization hyperspectral data of seed samples. Several detection models were used to perform single variety vitality detection modeling on the collected and processed seed dynamic polarization spectral data. The detection models included PLSR, BPNN, GRNN, SVR, RF, CNN, and LSTM. At the same time, the best modeling method was applied to perform hybrid modeling for each variety of grain seeds. Blending ensemble learning was used on the basis of a single model to integrate the ideal modeling methods. The best modeling method was applied to perform hybrid modeling for each variety of grain seeds.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. In the present invention, seed vitality can be detected based on static and dynamic polarization spectroscopy technologies, and the method is universal. At the same time, the dynamic polarization spectroscopy technology can effectively shorten the detection time and effectively solve the problems existing in traditional seed vitality detection methods, such as complex operation, long detection cycle and high professional requirements.

[0027] 2. In the present invention, polarization hyperspectral technology is introduced into the field of seed vigor detection for the first time, which improves the accuracy of traditional hyperspectral technology for seed vigor detection and realizes rapid and non-destructive detection of soybean seed vitality. It has certain research prospects and provides a new method for large-scale, non-contact, online seed vigor sorting lines. It not only provides a new method for the subsequent realization of online seed quality rapid sorting lines, but also provides a new method for seed vigor detection, which has very broad application prospects in the agricultural field.

[0028] 3. In the present invention, blending ensemble learning is adopted to improve the prediction accuracy of seed vitality model by fusing the ideal model of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0030] In the attached picture:

[0031] Figure 1 Schematic diagram of a seed vitality detection device based on polarization spectrum technology of the present invention;

[0032] Figure 2 This is a flow chart of the seed vitality detection method based on polarization spectrum technology of the present invention;

[0033] Numbers in the figure: 1. Long-stroke electronically controlled mobile platform; 2. Black cardboard; 3. Seed sample; 4. Halogen light source; 5. Light hood; 6. Polarizer; 7. Lens; 8. Hyperspectral sorter; 9. Dark box; 10. Analyzer; 11. First bracket; 12. Second bracket; 13. Computer. DETAILED DESCRIPTION

[0034] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0035] like Figure 1 As shown, a seed vitality detection device based on polarization spectrum technology includes a long-stroke electric-controlled mobile platform 1, a black cardboard 2, a seed sample 3, a halogen light source 4, a polarizer 6, a lens 7, a hyperspectral sorter 8, an analyzer 10, a first bracket 11, a second bracket 12 and a computer 13, and each component is installed in a dark box 9 to avoid external interference and ensure its concealment;

[0036] The black cardboard 2 is fixed on the long-stroke electric-controlled mobile platform 1, and the seed samples 3 are placed on the black cardboard 2 in a certain arrangement order. The long-stroke electric-controlled mobile platform 1 is used to adjust the travel distance that the seed samples 3 need to move, and then the long-stroke electric-controlled mobile platform 1 is used to move the samples to complete the line scan;

[0037] The hyperspectral sorter 8 is located at the upper end of the long-stroke electrically-controlled mobile platform 1, and is fixed by a first bracket 11. The halogen light source 4 is located on one side of the long-stroke electrically-controlled mobile platform 1, and is fixed by a second bracket 12, and the halogen light source 4 is connected to a computer 13. A light shield 5 is provided on the outside of the halogen light source 4, and the light shield 5 covers the outside of the halogen light source 4. The aluminum profile light shield 5 can be removed during operation, and the aluminum profile light shield 5 can be used to block the halogen light source 4 when not in operation. The halogen light source 4 can adopt a halogen tungsten light source.

[0038] A polarizer 10 is provided above the seed sample 3, the lower end of the hyperspectral sorter 8 is connected to the lens 7, and a polarizer 6 is provided at the lower end of the halogen light source 4. The polarizer 6 ensures that the light emitted by the halogen light source 4 reaches the lens 7 after being reflected by the polarizer 6 and the seed sample 3 and is captured by the hyperspectral sorter 8.

[0039] like Figure 2 As shown, the present invention also provides a detection method of a seed vitality detection device based on polarization spectrum technology, including the following situations:

[0040] Example 1: Cereal seed vitality detection based on static polarization spectroscopy;

[0041] The main steps of the seed vitality detection method based on static polarization spectroscopy are as follows:

[0042] Step a. Performing artificial aging treatment on the seed sample 3.

[0043] Step b. After collecting the polarization spectrum of the original seeds, a germination experiment is performed on the seeds in accordance with the national standard GT / T 3543.1-3543.7-1995.

[0044] Step c. Collecting spectral data. Turn on the halogen light source 4 and preheat it. Adjust the initial polarization angle of the polarizer 6 to 0°, that is, the polarization angles of the polarizer and the analyzer are both rotated to 0°. Then rotate the analyzer 10 from 0° to 90°, with an interval of 5°, to obtain a total of 19 seed polarization spectra at polarization angles.

