Calculation Method for Soybean Canopy Wilting Degree Using Fourier Transform of Multispectral Images

Abstract

The present invention belongs to a method for calculating the wilting degree of soybean canopy by using Fourier transform of multi-spectral images, and is realized through the following steps: acquiring soybean multi-spectral images, and performing denoising processing on the soybean reflection images; applying the iterative threshold method and the canopy extraction algorithm based on affine transformation to obtain the soybean canopy and conducting gray histogram statistical analysis on the canopy features of the soybean canopy; then performing Fourier transform on the soybean canopy, conducting spectral analysis on the canopy, extracting the direct current component in the spectrum of the multi-spectral image of the soybean canopy and calculating the proportion of the direct current component in the spectrum to obtain the wilting index of the soybean canopy, which can also be used to calculate the wilting degree caused by nutrient deficiencies such as nitrogen deficiency in soybeans or crop canopy wilting caused by saline-alkali stress, etc. The multi-spectral imaging technology has the advantages of low cost, convenient detection, high image resolution, etc. By obtaining the multi-spectral imaging information of crops, rapid non-destructive early detection of crop physiological and ecological information and canopy trait changes can be realized.

Description

Technical Field

[0001] The present invention belongs to the calculation of plant wilting degree using multi - spectral images, and particularly relates to a method for calculating the wilting degree of soybean canopy by using Fourier transform of multi - spectral images. Background Art

[0002] Soybean is a major oil crop in the world, providing important edible oil and high - protein feed for humans and animals, and playing an important role in China's national economy. Soybean has a large water requirement during growth and is the most sensitive leguminous crop to water stress. Drought will have a very serious impact on it. Water shortage will seriously affect the morphology of soybeans. When soybeans are under drought stress, a series of abnormal reactions will occur, such as leaf wilting, withering, and plant dwarfing. In severe cases, it will cause the death of the plant. Wilting is a phenomenon in which plant water deficit causes the young parts such as stems and leaves to droop, shrink, or curl because cells cannot maintain rigidity. It is a vital characteristic exhibited by plants under drought stress environments, related to the turgor pressure within the plant cell wall, and is an adaptive drought - avoidance mechanism based on insufficient water absorption by the roots. The wilting phenomenon of crop morphology reflects the water status within the body and is the most widely used plant drought - stress indicator. The change in the wilting degree of soybean leaf morphology is the external manifestation form of the comprehensive balance regulation of its internal water potential and complex production environment, and can directly and truly reflect the actual water deficit situation of soybean plants. The existing detection method and system for plant leaf water content based on multi - spectral images with the publication number CN102721651A also use multi - spectral images to detect the water content of leaves. However, the spectral channels are fewer, and the gray - scale texture feature quantity and leaf vegetation index of a single leaf of the plant are used as the basis for detecting the water content value of plant leaves. It is susceptible to the influence of individual factors of the selected test samples, has a large detection error, and is difficult to comprehensively detect the physiological and ecological traits of the plant canopy. For a method and system for calculating plant wilting degree with the publication number CN 112330694A, it calculates the proportion of the pixel number of the green part of maize in its visible area in the RGB image in the visible light as the wilting degree of maize, and there is a problem of being difficult to achieve early detection of canopy trait changes when plants are under water stress. Summary of the Invention

[0003] The present invention aims to overcome the deficiencies of the prior art and provides a method for calculating the wilting degree of soybean canopy by using Fourier transform of multi - spectral images.

[0004] The method for calculating the wilting degree of soybean canopy by using Fourier transform of multi - spectral images according to the present invention is realized through the following steps:

[0005] (1). Obtain the multi - spectral image of soybean, and obtain the reflection image of the multi - spectral image of soybean by the empirical linear method;

[0006] (2), Denoise the soybean reflection image;

[0007] (3), Apply the iterative threshold method and the canopy extraction algorithm based on affine transformation to obtain the soybean canopy and perform gray histogram statistical analysis on the canopy features of the soybean canopy;

[0008] (4), Then perform Fourier transform on the soybean canopy, perform spectral analysis on the canopy, extract the DC component in the spectrum of the multi-spectral image of the soybean canopy and calculate the proportion of the DC component in the spectrum, and obtain the wilting index of the soybean canopy.

[0009] As a further improvement of the present invention, it is specifically realized through the following steps:

[0010] (1), Use a Parrot Sequoia multi-spectral camera to obtain multi-spectral images of soybeans. The multi-spectral images are respectively green light spectral images, red light spectral images, red edge spectral images and near-infrared spectral images; according to the Sequoia calibration equation, use the empirical linear method to calculate the reflection images of the original multi-spectral images of soybeans, and respectively obtain the green light spectral reflection image, near-infrared spectral reflection image, red light spectral reflection image and red edge spectral reflection image of soybeans, and calculate the corresponding normalization values;

[0011] (2) Perform noise reduction processing on the green light spectral reflection image, near-infrared spectral reflection image, red light spectral reflection image and red edge spectral reflection image obtained in step (1) by using the median filtering method;

[0012] (3) Use the iterative threshold method to obtain the canopy of the near-infrared spectral reflection image of soybeans for the images processed in step (2). Take the near-infrared canopy image of soybeans as a template to identify the green light spectral reflection image, red light spectral reflection image and red edge spectral reflection image of soybeans. Apply the affine transformation calculation method to construct the affine transformation model of the reference image and the target area in the green light spectral reflection image, red light spectral reflection image and red edge spectral reflection image of soybeans, so as to extract the target area of the canopy of the green light spectral reflection image, red light spectral reflection image and red edge spectral reflection image of soybeans, perform gray histogram statistics on it and obtain and analyze the canopy features;

[0013] (4) Perform Fourier transform on the canopies of the near-infrared spectral reflection image, green light spectral reflection image, red light spectral reflection image and red edge spectral reflection image of soybeans obtained in step (3), perform spectral analysis on the canopy to obtain the amplitude spectrum diagram, and then obtain the energy spectrum diagram. Extract the DC component in the energy spectrum diagram and calculate the proportion of the DC component in the total energy of the spectrum to obtain the wilting index L of the soybean canopy as

[0014]

[0015] Wherein, ln(·) represents the natural logarithm function loge(·), and |F(0,0)| 2 represents the energy value of the DC component in the energy spectrum, and represents the total energy value of the frequency spectrum in the energy spectrum.

[0016] As a further improvement of the present invention, the Fourier transform calculation method in the step (4) is the time extraction radix-2 FFT calculation method.

