Method for Rapid and Non-Destructive Monitoring of Early Parasitism of Parasitic Weeds on Sunflowers Underground

Abstract

The present invention belongs to the technical field of crop weed control. In particular, the present invention provides a method for rapidly and non-destructively monitoring the early underground parasitism of the parasitic weed Orobanche cumana in sunflowers. For the first time, the present invention uses hyperspectral technology to rapidly and non-destructively monitor the method of Orobanche cumana infecting sunflowers at four early stages in different resistant sunflowers. Based on the ELM model of the physiological index sensitive band, infected plants and non-infected plants as well as plants of different varieties 10, 17, 24, and 31 days after sunflower planting are classified, with the accuracy rates being 95.83% and 95.83% respectively. Combining the spatial and spectral characteristics of the hyperspectral image, the 3-band image dataset obtains accuracy rates of 95.83% and 97.92% through the CNN model, indicating the potential of the multispectral imaging system in the detection task. Through these models, we can easily achieve early warning of the parasitic stage of Orobanche cumana.

Description

Technical Field

[0001] The present invention belongs to the technical field of crop weeds, and particularly relates to a method for rapidly and non-destructively monitoring the early underground parasitism of sunflower parasitic weeds. Background Art

[0002] Sunflower broomrape (Orobanche cumana Wallr) is an annual root parasitic weed that mainly parasitizes on the roots of sunflowers. By establishing a parasitic relationship with the sunflower roots, it absorbs the nutrients and water of the sunflowers to achieve complete parasitism. This disease is mainly distributed in arid and semi-arid regions around the world. Among them, the Mediterranean region, Africa, southeastern Europe and Asia are the most severely affected by broomrape. In China, it is mainly represented by Bayannur City, Inner Mongolia, and has been listed as one of the quarantine plants for entry into China.

[0003] Sunflower is an important economic crop in the world and also one of the five major oil crops in China. Once sunflower broomrape parasitizes on the roots of sunflowers, it will cause a decrease in the biomass and yield of sunflowers. In severe cases, the yield can be reduced by 80% or even result in a complete crop failure, and the oil content of sunflowers will also be severely affected. Worldwide, the control of broomrape has always been a major problem.

[0004] Since broomrape is a root parasitic weed, it has already had a serious impact on the growth of host plants during the underground growth stage of broomrape. Therefore, the critical period for controlling broomrape is the underground growth stage, that is, the early infection stage of sunflower broomrape. At present, the measures for controlling broomrape mainly include biological control, spraying herbicides, manual removal, using trap crops to induce eradication, chemical control, and breeding of resistant species, etc. However, each method has certain limitations to some extent. Therefore, there is an urgent need for a rapid and non-destructive method to diagnose this parasitic state and then take appropriate measures for timely control. This is to use diagnostic tools as early as possible after sunflower sowing and sunflower broomrape infection, which has a wide range of applications in finding effective control methods, thereby improving the control efficiency of sunflower broomrape. Summary of the Invention

[0005] The present invention provides a method for rapidly and non-destructively monitoring the early underground parasitism of sunflower parasitic weeds, providing an effective measure for rapidly diagnosing this parasitic state and then taking appropriate measures for timely control, which will greatly reduce the use of herbicides and protect the ecological environment.

[0006] The present invention for the first time uses hyperspectral technology to rapidly and non-destructively monitor the method of different resistant sunflowers being infected by parasitic weeds sunflower broomrape at four early stages.

[0007] In this invention, through pot experiments, three sunflowers with different resistances and Orobanche cumana Wallr. were co-cultured to simulate the action process of parasitic weeds in the field. It was found that with the parasitism of Orobanche cumana Wallr. on the highly susceptible sunflower SH361, the differences in plant growth between the control and the treatment became increasingly significant. Secondly, it was the moderately resistant variety Sanrui 3, while there were no significant differences in plant height, dry and fresh weights between the control and the treatment of the immune variety Tonghui 15 at all stages of parasitism.

[0008] Furthermore, a quantitative regression model (including visualization) of enzyme activity physiological indexes was established based on hyperspectral technology.

[0009] Furthermore, physiological indexes of sunflower leaves after hyperspectral images were completed were measured subsequently.

[0010] The physiological indexes include SOD antioxidant enzyme activity, glutathione reductase GR, glutathione content (GSSG + GSH, GSH), phenylalanine ammonia-lyase, polyphenol oxidase, and malondialdehyde content and ROS.

