Field crop leaf area index monitoring method and system based on machine vision

Abstract

The invention discloses a field crop leaf area index monitoring method and system based on machine vision, and the method is based on a hemispherical canopy image collected by a low-cost fisheye lens, and preferably selects a crop and soil or sky background classification color index to adapt to different crop growth periods and illumination conditions. A supervised classification result is used as a real classification result of crop canopy and background pixels, and color indexes suitable for different key growth periods and different illumination intensities are preferably selected for the two types of hemispherical canopy images; adopting an EM algorithm and a two-class Gaussian mixture model GMM to perform classification binarization extraction of crop pixels and background pixels; calculating the crop canopy clearance rate; and finally, solving the aggregation distribution effect of actual crop leaves by applying Poisson distribution hypothesis, and calculating a crop leaf area index at a hemispherical canopy image acquisition site and moment based on the crop canopy clearance rates in different zenith rings, thereby providing support for precise field management. And the method has the most important significance for improving the field precise management level.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural remote sensing information technology, and particularly relates to a method and system for monitoring the leaf area index of field crops based on machine vision. Background Art

[0002] The leaf area index (LAI) is defined as half of the cumulative value of the surface areas of all leaves of crops per unit area, and is a key indicator for characterizing the growth status of crops. It is of great significance for monitoring the growth of crops, simulating crop growth, and guiding precise fertilization and irrigation management in the field. Traditionally, direct measurement methods are mainly used to monitor the leaf area index of field crops, that is, destructive sampling is used to sample crop plants, and the surface areas of all leaves of the plants are measured. This method has the advantage of high measurement accuracy, but due to its destructive nature, time-consuming and laborious, and difficult to monitor online at fixed points, it is difficult to meet the precise monitoring requirements for the growth status of field crops under the background of current smart agriculture.

[0003] With the development of remote sensing technology and sensor technology, researchers have proposed an indirect measurement method for the crop leaf area index, that is, a camera equipped with a fish-eye lens is used to collect hemispherical canopy images from below the crop canopy upward or from above the crop canopy downward, and various image processing means are used to extract the gap ratios of the multi-angle crop canopy, and then the leaf area index is inverted using the canopy gap ratio. This method can also achieve precise monitoring of the crop leaf area index, and has the advantages of non-destructiveness, time-saving and labor-saving, easy automation, etc., and is more likely to meet the precise monitoring requirements for the growth status of field crops under the background of current smart agriculture, and can provide support for precise fertilization and irrigation management in the field.

[0004] Currently, there are some imported high-end instruments on the market that can monitor the crop leaf area index. For example, the LAI-2200 series plant canopy analyzer developed by LI-COR Company in the United States is the currently recognized gold standard for non-destructive measurement in the academic community and is the first choice instrument for researchers to monitor the crop leaf area index. However, this instrument cannot view the information of the measured ground objects and the original canopy image information, is expensive and cannot achieve fixed-point automatic online monitoring of the crop leaf area index.

[0005] In summary, in order to achieve low-cost, fixed-point online precise monitoring of the crop leaf area index under the background of current smart agriculture, it is urgent to develop a method and system for monitoring the leaf area index based on the crop canopy gap ratio, so as to provide technical and equipment support for precise fertilization and irrigation management in the field. Summary of the Invention

[0006] Objective of the Invention: The objective of the present invention is to solve the problems that the traditional monitoring methods for the leaf area index of field crops are time-consuming, laborious, costly, and difficult to meet the requirements of automated monitoring of the growth of field crops. A monitoring method and system for the leaf area index of field crops based on the canopy gap fraction are developed to provide technical and equipment support for precise fertilization, irrigation management, etc. in the field under the background of smart agriculture.

[0007] To meet the various objectives of the present application, the following technical solutions are adopted:

[0008] The present invention first provides a monitoring method for the leaf area index of field crops based on machine vision. The method includes the following steps:

[0009] S1: For the fixed-point online monitoring requirements of the leaf area index of field crops, at different growth stages and different light intensities, two types of hemispherical canopy images are collected using a camera equipped with a fish-eye lens from below the crop canopy upward and from above the crop canopy downward.