[0017] The present invention adopts a method for calculating the wilting degree of soybean canopy by multi-spectral image Fourier transform, which is not only limited to calculating the wilting degree of soybean canopy caused by water stress, but can also be used to calculate the wilting degree of soybean canopy caused by nutrient deficiency such as nitrogen deficiency or crop canopy wilting caused by saline-alkali stress. By adopting multi-spectral imaging technology, it has the advantages of low cost, convenient detection and high image resolution. By obtaining multi-spectral imaging information of crops, rapid non-destructive early detection of crop physiological and ecological information and canopy trait changes can be realized. The present invention combines multi-spectral imaging processing technology and Fourier transform method, providing a new idea for calculating the wilting degree of crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is the original multi-spectral image of soybean and the corresponding histogram;

[0019] Figure 2 is the multi-spectral reflection image of soybean and the corresponding histogram;

[0020] Figure 3 is the median filtering effect of the multi-spectral image of soybean and the corresponding histogram;

[0021] Figure 4 is the iterative threshold segmentation of the normal processed soybean canopy and the corresponding histogram;

[0022] Figure 5 is the iterative threshold segmentation of the drought processed soybean canopy and the corresponding histogram;

[0023] Figure 6 is the normal processed soybean canopy and the corresponding histogram;

[0024] Figure 7 is the drought processed soybean canopy and the corresponding histogram;

[0025] Figure 8 is the amplitude spectrum of the multi-spectral image of soybean;

[0026] Figure 9 is the energy spectrum of the multi-spectral image of soybean;

[0027] Figure 10 is the soybean canopy at V4 stage and the corresponding frequency spectrum diagram;

[0028] Figure 11 It is the soybean canopy at V5 stage and the corresponding spectral diagram. Detailed implementation mode

[0029] Example 1

[0030] The method for calculating the wilting degree of soybean canopy using multi - spectral image Fourier transform of the present invention is realized through the following steps:

[0031] (1). Use a Parrot Sequoia multi - spectral camera to obtain multi - spectral images of soybeans. The multi - spectral images are respectively a green - light spectral image, a red - light spectral image, a red - edge spectral image, and a near - infrared spectral image; according to the Sequoia calibration equation, use the empirical linear method to calculate the reflection images of the original multi - spectral images of soybeans, and respectively obtain the green - light spectral reflection image, near - infrared spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image of soybeans, and calculate the corresponding normalized values;

[0032] (2) Use the median filtering method to perform noise reduction processing on the green - light spectral reflection image, near - infrared spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image obtained in step (1);

[0033] (3) Use the iterative threshold method to obtain the soybean near - infrared spectral reflection image canopy for the images processed in step (2). Take the soybean near - infrared canopy image as a template for identifying the green - light spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image of soybeans. Apply the affine transformation calculation method to construct the affine transformation model of the target area between the reference image and the green - light spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image of soybeans, so as to extract the target area of the green - light spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image canopy of soybeans, perform gray - level histogram statistics on it, and obtain the analysis of canopy characteristics;

[0034] (4) Perform Fourier transform on the canopies of the soybean near - infrared spectral reflection image, green - light spectral reflection image, red - light spectral reflection image, and red - edge spectral reflection image obtained in step (3). Perform spectral analysis on the canopy to obtain the amplitude spectrum diagram, and then obtain the energy spectrum diagram. Extract the DC component in the energy spectrum diagram and calculate the proportion of the DC component in the total energy of the spectrum to obtain the wilting index L of the soybean canopy as

[0035]

[0036] In the formula, ln(·) represents the natural logarithm function loge(·), |F(0,0)| 2 represents the energy value of the DC component in the energy spectrum, represents the total energy value of the spectrum in the energy spectrum.

[0037] Example 2

[0038] Step 1: Obtain the reflection image of the soybean multispectral image

[0039] Under outdoor conditions with a temperature of 20 - 34°C, select the soybean variety Suinong 26 as the research object. The characteristics of this soybean variety are that the leaves are long and pointed. Soybeans are sensitive to light intensity, environmental temperature, and soil moisture content, and changes in the environment can easily cause morphological changes in plant leaves. When the water is sufficient, the plant leaves are flat and stretched; when suffering from drought stress, the plant leaves wilt. After the seeds are selected, they are disinfected, germinated, and sown. The specific operations are as follows: First, spread medium-sized soil blocks evenly at the bottom of a planting box made of polyvinyl chloride (PVC) material with a diameter of 0.3 m and a height of 0.18 m, fill it with selected non-saline fine soil until the planting box weighs 5 kg, evenly spread compound fertilizer, and then continue to put fine soil in the planting box until it weighs 8.32 kg. After emergence, when the seedlings grow to the V1 stage, thin the soybeans according to requirements. This experiment is divided into normal water supply treatment and drought treatment. At the V1 stage, water control is carried out on half of the soybean plants (no water is added all the time). On the 15th day after water control (V4 stage), the first data is collected for all the soybeans planted in the planting boxes, and on the 20th day (V5 stage), the second data is collected.

[0040] Use a Parrot Sequoia multispectral camera to collect the multispectral images of soybeans. This camera has a total of 5 imaging sensors, including 4 multispectral sensors and 1 RGB sensor. The four multispectral bands are green light (central wavelength 550 nm, band width 40 nm), red light (central wavelength 660 nm, band width 40 nm), red edge (central wavelength 735 nm, band width 10 nm), and near-infrared (central wavelength 790 nm, band width 40 nm). To more clearly observe the morphological characteristics of soybean wilting and ensure the integrity of the collected images, the experiment is carried out by shooting vertically, and the camera height is set at 1.6 meters from the soybean canopy. Obtain the green light spectral image, red light spectral image, red edge spectral image, and near-infrared spectral image of soybeans, and the image size is 1280 pixels × 960 pixels for all.

[0041] Since the Sequoia camera records DN (Digital Number) values, this method uses the empirical linear method to convert the DN values of the multispectral images into reflection values, obtaining the green light spectral reflection image, near-infrared spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of soybeans. The specific method is as follows:

[0042] (1) Calculation of the multispectral camera calibration equation

[0043] The linear inversion formula between the gray value and the reflectance of the multi-spectral image is calculated by the least squares method. The least squares method finds the best function match for the data by minimizing the sum of the squares of the errors. Using the least squares method, the unknown data can be easily obtained, and the sum of the squares of the errors between the obtained data and the actual data is minimized. Let the dependent variable y be related to the independent variable x, and now n sets of independent observation data (x i , y i ) are obtained. Then the model of the unary linear regression equation can be expressed as:

[0044]

[0045] In the formula, a is the linear relationship coefficient and b is the constant.

[0046] The least squares estimation is used to estimate the unknown parameters a and b of the unary linear regression equation. Let

[0047]

[0048] Then the least squares estimation of the regression coefficient is to make Q reach the minimum, specifically:

[0049]

[0050] Among them,

[0051] After obtaining , the expression of the unary regression equation can be written as:

[0052]

[0053] Since the reflectance and the gray value are parameters that linearly describe the illumination intensity of the standard gray scale, the linear conversion between the reflectance and the gray value of a certain band of the multi-spectral image can be expressed as

[0054] R i = kg i + b (5)

[0055] In the formula, R i represents the reflectance of the i-th band, g i represents the gray value of the i-th band, k is the linear relationship coefficient, and b is the constant.

[0056] According to the above least squares method principle, the calibration equations of the green light spectral reflectance image, the soybean near-infrared spectral reflectance image, the red light spectral reflectance image, and the red edge spectral reflectance image of the multi-spectral camera Sequoia are calculated as shown in Table 1.

[0057] Table 1 Sequoia Calibration Equation

[0058] Band Formula Green light spectral reflection image y = 0.0000085642x - 0.030957 Near-infrared spectral reflection image y = 0.0000035813x - 0.044039 Red light spectral reflection image y = 0.000020052x - 0.20474 Red edge spectral reflection image y = 0.000040562x - 0.47423

[0059] Note: y is the reflectance and x is the DN value.

[0060] (2) Acquisition of soybean multispectral reflection images

[0061] Perform channel analysis on the original soybean multispectral images to calculate the gray value distributions of the green light spectral reflection image, soybean near-infrared spectral reflection image, red light spectral reflection image, and red edge spectral reflection image, respectively, as Figure 1 shown.