[0011] Furthermore, a classification model for the presence or absence of parasitism and resistant varieties was obtained by hyperspectral technology.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] The present invention provides a method for rapid and non-destructive monitoring of the early parasitism of the parasitic weed Orobanche cumana Wallr. on sunflowers, and further discovers the identification of sunflower resistant varieties based on hyperspectral technology, the identification of Orobanche cumana Wallr. parasitism, and the rapid quantitative and visual characterization of physiological indexes. By establishing a model, it is expected to achieve rapid and early non-destructive monitoring of the parasitism of Orobanche cumana Wallr. in field production, providing broad and powerful technical support for the implementation of subsequent specific and effective prevention and control measures. Description of the Drawings

[0014] The drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0015] Figure 1 Shows the changes in plant height and dry and fresh weights of three sunflower varieties with different resistances at different stages of Orobanche cumana Wallr. parasitism. The data are the averages of three replicates (mean ± SE). * indicates significant differences at p-value < 0.05 in the Turkey multiple comparison test, ** p-value < 0.01, *** p-value < 0.001, **** p-value < 0.0001.

[0016] Figure 2Changes in the activities of SOD, GR antioxidant enzymes, non-antioxidant enzymes, phenylalanine ammonia-lyase, and polyphenol oxidase in three sunflower varieties with different resistances at different stages of Orobanche cumana parasitism. Data are the means of three replicates (mean ± SE). * indicates significant differences at p-value < 0.05 by Turkey's multiple comparison test, ** p-value < 0.01, *** p-value < 0.001, **** p-value < 0.0001.

[0017] Figure 3 Changes in malondialdehyde, H2O2, and O2 - in three sunflower varieties with different resistances at different stages of Orobanche cumana parasitism. Data are the means of three replicates (mean ± SE). * indicates significant differences at p-value < 0.05 by Turkey's multiple comparison test, ** p-value < 0.01, *** p-value < 0.001, **** p-value < 0.0001.

[0018] Figure 4 Are enzyme activity distribution maps of RGB images of sunflower canopies and GR, PPO, GSH, GSH+GSSG. The samples are (a) samples of the SH-CK group, (b) samples of the SH-OR group, (c) samples of the SR-CK group, (d) samples of the SR-OR group, (e) samples of the TH-CK group, (f) samples of the TH-OR group after 10 days of cultivation, and (g) samples of the SH-CK group, (h) samples of the SH-OR group, (i) samples of the SR-CK group, (j) samples of the SR-OR group, (k) samples of the TH-CK group, (l) samples of the TH-OR group after 31 days of cultivation.

[0019] Figure 5 (a) Visible-near infrared region, uninfected (black) and infected (gray) samples, (b) visible-near infrared region, uninfected and infected samples, (c) SH361 (black), SR3 (dark gray), and TH15 (light gray) samples in the visible-near infrared region, (d) SH361, SR3, TH15 samples in the visible-near infrared region. The shadows illustrate the standard errors of the reflectance.

[0020] Figure 6 Are the confusion matrices of the variety recognition test sets of (a) the ELM model based on fused data, (b) the ELM model based on CARS-extracted features of fused data, (c) the ELM model based on enzyme activity-sensitive features, and (d) the ELM model based on re-extracted features from joint features by SPA.

[0021] Figure 7Approximate canopy areas of samples at different stages. At this stage, dark gray indicates that the sample is infected, and light gray indicates that the sample is not infected. A, B, C, and D respectively represent the approximate canopy areas of sunflower leaves after co-culturing sunflowers with Orobanche cumana for 10 days, 17 days, 24 days, and 31 days. Detailed implementation manners

[0022] The present invention will be further described below in conjunction with specific embodiments. The following are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto.

[0023] Materials used in the present invention: Seeds of Orobanche cumana (the currently highest physiological race G in China) and sunflowers (SH361, Sanrui No. 3, Tonghui 15) were provided by the Plant Protection Institute of Inner Mongolia Academy of Agricultural and Animal Husbandry Sciences. The sunflower variety SH361 is 100% susceptible to Orobanche cumana, Sanrui No. 3 is highly resistant to Orobanche cumana but will parasitize, and Tonghui 15 is 100% immune to Orobanche cumana.

[0024] Example 1

[0025] 2 Materials and methods

[0026] 2.1 Pretreatment and germination of sunflower seeds

[0027] Taking highly susceptible, highly resistant, and immune sunflower varieties as experimental materials, select plump and healthy sunflower seeds, remove the outer seed coat, disinfect them in 75% anhydrous ethanol solution for 1 min, rinse several times with distilled water, and then put them into a culture dish containing two layers of filter paper with distilled water. Cultivate them under dark conditions at 25 °C for 48 h to promote germination and root growth. Co-culture them with Orobanche cumana respectively, and use the sunflowers of each variety that have not been co-cultured as a control. At the early stage of Orobanche cumana parasitism (co-cultured for 10 days), just parasitized (co-cultured for 17 days), one week after parasitism (co-cultured for 24 days), and two weeks after parasitism (co-cultured for 31 days), select plants with consistent growth to measure various indicators, and each group has at least six replicates.