[0010] S2. Using the supervised classification method for typical monitoring scenarios of the leaf area index of field crops, the distributions of crop pixels, soil background pixels, or sky background pixels in the two types of hemispherical canopy images collected in step S1 are extracted, and used as the true value verification data set for the classification of crop and soil background pixels or crop and sky background pixels.

[0011] S3. The two types of hemispherical canopy images collected in step S1 are respectively converted from the RGB color space to the HSV and Lab color spaces, and then 12 common color indices including the red R, green G, blue B, hue H, green-to-red a, yellow-to-green b channel indices, as well as the excess green index ExG, green-blue difference vegetation index GBDI, modified excess green vegetation index MExG, visible atmospheric impedance index VARI, red-green ratio vegetation index RGRI, and excess green and red difference index ExGR are respectively selected in the RGB, HSV, and Lab color spaces.

[0012] S4. Based on step S3, it is assumed that in the hemispherical canopy image collected from below the crop canopy upward, the crop and sky background pixels both show a Gaussian distribution, that is, the crop and sky background pixels in this hemispherical canopy image can be described by a two-class Gaussian mixture model GMM; similarly, in the image from above the crop canopy downward, the canopy and soil background pixels can also be described by a two-class Gaussian mixture model GMM.

[0013] S5. Based on steps S2 and S4, a confusion matrix is used to evaluate the classification accuracy; that is, taking the supervised classification result as the true classification result of crop canopy and background pixels, the accuracy, precision, recall, and F1-score are used to evaluate the automatic classification accuracy of crop and background pixels under each color index using the EM algorithm and the two-class Gaussian mixture model GMM. Furthermore, color indices suitable for different critical growth stages and different light intensities are preferably selected for the two types of hemispherical canopy images respectively;

[0014] S6. Based on steps S2, S4, and S5, for the two types of hemispherical canopy images respectively, classification binary extraction of crop pixels and background pixels is carried out using the EM algorithm and the two-class Gaussian mixture model GMM based on the optimal color index;

[0015] S7. The zenith angle in the range of 0° - 75° is equally divided into 5 parts, each zenith ring is 15°, and the crop canopy gap fraction is calculated;

[0016] S8. Based on step S7, the Poisson distribution hypothesis is applied to solve the aggregation distribution effect of actual crop leaves. Furthermore, the crop leaf area index at the acquisition location and time of the hemispherical canopy image is calculated based on the crop canopy gap fraction within different zenith rings.

[0017] Furthermore, the two-class Gaussian mixture model GMM described in step S4 is as shown in formula (1):

[0018]

[0019] In the formula, ω, μ, and σ respectively represent weight, mean, and standard deviation, and the subscripts v and b respectively represent crop canopy and soil or sky background, and N(μ, σ) represents the Gaussian distribution function.

[0020] Furthermore, for the evaluation of classification accuracy using the confusion matrix in step S5, the specific calculation methods of each accuracy evaluation index are as follows:

[0021]

[0022] In the formula, TP is the number of positive classes correctly predicted by the model as positive classes, FP is the number of negative classes wrongly predicted by the model as positive classes, TN is the number of negative classes correctly predicted by the model as negative classes, and FN is the number of positive classes wrongly predicted by the model as negative classes.

[0023] Further, the specific method of step S5 is as follows: The Expectation-Maximization algorithm, i.e., the EM algorithm, is used to classify the crop and background pixels in the two types of hemispherical canopy images based on the Gaussian mixture model. First, the hemispherical canopy image is converted into the corresponding color index, and on this basis, random initial values are set for the weight ω, mean μ, and standard deviation σ in the two Gaussian mixture models GMM. Secondly, the posterior probability is calculated based on the current initial values of ω, μ, and σ, i.e., the E-step is performed. Then, ω, μ, and σ are recalculated based on the posterior probability calculated in the E-step (M-step). Then, the log-likelihood function is calculated to determine whether convergence has occurred. If not, the above steps are repeated. If convergence occurs, the two Gaussian distributions are defined with the updated ω, μ, and σ, i.e., the specific distributions of the crop canopy pixels and the soil or sky background pixels are described. Finally, based on this, the intersection of the two Gaussian distributions is used as the classification threshold to extract the crop pixel distribution, i.e., the crop canopy pixels are assigned a value of 1, and the soil or sky background pixels are assigned a value of 0 to obtain a binary classification image.