[0062] From Figure 1 it can be seen that in the original soybean multispectral images, Figure 1 (a1) The gray frequency distribution of the green light image is between 0.1098 and 0.5255. The corresponding histogram (b1) reaches a peak at 0.3843, with a peak value of 26552, the sum of image pixels is 397896, the average gray value is 0.3238, the histogram is bimodal, and it is at the trough position when the gray value is 0.2941. Figure 1 (a2) The gray frequency distribution of the near-infrared image is between 0.1176 and 0.8431. Its histogram (b2) reaches a peak at 0.3961, with a peak value of 37358, the sum of image pixels is 433785, the average gray value is 0.3530, the histogram is bimodal, and it is at the trough position at 0.3098. Figure 1 (a3) The gray frequency distribution of the soybean red light image is between 0.1020 and 0.5686. The corresponding histogram (b3) reaches a peak at 0.4314, with a peak value of 25737, the sum of image pixels is 414746, the average gray value is 0.3375, the histogram is bimodal, and it is at the trough position at 0.3098. Figure 1 (a4) The gray frequency distribution of the red edge image is between 0.0980 and 0.4667. The corresponding histogram (b4) reaches a peak at 0.2510, with a peak value of 61388, the sum of image pixels is 266031, the average gray value is 0.2165, the histogram is bimodal, and it is at the trough position at 0.2039.

[0063] According to the Sequoia calibration equation established in Table 1, calculate the reflection images of the original soybean multispectral images, respectively obtain the soybean green light spectral reflection image, soybean near-infrared spectral reflection image, red light spectral reflection image, and red edge spectral reflection image, and calculate the corresponding gray value histograms, as Figure 2 shown.

[0064] From Figure 2 it can be seen that in the soybean multispectral reflection images, Figure 2(a1) The gray - level frequency distribution of the green - light image is between 0.0314 and 0.2706. The corresponding histogram (b1) reaches a peak at 0.1765, with the peak value being 48461, the sum of image pixels being 185262, and the average gray - level value being 0.1508. The histogram is bimodal, and the left peak is lower than the right peak. The envelope line is basically consistent with the trend of the original image histogram. The distance between the two peaks has decreased by 0.0588, and it is at the trough position at 0.1373, shifted 0.1568 to the left compared to the original image. Figure 2 (a2) The gray - level frequency distribution of the near - infrared image is between 0 and 0.1567. Its histogram (b2) reaches a peak at 0.0471, with the peak value being 163849, the sum of image pixels being 47753, and the average gray - level value being 0.0394. The histogram is bimodal, and the left peak is lower than the right peak. The envelope line is basically consistent with the trend of the original image histogram. The distance between the two peaks has decreased by 0.1137, and it is at the trough position at 0.0275, shifted 0.2823 to the left compared to the original image. Figure 2 (a3) The gray - level frequency distribution of the red - light image is between 0 and 0.5137. The corresponding histogram (b3) reaches a peak at 0.3490, with the peak value being 22535, the sum of image pixels being 293878, and the average gray - level value being 0.2449. The histogram is bimodal, and the left peak is lower than the right peak. The envelope line is basically consistent with the trend of the original image histogram. The distance between the two peaks has increased by 0.1019, and it is at the trough position at 0.2039, shifted 0.1059 to the left compared to the original image. Figure 2 (a4) The gray - level frequency distribution of the red - edge image is between 0 and 0.7059. Its histogram (b4) reaches a peak at 0.1765, with the peak value being 29350, the sum of image pixels being 141031, and the average gray - level value being 0.1569. The number of adjacent gray - level values in the histogram fluctuates greatly, but the envelope line is still basically consistent with the trend of the original image histogram. It is at the trough position at 0.0627, shifted 0.1412 to the left compared to the original image. After analysis, after converting the original multi - spectral image of soybeans into a reflectance image, the histograms of the four spectral reflectance images are translated 0.0980 - 0.1216 to the left as a whole. The regular change of the characteristic values of the reflectance image is basically consistent with that of the original image. Therefore, this method uses the multi - spectral reflectance image to carry out the calculation research on the wilting degree of soybeans.

[0065] Step 2: Denoising of the multi - spectral image of soybeans

[0066] To overcome the influence of noise during the image acquisition process, this application applies median filtering to perform noise reduction processing on the multi - spectral image and conducts histogram statistical analysis to remove the influence of interference factors on canopy extraction.

[0067] Median filtering replaces the central pixel value with the median of all pixels within the window. The specific implementation method is to first sort all pixel values within the filtering window by size, and then select the median of the sorting result as the central pixel value of the output. Such a processing method makes isolated noise pixel points have little impact on the filtering result, thus achieving a denoising effect. In practical applications, in order to balance denoising and protecting image detail information, it is necessary to select the window size according to the local statistical characteristics or features of the image.

[0068] Let the gray value of the pixel at position (x, y) in the image be f(x, y). After median filtering, the pixel value at this point is g(x, y). Then we have

[0069] g(x,y) = Med{f(x,y)} (x,y) ∈ Z (6)

[0070] Among them, Z represents the set of pixel points covered by the neighborhood template window centered at the point (x, y), and Med represents taking the median value of all numerical values in a set.

[0071] According to the median filtering formula (6), a 3×3 filtering window is selected to perform noise reduction processing on the four spectral reflection images of the soybean multispectral image and conduct histogram statistical analysis, as Figure 3 shown.

[0072] After median filtering the soybean multispectral image, it can be seen that Figure 3 (a1) The gray frequency distribution of the soybean green spectral reflection image is between 0.0275 and 0.2745. The corresponding histogram (b1) of the image reaches a peak at 0.1765, the peak value is 49549, the sum of image pixels is 184950, the average gray value is 0.1505. After filtering, the peak value increases by 1088, the sum of pixels decreases by 312, and the average gray value remains basically unchanged. Figure 3 (a2) The gray frequency distribution of the near-infrared spectral reflection image is between 0 and 0.1569. The corresponding histogram (b2) of the image reaches a peak at 0.0471, the peak value is 176518, the sum of image pixels is 47577, the average gray value is 0.0392. After filtering, the peak value increases by 12669, the sum of pixels decreases by 176, and the average gray value remains basically unchanged. Figure 3 (a3) The gray frequency distribution of the red light spectral reflection image is between 0 and 0.5333. The corresponding histogram (b3) of the image reaches a peak at 0.3490, the peak value is 23388, the sum of image pixels is 293164, the average gray value is 0.2441. After filtering, the peak value increases by 853, the sum of pixels decreases by 714, and the average gray value remains basically unchanged. Figure 3(a4) The gray frequency distribution of the red-edge spectral reflection image is between 0 and 0.7608. The corresponding histogram (b4) of the image reaches a peak at 0.1686, with the peak value being 32125. The sum of image pixels is 139525, and the average gray value is 0.1555. After filtering, the peak value increases by 2775, the pixel sum decreases by 1506, and the average gray value remains basically unchanged.

[0073] Table 2 is the evaluation index for median filtering preprocessing of multi-spectral images

[0074]

[0075] As can be seen from Table 2, the maximum value of PSNR for the multi-spectral image processed by median filtering is 49.074 for the near-infrared spectral reflection image, and the average value of the four spectral reflection images is 40.7570. The maximum value of SSIM is 0.989 for the near-infrared spectral reflection image, and the average value of the four spectral reflection images is 0.9333. The minimum value of MAE is 0.257 for the near-infrared spectral reflection image, and the average value of the four spectral reflection images is 0.9678. When denoising the multi-spectral image, the multi-spectral image processed by median filtering retains the details such as the edges of the soybean multi-spectral image in terms of PSNR, SSIM, and MAE, and can relatively completely preserve the original information of the soybean multi-spectral image.