[0028] 2.2 Establishment of co-culture system of sunflower and Orobanche cumana

[0029] Transplant the germinated seeds with the same root length into a seedling-raising pot with a substrate of peat soil and vermiculite (mass ratio 1:1). Take 200 mg of Orobanche cumana seeds and evenly mix them with 0.5 kg of the above substrate for co-culturing sunflowers and Orobanche cumana. Field trials showed that the sunflower variety SH361 is highly susceptible to Orobanche cumana, Sanrui No. 3 is highly resistant to Orobanche cumana, and TH15 is highly immune to Orobanche cumana. The culture conditions are as follows: light cycle 16 / 8 h, light intensity 300 μmol m -2 s -1 , day and night temperature is 24 / 20 °C, and relative humidity is 60%-70%.

[0030] 3. Morphological indicators

[0031] The morphological indicators include the above-ground and underground growth conditions of sunflowers, as well as the height, fresh weight, and dry weight of sunflowers. The fresh samples were placed in an oven at 70 ± 5 °C for 5 days to measure the dry weight.

[0032] The results are as Figure 1 shown.

[0033] Before the parasitism of Orobanche cumana Wallr. on sunflowers (when sunflowers and Orobanche cumana Wallr. were co-cultured for 10 days), there were no significant differences between the controls and treatments of each sunflower variety in the above-ground parts, and there was no parasitism of Orobanche cumana Wallr. among the varieties and treatments in the underground roots; after 17 days of co-culture of sunflowers and Orobanche cumana Wallr., the growth of the controls of highly susceptible and highly resistant varieties in the above-ground parts was slightly higher than that of the treatments, and Orobanche cumana Wallr. parasitism occurred in the underground roots. Moreover, the size and quantity of Orobanche cumana Wallr. on the highly susceptible varieties were greater than those on the highly resistant varieties. There were no significant differences between the controls and treatments of the immune variety in the above-ground and underground parts, and there was no Orobanche cumana Wallr. parasitism; after 24 days of co-culture of sunflowers and Orobanche cumana Wallr., the growth of the controls of highly susceptible and highly resistant varieties in the above-ground parts was significantly higher than that of the treatments, the size of Orobanche cumana Wallr. in the underground roots increased, and the size and quantity of Orobanche cumana Wallr. on the highly susceptible varieties were greater than those on the highly resistant varieties. There were no significant differences between the controls and treatments of the immune variety in the above-ground and underground parts, and there was no Orobanche cumana Wallr. parasitism; after 31 days of co-culture of sunflowers and Orobanche cumana Wallr. (two weeks after parasitism), the growth of the controls of highly susceptible and highly resistant varieties in the above-ground parts was significantly higher than that of the treatments, the size of Orobanche cumana Wallr. in the underground roots increased, and the size and quantity of Orobanche cumana Wallr. on the highly susceptible varieties were greater than those on the highly resistant varieties. The Orobanche cumana Wallr. on the highly susceptible varieties was about to emerge. There were no significant differences between the controls and treatments of the immune variety in the above-ground and underground parts, and there was still no Orobanche cumana Wallr. parasitism.

[0034] It can be Figure 1 seen that after being parasitized by Orobanche cumana Wallr., the plant height of sunflower plants decreased significantly in the highly susceptible variety SH361, and the degree of decrease increased with the prolongation of the parasitism time; the fresh weight and dry weight decreased significantly in the highly susceptible and highly resistant varieties compared with the controls one week after the parasitism of Orobanche cumana Wallr., and there were also significant differences in plant height, dry weight, and fresh weight among different varieties at the same time period and among treatments.

[0035] 4. Hyperspectral imaging system and image acquisition

[0036] The hyperspectral images of the sunflower canopy were acquired by a self-built visible and near-infrared hyperspectral imaging system. The hyperspectral imaging system consists of a visible and near-infrared (Vis-NIR) imaging spectrometer (ImSpector V10E; Spectral Imaging Ltd., Oulu, Finland) and a short-wave near-infrared (SWIR) imaging spectrometer (ImSpector N17E, Spectral Imaging Ltd., Oulu, Finland), (covering the bands of 380 - 1023 nm (resolution 2.8 nm) and 874 - 1734 nm (resolution 3.35 nm)), two high-performance cameras (Hamamatsu; Hamamatsu City, Japan & Xeva 992; Xenics Infrared Solutions, Leuven, Belgium) (resolutions of 672×512 (spatial × spectral) pixels and 320×256 (spatial × spectral) pixels respectively), two camera lenses (OLES22; Specim, Spectral Imaging Ltd., Oulu, Finland), two 150W halogen lamps (Fiber-Lite DC950 Illuminator; Dolan Jenner Industries Inc., Boxborough, MA, USA) and a stepping motor-driven conveyor belt.