[0024] Further, the calculation of the crop canopy gap fraction in step S7 is specifically calculated as follows:

[0025]

[0026] where p0(θ) is the canopy gap fraction at the zenith angle θ, N s (θ) is the number of sky or soil background pixels at the zenith angle θ, N ns (θ) is the number of crop pixels at the zenith angle θ.

[0027] Further, the calculation of the crop leaf area index at the acquisition location and time of the hemispherical canopy image based on the crop canopy gap fraction within different zenith rings in step S8 is specifically calculated as follows:

[0028]

[0029] where p0(θ i ) is the crop canopy gap fraction of the i-th zenith ring, cosθ i is the optical path length of the light passing through the canopy under the i-th zenith ring, sinθ i dθ i is the weight factor and satisfies

[0030] The present invention also provides a machine vision-based monitoring system for the leaf area index of field crops. This system is used to run the above method. The system includes a central processing unit, a fish-eye camera, a fish-eye camera position and attitude adjustment module, a communication module, an instruction input module, a processing result display module, and a data storage module;

[0031] The fish-eye camera collects two types of hemispherical canopy images from below and above the crop canopy through the fish-eye camera position and attitude adjustment module, and uses the camera serial interface to transmit image data, control signals, and clock signals to the central processor;

[0032] The communication module receives the hemispherical canopy images and the processing results of the crop leaf area index from the central processor, and remotely transmits data such as hemispherical canopy images and crop leaf area index through the network, and receives control instructions from the remote control platform, including leaf area index acquisition, image transmission, and query of the intermediate processing process;

[0033] The instruction input and processing result display module uses a touch display panel to receive control instructions for on-site leaf area index acquisition, image transmission, and query of the intermediate processing process, and displays the corresponding data processing results;

[0034] The data storage module uses an SD module to store hemispherical canopy images, their intermediate processing processes, and leaf area index data, and conducts data interaction with each module through the central processing module;

[0035] The central processor adopts an arm architecture, embeds a linux operating system, and is equipped with CSI, a display serial interface, and a microSD interface, and is used to implement the leaf area index of field crops based on the crop canopy gap ratio and data interaction with each module.

[0036] Furthermore, the fish-eye camera position and attitude adjustment module includes a main body support structure, a vertical adjustment structure for the fish-eye camera support cross-arm, and a vertical flipping structure for the fish-eye camera;

[0037] The main body support structure includes a support frame and a main rod. The main rod is vertically fixed on the support frame and is used to place the main control box and the lead screw assembly; the overall combination of the fish-eye camera support cross-arm, the flipping stepping motor, and the fish-eye camera is connected to the lead screw assembly through a nut;

[0038] The vertical adjustment structure for the fish-eye camera support cross-arm includes a stepping motor. By controlling the forward and reverse rotation of the lead screw through the stepping motor, the overall combination of the fish-eye camera support cross-arm, the flipping stepping motor, and the fish-eye camera is driven to move up and down, realizing precise movement in the vertical direction;

[0039] The up-and-down flipping structure of the fisheye camera includes a flipping stepper motor. The flipping stepper motor is connected to the fisheye camera through a fisheye camera support cross arm. By driving the fisheye camera support cross arm connected to the fisheye lens to flip around the horizontal axis by the flipping stepper motor, the fisheye camera can collect two types of hemispherical half-layer images from below the crop canopy upwards and from above the crop canopy downwards. That is, to collect the hemispherical canopy image from below the crop canopy upwards, the stepper motor controls the fisheye camera support cross arm to move downward on the lead screw assembly until it is below the crop canopy and close to the ground. Then, through the up-and-down flipping structure of the fisheye camera, it receives the control signal from the central processing unit and drives the flipping stepper motor to flip upwards, finally realizing the collection of the hemispherical canopy image by the fisheye camera from below the crop canopy upwards. To collect the hemispherical canopy image from above the crop canopy downwards, the stepper motor controls the fisheye camera support cross arm to move upwards on the lead screw assembly until it is above the crop canopy. Then, through the up-and-down flipping structure of the fisheye camera, it receives the control signal from the central processing unit and drives the flipping stepper motor to flip downwards, finally realizing the collection of the hemispherical canopy image by the fisheye camera from above the crop canopy downwards.