[0076] Step 3: Extraction of soybean canopy multi-spectral images

[0077] (1) Use the iterative threshold method to obtain the soybean near-infrared spectral reflection image canopy

[0078] The iterative threshold method is used to segment the soybean multi-spectral image canopy, and the optimal value is obtained through various maximum or minimum decision functions. No matter which algorithm is used to select the threshold T, for an original soybean image f(x, y), the segmented image can be defined as:

[0079]

[0080] In the formula, the image g(x, y) is a binary image, which realizes the division of the foreground and background regions of the original image.

[0081] The iterative threshold method is an algorithm that can automatically select the threshold based on image data. First, an initial threshold is selected, and then the threshold is continuously updated according to a certain strategy until the convergence criterion given by the algorithm is met. Take the median of the gray range of the soybean image as the initial threshold T0 (assuming the image has L grayscales), and the iterative formula of this algorithm is:

[0082]

[0083] In the formula, h k is the number of pixels with a gray value of k in the image; when iterating to Ti+1 = T i When it is determined that T i is the segmentation threshold.

[0084] After segmenting the soybean canopy, the image is subjected to an opening operation, that is, in the image, an erosion operation is first performed using a structuring element and then a dilation operation is performed to refine the image to eliminate isolated point burrs, and the soybean canopy image information is reliably retained. Among them, the opening operation formula is:

[0085]

[0086] In the formula, X Θ S represents the erosion operation, represents the dilation operation.

[0087] When using the iterative threshold method to segment the normally processed soybean multispectral image, the thresholds of the near-infrared multispectral image are 0.0764 respectively. The near-infrared multispectral image of the soybean canopy is calculated and obtained. At the same time, the gray value histogram of the near-infrared multispectral image of the soybean canopy is calculated, and the effect of the soybean canopy multispectral image and the corresponding histogram are obtained, as Figure 4 shown.

[0088] From Figure 4 it can be seen that in the soybean canopy multispectral image segmented by the iterative method, Figure 4 (a2) The gray frequency distribution of the near-infrared soybean canopy image is between 0.0745 and 0.1725. The corresponding histogram (b2) reaches a peak at 0.1098, the peak value is 3130, the sum of all canopy pixels is 4788, and the average gray value is 0.1164. The target canopy area is segmented relatively completely by the iterative method.

[0089] When using the iterative threshold method to segment the drought-treated soybean multispectral image, the thresholds of the near-infrared multispectral image are 0.0905 respectively. The near-infrared multispectral image of the soybean canopy is calculated and obtained. At the same time, the gray value histogram of the soybean canopy multispectral image is calculated, and the effect of the soybean canopy multispectral image and the corresponding histogram are obtained, as Figure 5 shown.

[0090] From Figure 5 it can be seen that in the soybean canopy multispectral image segmented by the iterative method, Figure 5 (a2) The gray frequency distribution of the near-infrared soybean canopy image is between 0.0902 and 0.1961. The corresponding histogram (b2) reaches a peak at 0.1922, the peak value is 9326, the sum of all canopy pixels is 3658, and the average gray value is 0.1688. The target canopy area is segmented relatively completely.

[0091] The iterative threshold method has the most ideal segmentation effect on the near-infrared spectral reflection image of soybeans. When segmenting the red-edge spectral reflection image, although most of the target canopy area can be segmented, there will be cases where some leaves are missing or some background areas are retained. The iterative threshold method has an unsatisfactory segmentation effect on the green-light spectral reflection image and the red-light spectral reflection image of soybeans. Since the sizes of the green-light, near-infrared, red-light, and red-edge spectral reflection images of soybeans are all 1280 pixels × 960 pixels, therefore, the near-infrared image of the soybean canopy is used as a template to segment the canopies of the green-light, red-light, and red-edge spectral reflection images of soybeans.

[0092] (2) Extraction of multi-spectral image canopy based on a standard template

[0093] Use the near-infrared reference image of soybeans obtained in the above step (1) as a template to identify the green-light spectral reflection image, red-light spectral reflection image, and red-edge spectral reflection image of soybeans. Apply affine transformation to calculate and select the optimal registration parameters of the heterologous images, and construct an affine transformation model of the reference image and the target areas in the green-light spectral reflection image, red-light spectral reflection image, and red-edge spectral reflection image of soybeans, so as to extract the target areas of the canopies of the green-light spectral reflection image, red-light spectral reflection image, and red-edge spectral reflection image of soybeans.

[0094] Affine transformation is a spatial transformation superimposed by two simple transformations, namely linear transformation and translation transformation. Its spatial changes include spatial position changes such as translation, rotation, scaling, shearing, and reflection

[22] . For the image undergoing affine transformation, the collinear points before and after the transformation still have a collinear relationship, the parallel lines still remain parallel, and the ratio of the variable line segments remains unchanged. Therefore, the image also has the characteristics of collinearity, parallelism, and invariance of collinear ratio after affine transformation

[23] .

[0095] The affine transformation in the two-dimensional Euclidean space can be expressed as:

[0096]

[0097] All geometric transformations can be represented in the form of the product of a square matrix and a column vector by writing them in homogeneous form. Equation (10) written in homogeneous coordinate form is:

[0098]

[0099] In the formula, (x, y) and (x′, y′) are the coordinates of two corresponding points in the plane, and (a0, a3) T is the translation vector, is the matrix of the combined transformation of rotation, scaling, and shearing, and a i are all real numbers.

[0100] The registration method for the near-infrared canopy reference image of soybean with the green light spectral reflectance image, red light spectral reflectance image, and red-edge spectral reflectance image of soybean is to use the template matching calibration method. The extracted near-infrared image of the soybean canopy is used as the reference image for canopy extraction of the green light, red light, and red-edge spectral reflectance images of soybean. Through the affine transformation algorithm, the matching of the near-infrared canopy reference image with the green light, red light, and red-edge spectral reflectance images is completed, and then the soybean canopy is accurately extracted.

[0101] First, the feature points (x, y) and (x′, y′) are calibrated in the regions of the near-infrared reference image of soybean, the green light spectral reflectance image of soybean, the red light spectral reflectance image of soybean, and the red-edge spectral reflectance image of soybean respectively. Then, the corresponding point pairs (x, y) and (x′, y′) in the canopy target images of the near-infrared reference image and the green light, red light, and red-edge spectral reflectance images are found. Finally, a matching model is established through the corresponding feature points to achieve the matching of multi-source images. During the image registration process, linear geometric transformation is adopted, and the analysis of the successive distortion results of the image to be registered is used as the basis for registering the reference image. Then, according to the geometric distortion situation between the near-infrared reference image of the soybean canopy and the target images of the green light and red light spectral reflectance images, the best-fitting geometric transformation model between the two images, that is, the affine transformation model, is selected. The present application adopts the most commonly used affine transformation model, and the mathematical expression for mapping the points in the reference image (x, y) to the image to be registered (x′, y′) is:

[0102]

[0103] In the formula, (t x , t y ) T is the translation amount.

[0104] In the present application, taking the near-infrared image of the soybean canopy as the reference image, the affine transformation model for matching the green light spectral reflectance image of soybean is:

[0105]

[0106] In the formula, (x, y) is the near-infrared spectral reflectance image of the soybean canopy, (x′, y′) is the green light spectral reflectance image of soybean, (-42, -30) T is the translation amount.

[0107] The affine transformation model for matching the red light spectral reflectance image of soybean is:

[0108]

[0109] In the formula, (x, y) is the near-infrared spectral reflectance image of the soybean canopy, (x′, y′) is the red light spectral reflectance image of soybean, (-30, 7) T is the translation amount.