[0037] The Vis-NIR images and SWIR images of the sunflower canopy were collected respectively. The potted sunflowers were placed on the conveyor belt. When collecting the Vis-NIR images, the distance between the lens and the sunflower canopy was adjusted to 27 cm, the camera exposure time was set to 35 ms, and the conveyor belt speed was set to 3.2 mm / s to ensure clear and distortion-free imaging. When collecting the SWIR images, the above parameters were set to 24 cm, 3.2 ms and 15 mm / s respectively.

[0038] The reflectance hyperspectral images were obtained after the original spectra were corrected by the black and white plate. The correction formula is as follows (Zhou & Leul, 1999):

[0039]

[0040] Ic is the corrected image, Iraw is the original image, W is the white plate image, and B is the black plate image.

[0041] In hyperspectral images, sunflower canopy leaves are the region of interest (ROI). The ROI is segmented by the threshold segmentation method, and the pixel spectra within the ROI range are extracted. Since there is random noise at the beginning and end of the spectra, the ranges of 408 - 1023 nm (488 bands) and 874 - 1734 nm (256 bands) are intercepted for subsequent research. The average spectrum of the pixel spectra of the ROI in each sample image is used as the sample spectrum. The Vis-NIR and SWIR spectra are directly spliced for low-level fusion. In subsequent research, models are established based on the Vis-NIR, SWIR, and fused spectra as input data respectively.

[0042] Example 1 Determination of Physiological Parameters

[0043] For three sunflower varieties with different resistances, including the highly susceptible variety SH361, the susceptible-resistant variety Sanrui 3, and the immune variety Tonghui 15, physiological indicators including SOD antioxidant enzyme activity, glutathione reductase GR, glutathione content (GSSG + GSH, GSH), phenylalanine ammonia-lyase, polyphenol oxidase, malondialdehyde content, and ROS were detected at four underground growth stages of sunflower broomrape during the early stage of parasitism, including before parasitism, just after parasitism, one week after parasitism, and two weeks after parasitism (a total of 10 days, 17 days, 24 days, and 31 days of cultivation).

[0044] 1. Determination of SOD Enzyme

[0045] The SOD activity was determined by the NBT method (Zhang et al., 2008). The reaction system was a total of 3 mL, containing 50 mM phosphate buffer (pH 7.8), 13 mM L-methionine, 75 μM NBT, 0.1 mM EDTA, 2 μM riboflavin, and 100 μl of enzyme extract. The reaction mixture was reacted for 20 min under the condition of 4000 lx, and colorimetric determination was carried out at a wavelength of 560 nm. The SOD activity was calculated with 50% inhibition of NBT photoreduction as one enzyme activity unit.

[0046] 2. Determination of GR

[0047] The method for determining GR activity was modified according to the method of Jiang and Zhang (2002). The reaction system was 3 mL, including 2.7 mL of 50 mM sodium phosphate buffer (pH 7.8, containing 2 mM of EDTA-Na2), 100 μl of 2.4 mM NADPH, 100 μl of 10 mM oxidized glutathione (GSSG), and 100 μL of supernatant. The blank zero-adjusting tube was added with 100 μL of extract (50 mM potassium phosphate buffer (pH 7.8)). The absorbances at 340 nm were measured for 10 s and 190 s in the reaction system, denoted as A1 and A2, and △A = A1 - A2. The extinction coefficient was 6.2 mM -1cm -1 。

[0048] 3. Determination of GSH + GSSG Content

[0049] The determination of GSH + GSSG content was modified according to the method of Law et al. (1983). 700 μL of 0.3 mM NADPH, 100 μL of 6 mM DTNB (5,5'-dithiobis(2-nitrobenzoic acid)), 50 μL of GR (10 U / mL), and 150 μL of supernatant were used. The change in absorbance at 412 nm was measured (using phosphate buffer instead of DTNB reagent as the blank control). In the blank tube, 120 μl of distilled water was used instead of the supernatant. The change in absorbance at 412 nm was measured.

[0050] 4. Determination of GSH Content

[0051] 1.5 mL of PBS (sodium phosphate buffer, 0.1 M, containing 5 mM EDTA, pH 8.0) mixture was added to 500 μL of supernatant. The mixture was adjusted to pH 8.0 with NaOH. 0.5 mL of the above mixture was taken and added to 1.4 mL of 0.1 M PBS (pH 8.0) and 100 μL of OPT (o-phthalaldehyde, 0.1 g kg -1 ) and mixed well. The reaction was carried out at room temperature for 15 min, and the change in absorbance at 412 nm was measured (using phosphate buffer instead of OPT reagent as the blank control).