[0040] The beneficial effects of the present invention compared with the prior art are as follows:

[0041] (1) For the monitoring requirements of leaf area index of field crops under different growth stages and lighting conditions, this method optimizes the color index suitable for classifying crop pixels and soil background pixels or sky background pixels in two types of hemispherical canopy images, improving the applicability and robustness of the actual monitoring of leaf area index of field crops for different application scenarios.

[0042] (2) Considering that in the actual canopy, the leaves do not show a random distribution but an aggregated distribution along the branches and stems, this method applies the Poisson distribution hypothesis to solve the aggregated distribution effect of actual crop leaves, improving the monitoring accuracy of the leaf area index of field crops.

[0043] (3) The present invention is a self-contained system and uses a low-cost fisheye camera and an embedded processor to realize the low-cost, accurate, and online monitoring of the leaf area index of field crops, solving the problems of time-consuming, laborious, high cost, and difficulty in meeting the demand for automated monitoring of the growth of field crops in traditional methods for monitoring the leaf area index of field crops, and having stronger practical application value in the precise field management of field crops. Description of the Drawings

[0044] Figure 1 It is a flow chart of a method for monitoring the leaf area index of field crops based on machine vision according to the present invention.

[0045] Figure 2 It is a schematic diagram of the zenith angle division of two types of hemispherical canopy images according to the present invention.

[0046] Figure 3Flowchart of the crop canopy extraction method based on the EM algorithm and Gaussian mixture model of the present invention.

[0047] Figure 4 Block diagram of a field crop leaf area index monitoring system based on machine vision of the present invention.

[0048] Figure 5 Structural diagram of a field crop leaf area index monitoring system based on machine vision of the present invention.

[0049] Figure 6 Structural diagram of the upper and lower adjustment of the cross arm of a field crop leaf area index monitoring system based on machine vision of the present invention.

[0050] Figure 7 Structural diagram of the lens flipping adjustment of a field crop leaf area index monitoring system based on machine vision of the present invention. Detailed implementation manners

[0051] As Figure 1 shown, a method for monitoring the leaf area index of field crops proposed to meet one of the purposes of this application is characterized in that a color index is preferably selected for the on-line automatic monitoring requirements of the leaf area index under different growth periods and lighting conditions of field crops, etc. Based on this, a crop pixel segmentation method is used to extract crop pixels and calculate the crop canopy gap ratio in the hemispherical canopy image, and then the automatic monitoring of the leaf area index of field crops based on the optimal color index of the hemispherical canopy image is realized. The specific steps are as follows:

[0052] Step 1: For the fixed-point on-line monitoring requirements of the leaf area index of field crops, at different key growth periods (such as the overwintering period, greening period, jointing period, heading period, flowering period, etc. of wheat) and under different light intensities (such as sunny, cloudy, overcast, etc.), a camera equipped with a fish-eye lens is used to collect two types of hemispherical canopy images from below the crop canopy upwards and from above the crop canopy downwards, as Figure 2 (a) shown.

[0053] Step 2: On the basis of Step 1, first, a supervised classification method is used for the typical field crop leaf area index monitoring scenario to extract the distribution of crop pixels, soil background pixels or sky background pixels in the two types of hemispherical canopy images, and this is used as the classification true value verification data set of crop and soil background pixels or crop and sky background pixels.

[0054] Step 3: Based on step 1 and step 2, the two types of hemispherical canopy images collected are converted from RGB color space to HSV and LAB color space respectively, and then 12 common color indices including R (red), G (green), B (blue), H (hue), green to red a, yellow to green b channel index, excess green index (ExG), green blue difference vegetation index (GBDI), modified excess green vegetation index (MExG), visible atmospheric resistive index (VARI), red green ratio vegetation index (RGRI), and excess green minus excess red (ExGR) are selected in RGB, HSV and LAB color spaces respectively.