[0110] The affine transformation model for matching the soybean red-edge spectral reflection image is as follows:

[0111]

[0112] where \((x, y)\) is the near-infrared spectral reflection image of the soybean canopy, \((x′, y′)\) is the soybean red-edge spectral reflection image, and \((-6, -20)\) T is the translation amount.

[0113] After the template coordinate transformation, the target image recognition of the soybean green, red, and red-edge spectral reflection images is divided into the following three processes.

[0114] (a) Read the elements \(M\) of the \(i\)-th row and \(j\)-th column of the segmented near-infrared reference image matrix in sequence ij and the elements \(N\) of the \(i\)-th row and \(j\)-th column in the green, red, and red-edge spectral reflection image matrices ij values.

[0115] (b) Determine whether the value of \(M\) ij is 0. If \(M\) ij is 0, then set the value of \(N\) ij to 0; if the value of \(M\) ij is not 0, then the value of \(N\) ij remains unchanged.

[0116] (c) Output the matrix \(N\) ij , denoted as the image after the segmentation of the canopy area and background area of the soybean green, red, and red-edge spectral reflection images. Finally, the soybean green spectral reflection image, red, and red-edge spectral reflection image canopies are obtained.

[0117] When using affine transformation to segment normal soybean multispectral images, the normal processed soybean near-infrared canopy image is used as a template to calculate and obtain the green, red, and red-edge multispectral images of the soybean canopy. At the same time, calculate the gray value histogram of the soybean canopy multispectral image to obtain the effect of the soybean canopy multispectral image and the corresponding histogram, as Figure 6 shown.

[0118] From Figure 6 it can be seen that in the soybean canopy multispectral image segmented by affine transformation, Figure 6 (a1) The gray frequency distribution of the green soybean canopy image is between 0.0353 and 0.2392. The corresponding histogram (b1) reaches a peak at 0.1255, the peak value is 3081, the sum of all canopy pixels is 4932, and the average gray value is 0.1282. The target canopy area is segmented relatively completely by affine transformation. Figure 6(a2) The gray - level frequency distribution of the red - light soybean canopy image is between 0 and 0.2941. Its histogram (b2) reaches a peak at 0.0118, with a peak value of 2556. The sum of all canopy pixels is 974, and the average gray - level value is 0.0369. The color of the canopy and the background area is too close, and most of the target canopy area is segmented out. Figure 6 (a3) The gray - level frequency distribution of the red - edge soybean canopy image is between 0 and 0.8078. The corresponding histogram (b3) reaches a peak at 0.4667, with a peak value of 673. The sum of all canopy pixels is 21437, and the average gray - level value is 0.4544. The target canopy area is segmented out relatively completely by using affine transformation.

[0119] When using affine transformation to segment the multi - spectral image of drought - treated soybeans, taking the near - infrared canopy image of drought - treated soybeans as a template, calculating and obtaining the green - light, red - light, and red - edge multi - spectral images of the soybean canopy. At the same time, calculating the gray - level value histogram of the multi - spectral image of the soybean canopy, obtaining the effect of the multi - spectral image of the soybean canopy and the corresponding histogram, as Figure 7 shown.

[0120] From Figure 7 it can be seen that in the multi - spectral image of the soybean canopy segmented by using affine transformation, Figure 7 (a1) The gray - level frequency distribution of the green - light soybean canopy image is between 0.0628 and 0.2980. Its histogram (b1) reaches a peak at 0.1765, with a peak value of 979. The sum of all canopy pixels is 3266, and the average gray - level value is 0.1766. The target canopy area can be segmented out relatively completely by using affine transformation. Figure 7 (a2) The gray - level frequency distribution of the red - light soybean canopy image is between 0 and 0.4078. The corresponding histogram (b2) reaches a peak at 0.0667, with a peak value of 682. The sum of all canopy pixels is 2248, and the average gray - level value is 0.1227. The target canopy area is segmented out relatively completely. Figure 7 (a3) The gray - level frequency distribution of the red - edge soybean canopy image is between 0 and 0.7804. Its histogram (b3) reaches a peak at 0.5451, with a peak value of 305. The sum of all canopy pixels is 9351, and the average gray - level value is 0.4763. The soybean canopy is segmented out relatively completely by using affine transformation.

[0121] In the segmentation results of the soybean multi - spectral image, calculating the effective segmentation rate, over - segmentation rate, under - segmentation rate, and algorithm segmentation time of the soybean multi - spectral image, and conducting statistical analysis on the four multi - spectral canopy images of green - light, near - infrared, red - light, and red - edge. Among them, the higher the effective segmentation rate and the lower the over - segmentation rate and under - segmentation rate, the better the algorithm segmentation performance. The results are shown in Table 3.

[0122] Table 3 Results of image segmentation evaluation indicators

[0123]

[0124]

[0125] As can be seen from the statistical results in Table 3, the average effective segmentation rate of the soybean image canopy using the iterative threshold method and the canopy extraction algorithm based on affine transformation is above 95%, and the recognition effect is good; both algorithms effectively distinguish the canopy and background in the four spectral band images, with an average effective segmentation rate of 97.02%, an average under-segmentation rate of 2.64%, and an average over-segmentation rate of 1.83%; the effective segmentation rate of the iterative threshold method for the near-infrared spectral reflection image is the highest at 98.17%, and the under-segmentation rate is the lowest at 1.06%; the effective segmentation rates of the canopy extraction algorithm based on affine transformation for the green light spectral reflection image, red light spectral reflection image, and red edge spectral reflection image are all above 95%. Among them, due to the close color of the canopy and background regions in the red light reflection image, the under-segmentation rate is the highest at 4.29%; in terms of calculation speed, the average processing times of the iterative threshold method and the canopy extraction algorithm based on affine transformation are 0.8703 seconds and 0.7301 seconds respectively; considering the extraction effect of the soybean canopy image comprehensively, the extraction effects of both algorithms on the soybean canopy have reached an ideal state.

[0126] Step 4: Calculation method of soybean wilting degree based on Fourier transform

[0127] (1) The Fourier transform method can transform the image signal information from the time domain to the frequency domain. Information that is difficult to observe in the time domain becomes easier to observe after being transformed to the frequency domain. After being decomposed into sine signals or cosine functions of different frequencies and then superimposed, the structural analysis of its individual components can be realized.

[0128] Performing a Fourier transform on the soybean canopy multispectral image means transforming the soybean canopy multispectral image from the time domain space to the frequency domain space, so as to utilize the Fourier spectrum characteristics for processing and analysis. Since the image is a non-periodic discrete signal, the discrete Fourier transform method time extraction radix-2 FFT calculation method is used to perform the Fourier transform on the soybean multispectral image, and programming is used to perform the Fourier transform on the soybean multispectral image.

[0129] The process of implementing the Fourier transform of the soybean canopy multispectral image through programming is as follows: First, since the time domain axis extraction radix-2 FFT algorithm is used, this algorithm requires the size of the image to be a multiple of 2 × a multiple of 2. Therefore, the size of the soybean canopy multispectral image matrix used is 128 pixels × 128 pixels; then, reverse the order of each row of the image matrix and perform the butterfly operation; finally, reverse the order of each column of the matrix obtained in the previous step and perform the butterfly operation, and finally obtain the result of the Fourier transform of the soybean multispectral image.