[0052] 5. Determination of PAL

[0053] 0.1 g of fresh sample was weighed and added to 2 ml of boric acid buffer containing 5 mmol / L mercaptoethanol (containing a small amount of polyvinylpyrrolidone (PVP)) and ground in an ice bath. Then it was centrifuged (10000 rpm, 4 °C) for 15 min, and the supernatant was the crude enzyme extract. For the determination of phenylalanine ammonia-lyase: 1 mL of supernatant was added with 1 mL of 0.02 mol / L phenylalanine and 2 mL of distilled water (0.05 mol / L borate buffer solution), and 1 mL of distilled water (borate buffer solution) was used instead of the enzyme solution as the control. The reaction solution was incubated in a constant temperature water bath at 30 °C for 30 min, and then 0.2 mL of 6 mol / L HCl was added to terminate the reaction. The absorbance OD value was measured at 290 nm. The amount of enzyme required for an absorbance change of 0.01 per hour at 290 nm was defined as one unit (equivalent to the formation of 1 μg of cinnamic acid per milliliter of reaction mixture).

[0054] 6. Determination of PPO

[0055] Weigh 0.5 g of the material to be tested, add 0.30 g of PVPP, an appropriate amount of quartz sand and 10 mL of citric acid-phosphate buffer (pH 5.6). Grind it into a homogenate in an ice bath, then place it in a refrigerator at 4°C for extraction for 12 h. Then centrifuge at 8000 rpm for 15 min, and the supernatant is the crude enzyme solution. For the determination of polyphenol oxidase: Mix 1 mL of the crude enzyme solution with 3 mL of the reaction mixture (citric acid-phosphate buffer (pH 6.0) + 0.1% proline solution + 1.5% catechol solution mixed in a ratio of 10:2:3) in a constant temperature water bath at 37°C for 5 min respectively, and then mix them (Note: The above two solutions are mixed after water bath respectively, and the enzyme activity is measured once here). After reacting for 10 min, immediately add 3 mL of 1 mol / L metaphosphoric acid solution to terminate the reaction, and measure the absorbance at 460 nm. In the blank control solution, use the pH 6.0 buffer instead of catechol.

[0056] 7. Determination of MDA

[0057] The content of malondialdehyde was determined according to the method modified by Zhou and Leul (1999) using the thiobarbituric acid colorimetric method. Add 2 mL of the supernatant enzyme solution to 5 mL of 0.5% thiobarbituric acid (TBA, prepared with 10% trichloroacetic acid) solution and react in a water bath at 95°C for 30 min, then immediately place it in an ice bath; centrifuge at 5000 g for 10 min, take the supernatant and perform colorimetry at 532 nm and 600 nm, and the difference value is used to calculate the MDA content, and the extinction coefficient is 155 mM-1cm -1 .

[0058] 8. Determination of H2O2

[0059] The determination of H2O2 was carried out according to the method of Velikova et al. (2000). The reaction system was 2 mL, including 0.5 mL of the supernatant, 0.5 mL of 10 mM potassium phosphate buffer (pH 7.0) and 1 mL of 1 M potassium iodide. React at room temperature (28°C) for 1 h, and measure the absorbance at 390 nm. Calculate according to the standard curve.

[0060] 9. O2 - Determination

[0061] O2 - The determination of O2 was modified according to the method of Jiang and Zhang (2001). Mix 0.5 mL of the sample extract with 1 mL of 50 mM sodium phosphate buffer (pH 7.8) and 0.5 mL of 10 mM hydroxylamine hydrochloride, shake well, incubate at 25°C for 1 h, add 2 mL of 17 mM sulfanilic acid and 2 mL of 7 mM α-naphthylamine, mix, incubate at 25°C for 20 minutes, and measure the absorbance at 530 nm with a spectrophotometer.

[0062] Results:

[0063] Figure 3 andFigure 4 Shown separately are the changes in antioxidant enzyme activities, MDA, and ROS in the leaves of sunflowers with different resistance varieties during the four stages of Orobanche cumana parasitism.