[0055] Step 4: Based on step 3, it is assumed that in the hemispherical canopy image collected from below the crop canopy, the crop and sky background pixels are both Gaussian distributed, that is, the crop and sky background pixels in the hemispherical canopy image can be described by two types of Gaussian mixture models (GMM), as shown in formula (1); similarly, in the image collected from above the crop canopy, the canopy and soil background pixels can also be described by two types of Gaussian mixture models (GMM).

[0056]

[0057] Where ω, μ and σ represent weight, mean and standard deviation respectively, subscripts v and b represent crop canopy and soil or sky background respectively, and N(μ,σ) represents Gaussian distribution function.

[0058] Step 5: Based on Step 2 and Step 4, the confusion matrix is used to evaluate the classification accuracy; that is, the supervised classification results are taken as the true classification results of crop canopy and background pixels, and the accuracy (Accuracy), precision (Precision), recall (Recall) and F1 score (F1-score) are used to evaluate the automatic classification accuracy of crop and background pixels using the EM algorithm and GMM model under each color index, and then the color index suitable for different key growth periods and different light intensities is selected for the two types of hemispherical canopy images. The specific calculation method of each accuracy evaluation index is as follows:

[0059]

[0060]

[0061] Wherein, TP is the number of positive classes correctly predicted as positive classes by the model, FP is the number of negative classes incorrectly predicted as positive classes by the model, TN is the number of negative classes correctly predicted as negative classes by the model, and FN is the number of positive classes incorrectly predicted as negative classes by the model.

[0062] Step Six: Based on Steps Three, Four, and Five, for the two types of hemispherical canopy images respectively, using the optimal color index, the EM algorithm and the GMM model are adopted for the classification and binary extraction of crop pixels and background pixels, that is, as Figure 3 shown, this solution uses the Expectation-Maximum (EM) algorithm based on the Gaussian mixture model for the classification of crop and background pixels in the two types of hemispherical canopy images; first, the hemispherical canopy image is converted into the corresponding color index, and on this basis, random initial values are set for the weight ω, mean μ, and standard deviation σ in the GMM; secondly, the posterior probability is calculated according to the current initial values of ω, μ, and σ, that is, the E-step is performed; then, ω, μ, and σ are recalculated based on the posterior probability calculated in the E-step (M-step); then the log-likelihood function is calculated to determine whether it converges. If it does not converge, the above steps are repeated. If it converges, the two updated Gaussian mixture distributions are defined by the updated ω, μ, and σ, that is, the specific description of the crop canopy pixel distribution and the soil or sky background pixel is realized; finally, based on this, the intersection point of the two Gaussian distributions is used as the classification threshold to extract the crop pixel distribution, that is, the crop canopy pixels are assigned a value of 1, and the soil or sky background pixels are assigned a value of 0 to obtain the binary classification image, as Figure 2 (b) shown.

[0063] Step Seven: Based on Step Six, the zenith angles in the range of 0° - 75° are equally divided into 5 parts, and each zenith ring is 15°, as Figure 3 shown, and the crop canopy gap ratio is calculated. The specific calculation method is as follows:

[0064]

[0065] Wherein, p0(θ) is the canopy gap ratio at the zenith angle θ, N s (θ) is the number of sky or soil background pixels at the zenith angle θ, N ns (θ) is the number of crop pixels at the zenith angle θ.

[0066] Step 8: On the basis of Step 7, considering that in the actual canopy, the leaves do not show a random distribution but an aggregated distribution along branches and stems, this solution applies the Poisson distribution hypothesis to solve the aggregated distribution effect of actual crop leaves, and then calculates the crop leaf area index at the acquisition location and time of the hemispherical canopy image based on the crop canopy gap fraction in different zenith rings. The specific calculation method is as follows:

[0067]

[0068] In the formula, p0(θ i ) is the crop canopy gap fraction of the i-th zenith ring, cosθ i is the optical path length of the light passing through the canopy under the i-th zenith ring, sinθ i dθ i is the weight factor and satisfies