[0130] (2) Spectral spectrum of soybean canopy multispectral image

[0131] After performing Fourier transforms on the near-infrared spectral reflection image, green light spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of the soybean canopy respectively, amplitude spectra and energy spectra can be obtained. The amplitude spectrum calculation formula is as follows:

[0132]

[0133] The energy spectrum calculation formula is as follows:

[0134]

[0135] In the formula, R(u, v) ， I(u, v) respectively represent the real part and the imaginary part of F(u, v), and F(u, v) represents the frequency spectrum of the image f(x, y).

[0136] The spectral characteristics of the soybean canopy multispectral image after Fourier transform are distributed in the four corners of the image. To analyze the image more conveniently, using the translation property of the Fourier transform and the properties of exponents, according to Equation (18), the origin of the amplitude spectrum is translated to the middle of the image. In the shifted-frequency amplitude spectrum, its center point is called the DC component, which reflects the average brightness of the original image. Different points at the same distance from the center point have the same frequency and different directions. The closer to the center point, the lower the frequency; the farther from the center point, the higher the frequency.

[0137]

[0138] In the formula, represents the Fourier transform, f(x, y) represents an image of size M×N, represents translating the origin of the frequency spectrum F(u, v) from (0, 0) to the center point under the frequency coordinates

[0139] Performing Fourier transform analysis on the four spectral reflection images of the soybean canopy multispectral image, obtaining the Fourier transform amplitude spectrum, and then translating the frequency spectrum to the origin of the image according to formula (18), obtaining the shifted-frequency amplitude spectrum and phase spectrum, as Figure 8 shown.

[0140] Figure 8 (a1~d1) are the images of the four spectral reflection images of the soybean canopy multispectral image, Figure 8 (a2~d2) are the amplitude spectra of the corresponding soybean canopy multispectral images (a1~d1) after Fourier transform, Figure 8 (a4~d4) represent the shifted-frequency amplitude spectra of the soybean canopy multispectral images (a1~d1) after Fourier transform. From Figure 8It can be seen from (a2~d2) and (a4~d4) that the amplitude spectrum shows the frequency distribution state of the soybean multispectral image. After the Fourier transform and frequency shift of the soybean image, the low-frequency components change from being distributed at the four corners of the image to being distributed in the middle of the image. Usually, the amplitude spectrum determines the number of frequency components contained in an image. Through analysis, Figure 8 (a4) The total number of frequency components contained in the amplitude spectrum of the green light spectral reflection image of soybeans is 5,646,242. Figure 8 (b4) The total number of frequency components contained in the amplitude spectrum of the near-infrared spectral reflection image is 5,012,815. Figure 8 (c4) The total number of frequency components contained in the amplitude spectrum of the red light spectral reflection image is 3,771,802. Figure 8 (d4) The total number of frequency components contained in the amplitude spectrum of the red-edge spectral reflection image is 22,279,084.

[0141] Using the amplitude spectrum diagram obtained in the previous step, the frequency spectrum components can be further calculated, and the distribution status and energy information of each frequency spectrum component in the amplitude spectrum can be observed more clearly visually.

[0142] As can be seen from formula (17), the energy spectrum is defined as the square of the amplitude spectrum. The energy spectrum characterizes the energy characteristics of the image and describes the distribution of signal energy in each frequency range. Usually, the image signal is generally band-limited, and the energy spectrum of the image decreases rapidly with the increase of frequency, resulting in a large difference in the energy between the low-frequency part and the high-frequency part. In order to clearly display the low-frequency energy and high-frequency energy at the same time, the energy spectrum after Fourier transform is subjected to coordinate shift and logarithmic processing here. The Fourier transform is performed on the canopy reflection images of soybeans under normal treatment and drought treatment respectively, and the zero-order spectrum coefficient is located at the center of the two-dimensional spectrum coefficient square matrix to obtain the energy spectrum reflecting the spectrum energy information as Figure 9 shown.

[0143] Figure 9 (a1~d1) are the canopy multispectral images of soybeans under normal treatment. Figure 9 (a2~d2) represent the energy spectra of the canopy multispectral images of soybeans under normal treatment (a1~d1). Figure 9 (a3~d3) are the canopy multispectral images of soybeans under drought treatment. Figure 9 (a4~d4) represent the energy spectra of the canopy multispectral images of soybeans under drought treatment (a3~d3). From Figure 9It can be intuitively seen from (a2~d2) and (a4~d4) that a series of regions with different colors appear on the energy spectrum diagram. The differences in these colors indicate the spectral energy distribution of the soybean multispectral image. Where the spectral energy is large, the color at that point tends to be bright, and vice versa. Thus, it can be concluded that most of the spectral energy of the soybean canopy multispectral image is concentrated near the low-frequency components of the energy spectrum diagram, and the energy diffuses from the center to the surrounding areas. The energy of the high-frequency components is relatively weak and is distributed along the edges of the energy spectrum diagram. Calculate the spectral energy of the soybean canopy multispectral images under normal treatment and drought treatment respectively according to formulas (19~21), and analyze Figure 9 the energy spectrum diagram of (a2~d2) under normal treatment and Figure 9 the distribution status and characteristics of the spectral energy on the energy spectrum diagram of (a4~d4) under drought treatment.

[0144] On the energy spectrum diagram, a series of concentric circles are drawn with the spectral center as the center and R as the radius. The percentage of the spectral energy contained within circles of different radii in the total spectral energy is used to characterize the distribution status of the corresponding spectral energy

[27] , and the calculation formula is as follows:

[0145]

[0146] β = 100×(E R / E T ) (21)

[0147] In the formula, u and v represent frequency components, M and N represent the image size, |F(u, v)| represents the spectrum, and |F(u, v)| 2 represents the energy spectrum.

[0148] Analyze and calculate the spectral energy of the soybean green light spectral reflection images under normal treatment and drought treatment at different spectral radii according to formulas (19~21), as shown in Table 4.

[0149] Table 4 Distribution Table of Soybean Spectral Energy in Green Light Spectral Reflection Images

[0150]

[0151] It can be seen from Table 4 that when the spectral radius is 15, the energy of the soybeans under normal treatment in the green light spectral reflection image reaches 90% of the total spectral energy. When the radius is 50, the energy reaches 99%. For the soybeans under drought treatment, when the spectral radius is 15, the energy only reaches 78% of the total spectral energy, and when the radius is 55, the energy reaches 99%. Therefore, from Figure 8 the energy spectra of the green light spectral reflection images of (a2) and (a4), it can be seen that the yellow areas of the soybeans under drought treatment are more widely distributed than those of the soybeans under normal treatment.

[0152] Similarly, according to formulas (19 - 21), the spectral energies of the near-infrared spectral reflection images of soybeans under normal treatment and drought treatment at different spectral radii were analyzed and calculated, as shown in Table 5.

[0153] Table 5 Distribution Table of Spectral Energy of Soybean in Near-Infrared Spectral Reflection Image

[0154]

[0155] As can be seen from Table 5, when the spectral radius is 15, the energy of soybeans in the near-infrared spectral reflection image under normal treatment reaches 91% of the total spectral energy. When the radius is 50, the energy reaches 99%. For the soybeans under drought treatment, when the spectral radius is 15, the energy only reaches 82% of the total spectral energy, and when the radius is 55, the energy reaches 99%. Therefore, from Figure 8 the energy spectra of the (b2) and (b4) near-infrared spectral reflection images, the yellow area of the soybeans under drought treatment is more widely distributed than that of the soybeans under normal treatment.

[0156] Similarly, according to formulas (19 - 21), the spectral energies of the red-light spectral reflection images of soybeans under normal treatment and drought treatment at different spectral radii were analyzed and calculated, as shown in Table 6.