[0064] Compared with the control group, after 17 days of co-culture with Orobanche cumana, the activities of SOD, PAL, and PPO in the leaves of highly susceptible and highly resistant sunflower varieties were significantly higher than those of the control, indicating that under Orobanche cumana infection, the production of ROS and others was inhibited by increasing the activities of antioxidant enzymes; under the same conditions, compared with the positive control, the activity of GR was significantly enhanced in the highly resistant variety Sanrui No. 3, and increased by 100.89% compared with the control; after 24 days of co-culture, compared with the positive control, the activity of GR was also significantly enhanced in both highly susceptible and highly resistant varieties; compared with the control group, after one week (24 days of co-culture) of Orobanche cumana parasitism, the PAL activity was significantly increased in both highly susceptible and highly resistant varieties, and after two weeks of parasitism, PAL was significantly increased in the highly resistant variety, but significantly decreased in the highly susceptible variety ( Figure 3 ).

[0065] Compared with the positive control, after 24 days of co-culture of sunflower with Orobanche cumana, the MDA content and the contents of H2O2 and O2 in highly susceptible and highly resistant varieties - both increased significantly; after 31 days of co-culture of sunflower with Orobanche cumana, the MDA content of the immune variety increased significantly, while the highly susceptible and highly resistant varieties decreased compared with the control group respectively ( Figure 4 ).

[0066] Generally speaking, different resistance varieties respond to the stress of Orobanche cumana infection by regulating the rapid accumulation of antioxidant enzymes, MDA, and ROS to resist the damage caused by stress.

[0067] Example 2 Rapid Quantitative and Visual Characterization of Physiological Indexes

[0068] For 9 physiological indexes including SOD, GR, MDA, H2O2, O2-, PAL, PPO, GSH, and GSH+GSSG, partial least squares regression (PLSR), least squares-support vector machine (LS-SVM), and radial basis function neural network (RBFNN) models were established based on Vis-NIR, SWIR, and fusion spectra respectively, and the optimal quantitative models for each index were optimized.

[0069] The characteristic bands of 9 physiological indexes were extracted by successive projections algorithm (SPA) and competitive adaptive reweighted sampling method (CARS) respectively, and models were established based on the characteristic wavelength subsets respectively, and the sensitive bands with the highest correlation of 9 physiological indexes were optimized according to the model performance.

[0070] Hyperspectral images contain both the spatial and spectral information of the sunflower canopy, overcoming the defect that traditional chemical detection methods cannot characterize the distribution of enzyme activities in the canopy of the same sunflower plant. First, the spectra of canopy pixel points are preprocessed by the multiplicative scatter correction (MSC) method and Savitzky-Golay smoothing, and then the images of each band are preprocessed by median filtering to eliminate the influence of random noise. The preprocessed pixel point spectra are input into the optimized quantitative model of physiological indicators to obtain the values of physiological indicators (enzyme activities) corresponding to each pixel point in the sunflower canopy, and the values of physiological indicators at different positions in the sunflower canopy are visually characterized by false color maps.

[0071] Results:

[0072] Partial least squares regression (PLSR), least squares support vector machine (LS-SVM), and radial basis function neural network (RBFNN) models for nine physiological indicators were established based on visible near-infrared (Vis-NIR), short-wave infrared (SWIR), and fused spectra, respectively. The methods, parameters, and results of the optimal models are shown in Table 1. Acceptable results (RPDs are all higher than 1.4) were obtained for the optimal models of each indicator, and excellent results were achieved for the optimal models of GR, PPO, GSH, and GSH+GSSG, with RPDs of 2.12, 3.12, 4.33, and 3.75, respectively. The results of the optimal models established by extracting features based on competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) are listed in Table 1, and the sensitive bands related to the nine physiological indicators are listed in Table 2 according to the optimal models.

[0073] Since relatively accurate results were obtained for the optimal models of the four indicators of GR, PPO, GSH, and GSH+GSSG, these four indicators were selected for distribution visualization research. Hyperspectral images of the canopies of a total of 12 sunflowers cultivated for 10 days and 31 days in 6 treatment groups were randomly selected, and quantitative distribution prediction maps of the four physiological indicators were obtained through pixel point spectra and the optimal models, as Figure 4 shown. Different enzyme activities have different distribution characteristics. GR and PPO are more distributed at the tips of the leaves, while the distributions of GSH and GSH+GSSG are more uniform.

[0074] The results show that sunflowers of different resistant varieties also exhibit different enzyme activity distribution characteristics at different stages of Orobanche cumana parasitism. The visualization of enzyme activity distribution based on hyperspectral is of great significance for quickly perceiving the physiological state of sunflowers under stress.

[0075] Table 1 Optimal quantitative models for nine physiological indicators based on full spectra and features

[0076]

[0077] 1 The model parameter of PLSR is nLv, the model parameters of LS-SVM are gam and sig2, and the model parameter of RBFNN is the number of hidden layer nodes.