[0045] Step d. Spectral data preprocessing: The USB4000 spectrometer used collects spectrum in the range of 350-1000 nm. Considering the noise influence of the halogen light source 4 on spectrum collection, 550-950 nm is selected as the effective band. After removing the singular value, the wavelet denoising method is used to preprocess the polarization spectrum of the effective band.

[0046] Step e. Characteristic polarization angle selection. Select the polarization mean spectral data of the seed sample solution without aging (0d) and the seed sample solution with the longest aging time (6d), calculate the difference contribution value (DCV) of 19 groups of polarization angles, and select the best detection characteristic polarization angle:

[0047]

[0048] In the above formula, TD0,λ i With TD6,λ i They represent the polarization spectral transmittance of the solution of seeds soaked in unaged and aged seeds for 6 days at wavelength λ, respectively, and the wavelength range is 550nm~950nm in the pretreatment band. The formula is used to find the characteristic polarization angle when the difference cumulative value is the largest.

[0049] Step f. Using the principal component analysis method, the principal components and corresponding contribution rates of the polarization spectrum data matrix collected when the seeds are soaked for 10 minutes are obtained respectively, and the principal components are selected with a target cumulative contribution rate of 99% to complete the characteristic band extraction.

[0050] Step g. Predict seed vigor using polarization spectrum transmittance data obtained when seeds are soaked for 10 minutes. After preprocessing and feature extraction of polarization spectra of seeds with different aging degrees, regression prediction of seed vigor is performed based on four common machine learning methods: partial least squares regression, back propagation neural network, generalized regression neural network, and support vector regression.

[0051] Example 2: Cereal seed vitality detection based on dynamic polarization spectroscopy;

[0052] The main steps of the grain seed vitality detection method based on dynamic polarization spectroscopy are as follows:

[0053] Step a. Collect spectral data. Using the polarization detection device in the present application, after installing the optical fiber and the instrument, turn on the halogen light source 4 for preheating to minimize experimental errors. 61 sets of spectral data at different immersion times are collected for each sample. In order to explore whether the differences in seed membranes with different vitality gradients cause different substance leaching rates, the polarization spectral data are differentially processed to obtain dynamic polarization spectral data of the seeds.

[0054] Step b. Spectral data preprocessing. Considering the influence of noise on dynamic polarization spectral data, the 550-950nm band is used as the effective spectrum. First, the dynamic polarization spectral data is preprocessed by wavelet denoising, and the processed differential data will be used as the input of the seed vitality detection model. In order to avoid the contingency caused by random selection of samples in subsequent modeling, all seed samples of each variety are normalized and averaged.

[0055] Step c. Establish a vitality detection model. Three seed vitality modeling methods, namely random forest (RF), convolutional neural network (CNN) and long short-term memory artificial neural network (LSTM), were introduced. A total of 7 models (PLSR, BPNN, GRNN, SVR, RF, CNN, and LSTM) were used to perform single strain vitality detection modeling on the collected and processed seed dynamic polarization spectral data. At the same time, the best modeling method was applied to perform mixed modeling on each variety of grain seeds. In order to improve the accuracy of mixed modeling, blending ensemble learning was used on the basis of a single model to integrate the ideal modeling methods and improve the prediction accuracy of seed mixed modeling.

[0056] Example 3: Soybean seed vitality detection based on polarization hyperspectral.

[0057] The main steps of the soybean seed vitality detection method based on polarization hyperspectral are as follows:

[0058] Step a. Conducting a germination experiment on the cereal seeds according to the national standard GT / T 3543.1-3543.7-1995.

[0059] Step b. Collect spectral data. Open the SPECVIEW software to focus the lens 7, set the scanning distance to 25 cm, and the forward speed of the large-stroke electric-controlled mobile platform 1 (conveyor belt) to 0.36 cm / s. After all parameters are set, collect the spectrum of the seeds. Do not add a polarizer first, and use the hyperspectral sorter 8 to collect ordinary hyperspectral images of the seed sample 3. Then install a polarizer 10 in front of the lens 7, and install a polarizer 6 in front of the halogen light source 4, and collect polarized hyperspectral images of each sample when the polarizer in front of the rotating lens 7 is 0°, 45°, 90° and 135°.

[0060] Step c. Perform image analysis. First, perform black and white correction on the acquired hyperspectral and polarized hyperspectral images, then extract the most important target part of the acquired seed image and perform processing and analysis, which reduces processing time and improves the accuracy of seed vitality detection.

[0061] Step d. Spectral data analysis. To improve the detection speed and accuracy, the moving average filter is first used to smooth the vector data of each sample hyperspectral and polarization hyperspectral data collected, and then the seed hyperspectral and polarization hyperspectral (0°, 45°, 90° and 135°) data are preprocessed using four preprocessing methods: first-order derivative, least squares smoothing filter, multivariate scattering correction and normal variable correction. The original polarization hyperspectral data of soybean seeds are calculated to obtain the three polarization components and polarization states of I, Q and U, and then the spectral data is subjected to CARS dimensionality reduction.