[0157] Table 6 Distribution Table of Spectral Energy of Soybean in Red-Light Spectral Reflection Image

[0158]

[0159] As can be seen from Table 6, when the spectral radius is 35, the energy of soybeans in the red-light spectral reflection image under normal treatment reaches 89% of the total spectral energy. When the radius is 50, the energy reaches 98%. For the soybeans under drought treatment, when the spectral radius is 35, the energy reaches 90% of the total spectral energy, and when the radius is 50, the energy also reaches 98%. Therefore, from Figure 8 the energy spectra of the (c2) and (c4) red-light spectral reflection images, the distribution range of the yellow area of the soybeans under drought treatment is basically the same as that of the soybeans under normal treatment.

[0160] Similarly, according to formulas (19 - 21), the spectral energies of the red-edge spectral reflection images of soybeans under normal treatment and drought treatment at different spectral radii were analyzed and calculated, as shown in Table 7.

[0161] Table 7 Distribution Table of Spectral Energy of Soybean in Red-Edge Spectral Reflection Image

[0162]

[0163] As can be seen from Table 7, when the spectral radius is 15, the energy of soybeans with normal processing of the red-edge spectral reflection image reaches 90% of the total spectral energy. When the radius is 47, the energy reaches 99%. For soybeans under drought treatment, when the spectral radius is 15, the energy only reaches 81% of the total spectral energy, and when the radius is 50, the energy reaches 99%. Therefore, from Figure 8 the energy spectra of the (d2) and (d4) red-edge spectral reflection images, the yellow area of soybeans under drought treatment is more widely distributed than that of soybeans with normal processing.

[0164] By analyzing the spectral characteristics of the multi-spectral images of the soybean canopy, it can be seen that on the energy spectra of the normal processing and drought treatment, most of the spectral energy is concentrated in the low-frequency region at the center of the spectrum, that is, most of the energy after Fourier transform is concentrated on the low-order spectral coefficients. And at the same spectral radius, except for the red-light spectral reflection image, the percentage of the spectral energy of the multi-spectral image of the soybean canopy under drought treatment in the total energy is less than that of the normal processing. Therefore, based on this spectral characteristic, the present application constructs a soybean body wilt index based on Fourier transform to calculate the wilt degree.

[0165] (3) Construct the soybean body wilt index

[0166] The mathematical meaning of the Fourier spectrum method is that when the shape of the soybean leaf changes from stretched to wilted, it indicates that the slopes of each point on the leaf surface have changed to varying degrees. From the Fourier spectrum diagram, its low-frequency spectrum will decrease and the high-frequency spectrum will increase. For a planar graph, the characteristics of the image after Fourier transform are that the amplitudes corresponding to most frequencies are very small, but there is suddenly a highest point at the zero-frequency point, that is, the amplitude of the DC component is the largest. This point value represents the number of points with zero frequency. When the soybean is not wilted, its leaf surface can be approximately regarded as a plane, and at this time, the number of points with zero frequency is the largest, and the DC component of the two-dimensional Fourier spectrum is the largest. When the leaf wilts, the curvature of the leaf increases, the number of points with zero frequency decreases, and the value of the DC component decreases. Therefore, the wilt degree of soybeans can be investigated by the magnitude of the DC component after Fourier transform.

[0167] The data on the soybean canopy surface is a non-periodic discrete signal. Applying Fourier transform for calculation to obtain spectral information, the DC component in the spectrum represents the planar component of the surface, and the high-order harmonic components represent the surface components with different bending degrees on the surface. The proportion of the DC component in the spectrum can quantitatively characterize the wilt degree of plant leaves. Accordingly, the present application defines the soybean wilt index L based on the Fourier transform energy spectrum as

[0168]

[0169] In the formula, ln(·) represents the natural logarithmic function loge(·), |F(0,0)|2 represents the energy value of the DC component in the energy spectrum, represents the total energy value of the frequency spectrum in the energy spectrum.

[0170] Step Five:

[0171] During specific detection, the near-infrared spectral reflection image, green-light spectral reflection image, red-light spectral reflection image, and red-edge spectral reflection image of the soybean canopy are respectively input into Equation (22) to obtain the soybean wilting index L detected by the four spectra, and the average wilting index L is calculated 平 , when the average wilting index is 24 ≤ L 平 < 30, it is in a normal growth state, and when the average wilting index is 30 ≤ L 平 ≤ 36, it is in a drought growth state.

[0172] Next, the above method is used to calculate the wilting index of soybeans in the V4 and V5 stages:

[0173] Perform Fourier transform analysis on the multi-spectral images of the four spectral reflection images of the soybean canopy of the V4-stage soybeans under normal treatment and drought treatment. The soybean canopy and the corresponding frequency spectrum diagrams are as Figure 10 shown.

[0174] Meanwhile, find out Figure 10 the DC components of the frequency spectrum diagrams of the four spectral reflection images of the soybean multi-spectral image in , and calculate the wilting index of the four spectral reflection images of the soybean canopy according to Equation (22), as shown in Table 8.

[0175] Table 8 Wilting Index of V4-Stage Soybean Multi-Spectral Images

[0176]

[0177] It can be concluded from Table 8 that after performing Fourier transforms on the canopies of the four spectral reflectance images of the soybean multispectral images respectively, the DC component of the green light spectral reflectance image under normal processing is 14261113, the sum of all harmonics is 1411661941, and the wilting index is 24.51. The DC component of the drought-treated one is 3298957, the sum of all harmonics is 973584572, and the wilting index is 30.01. Since the wilting of the soybean canopy under drought treatment is more severe, the DC component value is on the low side, and the sum of all harmonics also decreases accordingly. The wilting index increases by 5.50 compared with that of the normally treated soybeans. The DC component of the near-infrared spectral reflectance image under normal processing is 11061788, the sum of all harmonics is 1225012265, and the wilting index is 25.10. The DC component of the drought-treated one is 2483014, the sum of all harmonics is 807574334, and the wilting index is 31.40. The wilting of the soybean canopy under drought treatment is more severe, the DC component value is on the low side, and the sum of all frequencies decreases accordingly. The wilting index increases by 6.30 compared with that of the normally treated soybeans in the unfolded state. The DC component of the red light spectral reflectance image under normal processing is 2335836, the sum of all harmonics is 850051803, and the wilting index is 38.70. The DC component of the drought-treated one is 1788030, the sum of all harmonics is 787908086, and the wilting index is 46.33. Since the wilting of the soybean canopy under drought treatment is more severe, the DC component value is on the low side, and the sum of all harmonics also decreases accordingly. The wilting index increases by 7.63 compared with that of the normally treated soybeans. The DC component of the red edge spectral reflectance image under normal processing is 11595933, the sum of all harmonics is 1360527797, and the wilting index is 25.66. The DC component of the drought-treated one is 1993135, the sum of all harmonics is 744932713, and the wilting index is 31.50. The wilting of the soybean canopy under drought treatment is more severe, the DC component value is on the low side, and the sum of all frequencies decreases accordingly. The wilting index increases by 5.84 compared with that of the normally treated soybeans in the unfolded state. Through analysis, the wilting indices of the green light, near-infrared, red light, and red edge spectral reflectance images of soybeans can all reflect the wilting state of soybeans.

[0178] Similarly, Fourier transform analysis was performed on the four spectral reflectance images of the soybean canopies under normal and drought treatments at the V5 stage. The soybean canopy and the corresponding frequency spectra are as Figure 11 shown.

[0179] Meanwhile, the DC components of the frequency spectra of the four spectral reflectance images of the soybean multispectral images in Figure 11 were extracted, and the wilting indices of the frequency spectra of the four spectral reflectance images of soybeans in Figure 11 were calculated according to formula (22), as shown in Table 9.