[0078] Table 2 Sensitive Bands of 9 Physiological Indicators Screened by CARS or SPA

[0079]

[0080] Example 3 Identification of Sunflower Resistant Varieties and Parasite Identification

[0081] 1. Identification Model Based on Sensitive Spectral Bands

[0082] Since the physiological indicators of sunflowers with different resistant varieties show significant differences under different experimental treatments, it is speculated that the sensitive bands related to physiological indicators are also highly correlated with the resistant variety characteristics and parasitic characteristics of sunflowers. Combining data-driven and experience-driven methods, the sensitive spectral bands of physiological indicators are combined with SPA and CARS feature extraction methods to screen the sensitive spectral feature bands of sunflower resistant varieties and parasitic characteristics, which are used as input data to establish extreme learning machine (ELM), support vector machine (SVM) and partial least squares discriminant analysis (PLS-DA) models, screen the feature bands highly sensitive to the resistant variety characteristics and parasitic characteristics of sunflowers, and preferably select spectral-driven sunflower resistant variety identification models and early parasite identification models with simple structures and high accuracies.

[0083] Results:

[0084] The average reflectance spectra of infected plants, uninfected plants, and three resistant varieties of sunflowers are as Figure 5 shown. The reflectances of various sunflowers show different characteristics in different bands. The confusion matrix of the variety identification test set of the ELM model based on fused data, CARS for extracting fused data features, enzyme activity sensitive features, and SPA for re-extracting features from joint features is as Figure 6 shown.

[0085] For sunflower broomrape identification, the models based on the sensitive bands of 295 physiological indicators and the optimal models established by re - extracting the CARS or SPA features are shown in Table 3. The prediction accuracies of the training set and the test set of the ELM model based on the sensitive bands of physiological indicators are 98.96% and 97.92% respectively. The precision, recall and F1 - score of the test set are 90.91%, 100% and 95.24% respectively, indicating that the sensitive bands of physiological indicators are also highly correlated with the parasitic characteristics of sunflower broomrape. After re - extracting the CARS and SPA features, the number of independent variables of the model is reduced to 31 and 14 respectively. The prediction accuracies of the ELM models based on these variables are 97.92% and 95.83% respectively. Considering the model complexity and accuracy comprehensively, the ELM based on the re - extraction of the SPA features of the sensitive bands is selected as the optimal model, and the sensitive bands corresponding to the parasitic characteristics of sunflower broomrape are 410, 414, 433, 560, 661, 686, 710, 908, 918, 941, 942, 1656, 1693, 1707 nm.

[0086] The results show that for the identification of sunflower resistant varieties, since the physiological index differences between different resistant varieties are not very significant, 62 characteristic bands are extracted from the fused spectrum by CARS and combined with the sensitive bands of 295 physiological indicators to form a combined feature containing 331 bands. Models based on the combined feature and the re - extraction of CARS or SPA features are established respectively, and the optimal results are shown in Table 3. The result of the ELM model based on the re - extraction of the SPA features of the combined feature is the best, and the accuracies of the training set and the prediction set are 96.88% and 95.83% respectively, and the number of independent variables of the model is reduced to 12. Considering the model complexity and accuracy comprehensively, the ELM based on the re - extraction of the SPA features of the combined feature is selected as the optimal model, and the sensitive bands corresponding to the characteristics of sunflower broomrape resistant varieties are 408, 414, 434, 558, 696, 911, 931, 959, 1023, 1460, 1656, 1707 nm.

[0087] Table 3 Optimal model results based on different input spectra and feature screening methods

[0088]

[0089] Note: The sensitive bands in the table refer to the 295 characteristic bands sensitive to physiological indicators.

[0090] 2. Identification model based on sensitive 3 - band images

[0091] The hyperspectral images of sunflower canopy are high-dimensional data, which combine phenotypic image information and spectral information. On the basis of extracting characteristic bands in 1.5.1, the spectral and image features are further fused to extract and recombine the sensitive 3-band images. The process is as follows: all 3-band combinations are extracted from the sensitive characteristic bands, and an ELM model based on all 3-band spectral combinations is established. According to the classification accuracy of the ELM model, the 3-band combination with the highest classification accuracy is selected from them. Finally, the reflectance images corresponding to each of these 3 bands are extracted and recombined into a 3-channel image to achieve the dimensionality reduction and effective information extraction of high-dimensional hyperspectral images. The image dataset is augmented by data enhancement methods such as rotation, flipping, and adding noise. Randomly select 30% of the images in the original training set as the validation set, and the remaining 70% is still used as the training set. Train the pre-trained ResNet18 network. After parameter optimization, set the initial learning rate to 0.01, the validation frequency to 15, the max epoch to 30, and the minibatch size to 64.