[0062] Step e. Establish a soybean seed vitality detection model. In order to find the optimal method among the four preprocessing methods of FD, SG, MSC and SNV in the soybean seed vitality detection research, five machine learning algorithms including SVR, BPNN, GRNN, PLSR and RF were used to model and analyze the collected soybean seed hyperspectral and polarization hyperspectral data.

[0063] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A seed vitality detection device based on polarization spectrum technology, characterized in that: It includes a long-stroke electric-controlled mobile platform, a black cardboard, a seed sample, a halogen light source, a polarizer, a lens, a hyperspectral sorter, an analyzer, a first bracket, a second bracket and a computer, and all the components are installed inside a dark box; The black cardboard is fixed on the long-stroke electric-controlled mobile platform, the seed samples are placed on the black cardboard in a certain arrangement order, the long-stroke electric-controlled mobile platform is used to adjust the travel distance that the seed samples need to move, and then the long-stroke electric-controlled mobile platform is used to move the samples to complete line scanning; The hyperspectral sorter is located at the upper end of the long-stroke electrically-controlled mobile platform, and is fixed by a first bracket, the halogen light source is located at one side of the long-stroke electrically-controlled mobile platform, and is fixed by a second bracket, and the halogen light source is connected to a computer; A polarizer is arranged above the seed sample, the lower end of the hyperspectral sorter is connected to a lens, and a polarizer is arranged at the lower end of the halogen light source, so that the light emitted by the halogen light source is ensured to reach the lens and be captured by the hyperspectral sorter after being reflected by the polarizer and the seed sample through the polarizer; A light shield is arranged outside the halogen light source, and the light shield covers the outside of the halogen light source.

2. The detection method of the seed vitality detection device based on polarization spectrum technology according to claim 1, characterized in that: The following steps are involved: S1, first collect polarization spectrum data of seed samples; S2, then performing image analysis on the collected polarization spectrum data; S3, then analyzing the spectral data, including preprocessing the polarization spectral data, and performing dimension reduction on the spectral data using a competitive adaptive reweighting algorithm to select characteristic wavelengths; S4. Finally, a seed vitality detection model is established. At the same time, blending ensemble learning is used on the basis of a single model to integrate the ideal modeling methods, apply the best modeling method, and perform mixed modeling on each variety of grain seeds.

3. The detection method of a seed vitality detection device based on polarization spectrum technology according to claim 2 is characterized in that: Step S1 specifically includes the following steps: First, the seed samples are artificially aged; a germination experiment is carried out on the seeds; then the seed vitality detection device is adjusted, the halogen light source is turned on for preheating, the polarization angle of the polarizer is adjusted, the lens is focused, and the forward speed of the large-stroke electrically controlled mobile platform is set. Without adding a polarizer, ordinary hyperspectral images of the seed samples are collected using a hyperspectral sorter, and then an analyzer is installed in front of the lens, and a polarizer is installed in front of the halogen light source. Polarized hyperspectral images of each sample are collected when the polarizer installed in front of the lens is rotated at different angles.

4. The detection method of a seed vitality detection device based on polarization spectrum technology according to claim 2 is characterized in that: In step S2, the following steps are specifically included: The collected hyperspectral and polarized hyperspectral images are subjected to black and white correction, and then the most important target part in the acquired seed image is automatically extracted by ROI.

5. The detection method of a seed vitality detection device based on polarization spectrum technology according to claim 2 is characterized in that: Step S3 specifically includes the following steps: A moving average filter is used to smooth the vector data of each sample hyperspectral and polarization hyperspectral data collected. Then, four preprocessing methods, including first-order derivative, least squares smoothing filter, multivariate scattering correction and normal variable correction, are used to preprocess the seed hyperspectral and polarization hyperspectral data. The polarization component and polarization state are calculated by calculating the original polarization hyperspectral data of the seed sample, and then the spectral data is subjected to CARS dimensionality reduction.

6. The detection method of a seed vitality detection device based on polarization spectrum technology according to claim 2 is characterized in that: Step S4 specifically includes the following steps: Five machine learning algorithms including SVR, BPNN, GRNN, PLSR and RF were introduced to model and analyze the collected hyperspectral and polarization hyperspectral data of seed samples. Several detection models were used to perform single variety vitality detection modeling on the collected and processed seed dynamic polarization spectral data. The detection models included PLSR, BPNN, GRNN, SVR, RF, CNN, and LSTM. At the same time, the best modeling method was applied to perform hybrid modeling for each variety of grain seeds. Blending ensemble learning was used on the basis of a single model to integrate the ideal modeling methods. The best modeling method was applied to perform hybrid modeling for each variety of grain seeds.

Images