[0180] Table 9 Wilting indices of soybean multispectral images at the V5 stage

[0181]

[0182] As can be seen from Table 9, after performing Fourier transforms on the canopies of the four spectral reflectance images of soybean multispectral images respectively, the DC component of the green light spectral reflectance image under normal processing is 7,289,586, the sum of all harmonics is 1,029,379,324, and the wilting index is 21.11. The DC component of the drought treatment is 4,376,578, the sum of all harmonics is 1,047,865,766, and the wilting index is 32.35. Since the soybean canopy under drought treatment is more wilted, the DC component value is lower, and the sum of all harmonics also decreases accordingly. The wilting index has increased by 11.24 compared to the soybeans under normal processing. The DC component of the near-infrared spectral reflectance image under normal processing is 5,042,122, the sum of all harmonics is 825,852,108, and the wilting index is 22.16. The DC component of the drought treatment is 3,106,403, the sum of all harmonics is 842,947,054, and the wilting index is 33.46. The soybean canopy under drought treatment is more wilted, the DC component value is lower, and the sum of all frequencies decreases accordingly. The wilting index has increased by 11.30 compared to the soybeans in the unfolded state under normal processing. The DC component of the red light spectral reflectance image under normal processing is 1,019,769, the sum of all harmonics is 513,051,543, and the wilting index is 34.77. The DC component of the drought treatment is 504,395, the sum of all harmonics is 455,920,675, and the wilting index is 37.07. Since the color of the soybean canopy under normal processing is too close to the color of the background area, some parts of the canopy area are missing after binarization, resulting in a lower DC component value and a higher wilting index after Fourier transform. The extraction effect of the soybean canopy under drought treatment is relatively ideal. Although the soybean canopy is more wilted, the canopy area is relatively complete after binarization, and the wilting index only differs by 2.30 from that under normal processing. The DC component of the red-edge spectral reflectance image under normal processing is 5,718,096, the sum of all harmonics is 906,497,754, and the wilting index is 22.70. The DC component of the drought treatment is 3,317,622, the sum of all harmonics is 908,765,001, and the wilting index is 35.09. Since the soybean canopy under drought treatment is more wilted, the DC component value is lower, and the sum of all harmonics also decreases accordingly. The wilting index has increased by 12.39 compared to the soybeans in the unfolded state under normal processing. Through analysis, the wilting indices of the green light, near-infrared, red light, and red-edge spectral reflectance images of soybeans can all reflect the wilting state of soybeans.

[0183] The leaf inclination angle refers to the angle between the normal of the ventral surface of the leaf and the main stem, and it is also the angle between the leaf surface and the ground plane. There is a significant correlation between the leaf inclination angle of the plant and the soil water content. Therefore, by analyzing the correlation between the change in the leaf inclination angle of soybeans and the wilting index, it can be known that when the correlation determination coefficient R of the wilting degree of the soybean canopy and the leaf inclination angle 2(Calculated according to the least squares principle) are all above 0.85 and can be used as quantitative indicators for the variation laws of ecological and morphological traits of soybean plants under drought stress. In this application, the wilting index of soybean in four spectral reflectance images of normal treatment and drought treatment at V4 and V5 stages was respectively analyzed for correlation with the average leaf inclination angle at the corresponding stage to verify the effectiveness of the wilting index calculation method of this application. The results are shown in Table 10.

[0184] Table 10 Correlation analysis of measured values of soybean wilting index and average leaf inclination angle

[0185]

[0186] As can be seen from Table 10, in the normal treatment at V4 stage, the determination coefficients R2 of the wilting index of the green light, near-infrared, red light, and red edge spectral reflectance images of soybean and the measured values of leaf inclination angle reached 0.88, 0.92, 0.90, and 0.94 respectively. The maximum value was 0.94 for the red edge spectral reflectance image, the minimum value was 0.88 for the green light spectral reflectance image, and the average value was 0.91. In the drought treatment, the determination coefficients R2 of the wilting index of the green light, near-infrared, red light, and red edge spectral reflectance images of soybean and the leaf inclination angle reached 0.91, 0.92, 0.85, and 0.86 respectively. The maximum value was 0.92 for the near-infrared spectral reflectance image, the minimum value was 0.85 for the red light spectral reflectance image, and the average value was 0.89. In the normal treatment at V5 stage, the determination coefficients R2 of the wilting index of the green light, near-infrared, red light, and red edge spectral reflectance images of soybean and the leaf inclination angle reached 0.86, 0.85, 0.86, and 0.87 respectively. The maximum value was 0.87 for the red edge spectral reflectance image, the minimum value was 0.85 for the near-infrared spectral reflectance image, and the average value was 0.86. In the drought treatment, the determination coefficients R2 of the wilting index of the green light, near-infrared, red light, and red edge spectral reflectance images of soybean and the measured values of leaf inclination angle reached 0.88, 0.89, 0.95, and 0.93 respectively. The maximum value was 0.95 for the red light spectral reflectance image, the minimum value was 0.88 for the green light spectral reflectance image, and the average value was 0.91. The results show that the soybean wilting index defined in this application has a high correlation with the leaf inclination angle and can be used to calculate the wilting degree of soybean, thereby reflecting the water shortage situation of soybean.

Claims

1. A method for calculating the wilting degree of soybean canopy using multi-spectral image Fourier transform, characterized in that Achieved through the following steps: (1) Obtain the multispectral images of soybeans using a Parrot Sequoia multispectral camera. The multispectral images are respectively a green light spectral image, a red light spectral image, a red edge spectral image, and a near-infrared spectral image. According to the Sequoia calibration equation, use the empirical linear method to calculate the reflection images of the original multispectral images of soybeans, and respectively obtain the green light spectral reflection image, near-infrared spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of soybeans, and calculate the corresponding normalized values; (2) Use the median filtering method to perform noise reduction processing on the green light spectral reflection image, near-infrared spectral reflection image, red light spectral reflection image, and red edge spectral reflection image obtained in step (1); (3) Use the iterative threshold method to obtain the canopy of the near-infrared spectral reflection image of soybeans for the images processed in step (2). Use the near-infrared canopy image of soybeans as a template for identifying the green light spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of soybeans. Apply the affine transformation calculation method to construct the affine transformation model of the target area between the reference image and the green light spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of soybeans, so as to extract the target area of the canopy of the green light spectral reflection image, red light spectral reflection image, and red edge spectral reflection image of soybeans, perform gray level histogram statistics on it, and obtain and analyze the canopy characteristics; (4)Perform Fourier transform on the canopy of the near-infrared spectral reflection image, green light spectral reflection image, red light spectral reflection image, and red-edge spectral reflection image of soybeans obtained in step (3), conduct spectral analysis on the canopy to obtain an amplitude spectrum diagram, and then obtain an energy spectrum diagram. Extract the DC component in the energy spectrum diagram and calculate the proportion of the DC component in the total energy of the spectrum to obtain the wilting index of the soybean canopy. The calculation formula is as follows: ， In the formula, represents the natural logarithm function , represents the energy value of the DC component in the energy spectrum, represents the total energy value of the frequency spectrum in the energy spectrum, represents the frequency component, represents the image size, represents the frequency spectrum, represents the energy spectrum.

2. The method for calculating the wilting degree of soybean canopy using multi-spectral image Fourier transform according to claim 1, characterized in that The Fourier transform calculation method in step (4) is the time extraction radix-2 FFT calculation method.

Images