[0092] Results:

[0093] The number of ROI pixels in the sunflower hyperspectral image canopy can approximately represent its canopy leaf area. The approximate leaf areas of different varieties at different times are as Figure 7 shown. Sunflowers of different resistant varieties show different trends of canopy leaf area change under different experimental treatments, and their image features can represent the phenotypic change information to a certain extent.

[0094] 76 and 40 3-band combinations are respectively formed from 14 characteristic bands sensitive to Orobanche cumana parasitism in sunflowers and 12 characteristic bands sensitive to sunflower resistant varieties. After establishing ELMs based on the three-band combinations respectively, the three-band combination with the highest classification accuracy is selected, as shown in Table 4. Based on all the selected three-band combinations, sensitive 3-band images are extracted from the hyperspectral images, and a CNN model based on the ResNet18 framework is established. The results of all models are shown in Table 4.

[0095] The results show that for the identification of Orobanche cumana parasitism in sunflowers, the model established with the three-band images of 411, 560, and 941 nm as input data has the highest classification accuracy of 95.83%, which is the same as the accuracy of the ELM model established based on 14 sensitive spectral bands. For the identification of sunflower resistant varieties, the model established with the three-band images of 1460, 1656, and 1707 nm as input data has the highest classification accuracy of 97.92%, which is higher than all the models based on spectra. It may be that the image features of sunflowers of different resistant varieties are relatively more significant.

[0096] Table 4 Results of CNN-based 3-band image sunflower variety and Orobanche cumana parasitism classification models

[0097]

Claims

1. A method for rapidly and non-destructively monitoring early underground parasitism of sunflower parasitic weeds, characterized in that: A method for modeling the parasitism status of sunflowers with different resistance levels at different parasitic stages using a hyperspectral imaging system through a CNN model; Specifically: for 9 physiological indexes including SOD, GR, MDA, H2O2, O2-, PAL, PPO, GSH, and GSH+GSSG, partial least squares regression, least squares-support vector machine, and radial basis function neural network models are established based on Vis-NIR, SWIR, and fusion spectra respectively, and the optimal quantitative models for each index are selected; The characteristic bands of 9 physiological indexes are extracted by successive projections algorithm and competitive adaptive reweighted sampling method respectively, and models are established based on the characteristic band subsets respectively, and the sensitive bands with the highest correlation of 9 physiological indexes are selected according to the model performance; The sensitive spectral characteristic bands of sunflower resistant varieties and parasitism characteristics are screened as input data, and extreme learning machine, support vector machine, and partial least squares discriminant analysis models are established to screen the characteristic bands highly sensitive to sunflower resistant variety characteristics and parasitism characteristics; Select the ELM extracted based on the SPA characteristics of sensitive bands as the optimal model, corresponding to the sensitive bands of sunflower Orobanche cumana parasitism characteristics; Select the ELM extracted based on the SPA characteristics of joint features as the optimal model, corresponding to the sensitive bands of sunflower Orobanche cumana resistant variety characteristics; Three-band combinations are respectively formed from the characteristic bands sensitive to sunflower Orobanche cumana parasitism and the characteristic bands sensitive to sunflower resistant varieties; after establishing ELM based on the three-band combinations respectively, the three-band combination with the highest classification accuracy is selected, and based on all the selected three-band combinations, sensitive three-band images are extracted from the hyperspectral images, and a CNN model based on the ResNet18 framework is established to identify sunflower resistant varieties and identify sunflower Orobanche cumana parasitism.

2. The method for rapid and non-destructive monitoring of early underground parasitism of sunflower parasitic weeds according to claim 1, characterized in that: A method for detecting and determining the parasitism status by using a hyperspectral imaging system to detect the activities of antioxidant enzymes, phenylalanine ammonia-lyase, and polyphenol oxidase.

3. The method for rapid and non-destructive monitoring of early underground parasitism of sunflower parasitic weeds according to claim 1, characterized in that: A method for detecting and determining the parasitism status by using a hyperspectral imaging system to detect the MDA content and ROS.

4. The method for rapid and non-destructive monitoring of early underground parasitism of sunflower parasitic weeds according to claim 1, characterized in that: A method for detecting and determining the parasitism status by using a hyperspectral imaging system to detect glutathione metabolism.

5. The method for rapid and non-destructive monitoring of early underground parasitism of sunflower parasitic weeds according to claim 1, characterized in that: Based on hyperspectra, a quantitative regression model of enzyme activity physiological indexes and a quantitative visualization characterization method for GR, PPO, GSH, and GSH+GSSG are established.

6. The method for rapid and non-destructive monitoring of early underground parasitism of sunflower parasitic weeds according to claim 1, characterized in that: A classification model for obtaining the presence or absence of parasitism and resistant varieties by hyperspectral technology.

Images