[0069] As Figure 4 shown, a monitoring system for the leaf area index of field crops proposed by the present invention mainly includes a central processor, a fish-eye camera, a fish-eye camera position and attitude adjustment module, a communication module, an instruction input module, a processing result display module, a data storage module, etc.; the fish-eye camera collects two types of hemispherical canopy images from below and above the crop canopy through its position and attitude adjustment module, and uses the Camera Serial Interface (CSI) to transmit image data, control signals, and clock signals to the central processor; the communication module receives the hemispherical canopy image and the processing result of the crop leaf area index from the central processor, and remotely transmits data such as the hemispherical canopy image and the crop leaf area index through the 4G network, and receives control instructions such as leaf area index acquisition, image transmission, and query of the intermediate process of processing from the remote control platform; the instruction input and processing result display module uses a touch display panel to receive control instructions such as on-site leaf area index acquisition, image transmission, and query of the intermediate process of processing, and displays the corresponding data processing results; the data storage module uses an SD module to store the hemispherical canopy image, its intermediate processing process, and leaf area index data, and performs data interaction with each module through the central processing module; the central processor uses an arm architecture, embeds a linux operating system, and comes with CSI, a Display Serial Interface (DSI), a microSD interface, etc., and can realize the leaf area index of field crops based on the crop canopy gap fraction and data interaction with each module.

[0070] As Figures 5 - 7As shown in the figure, a monitoring system for the leaf area index of field crops according to the present invention, the position and attitude adjustment module of its fisheye camera includes a main body support structure, an up-and-down adjustment structure of the fisheye camera support cross arm, and an up-and-down flipping structure of the fisheye camera;

[0071] The main body support structure includes a support frame 8 and a main rod 6. The main rod 6 is vertically fixed on the support frame 8 and is used to place the main control box and the lead screw assembly 5. The overall structure composed of the fisheye camera support cross arm 3, the flipping stepper motor 2, and the fisheye camera 4 is connected to the lead screw assembly 5 through a nut;

[0072] The up-and-down adjustment structure of the fisheye camera support cross arm includes a stepper motor 1. By controlling the forward and reverse rotation of the lead screw by the stepper motor 1, the overall structure composed of the fisheye camera support cross arm 3, the flipping stepper motor 2, and the fisheye camera 4 is driven to move up and down, realizing precise movement in the vertical direction;

[0073] The up-and-down flipping structure of the fisheye camera includes a flipping stepper motor 2. The flipping stepper motor 2 is connected to the fisheye camera 4 through the fisheye camera support cross arm 3. By driving the fisheye camera support cross arm 3 connected to the fisheye lens 4 to flip around the horizontal axis by the flipping stepper motor 2, the fisheye camera 4 can collect two types of hemispherical half-layer images from below the crop canopy upward and from above the crop canopy downward; that is, to realize the collection of hemispherical canopy images from below the crop canopy upward, the stepper motor 1 controls the fisheye camera support cross arm 3 to move downward on the lead screw assembly 6 until it is below the crop canopy and close to the ground, and then receives the control signal of the central processing unit through the up-and-down flipping structure of the fisheye camera, drives the flipping stepper motor 2 to flip upward, and finally realizes the collection of hemispherical canopy images by the fisheye camera 4 from below the crop canopy upward; to realize the collection of hemispherical canopy images from above the crop canopy downward, the stepper motor 1 controls the fisheye camera support cross arm 3 to move upward on the lead screw assembly until it is above the crop canopy, and then receives the control signal of the central processing unit through the up-and-down flipping structure of the fisheye camera, drives the flipping stepper motor 2 to flip downward, and finally realizes the collection of hemispherical canopy images by the fisheye camera from above the crop canopy downward.

[0074] Finally, based on the hemispherical canopy images collected by the low-cost fisheye lens, the classification color index of the crop and the soil or sky background is preferably selected to adapt to different crop growth periods and lighting conditions, and the leaf area index is accurately estimated considering the leaf aggregation effect on the basis of accurately obtaining the crop canopy gap rate, realizing the low-cost, accurate, and online monitoring of the leaf area index of field crops, which has high application value in field precise management.

Claims

1. A method for monitoring the leaf area index of field crops based on machine vision, characterized in that, The method includes the following steps: S1: For the on-site online monitoring requirements of the leaf area index of field crops, at different growth stages and different light intensities, a camera equipped with a fish-eye lens is used to collect two types of hemispherical canopy images from below the crop canopy upwards and from above the crop canopy downwards; S2. Using the supervised classification method for the monitoring scenario of the leaf area index of typical field crops, extract the distribution of crop pixels, soil background pixels or sky background pixels in the two types of hemispherical canopy images collected in step S1, and use this as the true classification verification dataset for the classification of crop and soil background pixels or crop and sky background pixels; S3. Convert the two types of hemispherical canopy images collected in step S1 from the RGB color space to the HSV and Lab color spaces respectively, and then select 12 commonly used color indices including the red R, green G, blue B, hue H, green-to-red a, yellow-to-green b channel indices, as well as the excess green index ExG, green-blue difference vegetation index GBDI, modified excess green vegetation index MExG, visible light atmospheric impedance index VARI, red-green ratio vegetation index RGRI, and excess green and red differential index ExGR in the RGB, HSV, and Lab color spaces respectively; S4. Based on step S3, assume that in the hemispherical canopy image collected from below the crop canopy upwards, the crop and sky background pixels both show a Gaussian distribution, that is, the crop and sky background pixels in this hemispherical canopy image can be described by two Gaussian mixture models GMM; similarly, in the image from above the crop canopy downwards, the canopy and soil background pixels can also be described by two Gaussian mixture models GMM; S5. Based on steps S2 and S4, use the confusion matrix to evaluate the classification accuracy; that is, take the supervised classification result as the true classification result of the crop canopy and background pixels, and use the accuracy, precision, recall, and F1-score to evaluate the automatic classification accuracy of the crop and background pixels using the EM algorithm and two Gaussian mixture models GMM under each color index, and then optimize the color indices applicable to different critical growth stages and different light intensities for the two types of hemispherical canopy images respectively; S6. Based on steps S2, S4, and S5, for the two types of hemispherical canopy images respectively based on the optimal color index, use the EM algorithm and two Gaussian mixture models GMM to perform binary extraction of crop pixels and background pixels; S7. Divide the zenith angle in the range of 0° - 75° into 5 equal parts, each zenith ring is 15°, and calculate the crop canopy gap ratio; S8. Based on step S7, apply the Poisson distribution hypothesis to solve the aggregation distribution effect of actual crop leaves, and then calculate the leaf area index of the hemispherical canopy image collection location and time based on the crop canopy gap ratio within different zenith rings.

2. The method for monitoring the leaf area index of field crops based on machine vision according to claim 1, wherein The two Gaussian mixture models GMM described in step S4 are shown in formula (1): In the formula, ω, μ, and σ respectively represent the weight, mean, and standard deviation, and the subscripts v and b respectively represent the crop canopy and soil or sky background, and N(μ,σ) represents the Gaussian distribution function.

3. The method for monitoring the leaf area index of field crops based on machine vision according to claim 1, characterized in that, As described in step S5, the classification accuracy is evaluated using a confusion matrix. The specific calculation methods for each accuracy evaluation index are as follows: In the formula, TP is the number of positive classes correctly predicted as positive by the model, FP is the number of negative classes incorrectly predicted as positive by the model, TN is the number of negative classes correctly predicted as negative by the model, and FN is the number of positive classes incorrectly predicted as negative by the model.

4. The method for monitoring the leaf area index of field crops based on machine vision according to claim 1, characterized in that, The specific method of step S5 is as follows: The Expectation-Maximization algorithm, i.e., the EM algorithm, is used to classify crop and background pixels in two types of hemispherical canopy images based on the Gaussian mixture model. First, the hemispherical canopy image is converted into the corresponding color index, and on this basis, random initial values are set for the weights ω, means μ, and standard deviations σ in the two Gaussian mixture models GMM. Secondly, the posterior probabilities are calculated based on the current initial values of ω, μ, and σ, i.e., the E-step is performed. Then, ω, μ, and σ are recalculated based on the posterior probabilities calculated in the E-step (M-step). Then, the log-likelihood function is calculated to determine whether convergence has occurred. If convergence has not occurred, the above steps are repeated. If convergence has occurred, the two Gaussian mixture distributions are defined with the updated ω, μ, and σ, i.e., the distribution of crop canopy pixels and the pixels of the soil or sky background are specifically described. Finally, based on this, the intersection point of the two Gaussian distributions is used as the classification threshold to extract the crop pixel distribution, i.e., the crop canopy pixels are assigned a value of 1, and the soil or sky background pixels are assigned a value of 0 to obtain a binary classification image.

5. A method for monitoring the leaf area index of field crops based on machine vision according to claim 1, characterized in that, As described in step S7, the calculation of the crop canopy gap fraction is as follows: where p0(θ) is the canopy gap fraction at zenith angle θ, N s (θ) is the number of sky or soil background pixels at zenith angle θ, N ns (θ) is the number of crop pixels at zenith angle θ.

6. The method for monitoring the leaf area index of field crops based on machine vision according to claim 1, wherein As described in step S8, the calculation of the crop leaf area index at the acquisition location and time of the hemispherical canopy image based on the crop canopy gap fraction within different zenith rings is as follows: where, p0(θ i ) is the crop canopy gap fraction of the i-th zenith ring, cosθ i is the optical path length of the light passing through the canopy under the i-th zenith ring, sinθ i dθ i is the weight factor and satisfies 7. A monitoring system for leaf area index of field crops based on machine vision, which is used to run the method described in any one of claims 1-6, characterized in that, The system includes a central processing unit, a fish-eye camera, a fish-eye camera position and attitude adjustment module, a communication module, an instruction input module, a processing result display module, and a data storage module; The fish-eye camera acquires two types of hemispherical canopy images upward from below the crop canopy and downward from above the crop canopy through the fish-eye camera position and attitude adjustment module, and uses the camera serial interface to transmit image data, control signals, and clock signals to the central processing unit; The communication module receives the hemispherical canopy image and the processing results of the crop leaf area index from the central processing unit, and remotely transmits data such as the hemispherical canopy image and the crop leaf area index through the network, and receives control instructions from the remote control platform, including leaf area index acquisition, image transmission, and query of the intermediate processing process; The instruction input and processing result display module uses a touch display panel to receive control instructions for on-site leaf area index acquisition, image transmission, and query of the intermediate processing process, and displays the corresponding data processing results; The data storage module uses an SD module to store the hemispherical canopy image, its intermediate processing process, and leaf area index data, and performs data interaction with each module through the central processing module; The central processing unit uses an arm architecture, embeds a linux operating system, and is equipped with a CSI, a display serial interface, and a microSD interface, and is used to implement the leaf area index of field crops based on the crop canopy gap fraction and data interaction with each module.

8. The monitoring system for the leaf area index of field crops based on machine vision according to claim 7, characterized in that, The fish-eye camera position and attitude adjustment module includes a main body support structure, a vertical adjustment structure for the fish-eye camera support cross arm, and an up-and-down flipping structure for the fish-eye camera; The main body support structure includes a support frame (8) and a main rod (6). The main rod (6) is vertically fixed on the support frame (8) and is used to place the main control box and the lead screw assembly (5). The whole composed of the fish-eye camera support cross arm (3), the flipping stepping motor (2), and the fish-eye camera (4) is connected to the lead screw assembly (5) through a nut; The vertical adjustment structure for the fish-eye camera support cross arm includes a stepping motor (1). By controlling the forward and reverse rotation of the lead screw by the stepping motor (1), the whole composed of the fish-eye camera support cross arm (3), the flipping stepping motor (2), and the fish-eye camera (4) is driven to move up and down, realizing precise movement in the vertical direction; The up-and-down flipping structure for the fish-eye camera includes a flipping stepping motor (2). The flipping stepping motor (2) is connected to the fish-eye camera (4) through the fish-eye camera support cross arm (3). By driving the fish-eye camera support cross arm (3) connected to the fish-eye lens (4) to flip around the horizontal axis by the flipping stepping motor (2), the fish-eye camera (4) can collect two types of hemispherical half-layer images from below the crop canopy upwards and from above the crop canopy downwards. That is, to realize the collection of hemispherical canopy images from below the crop canopy upwards, the stepping motor (1) controls the fish-eye camera support cross arm (3) to move downwards on the lead screw assembly (6) until it is below the crop canopy and close to the ground. Then, through the up-and-down flipping structure of the fish-eye camera, it receives the control signal from the central processing unit and drives the flipping stepping motor (2) to flip upwards, finally realizing the collection of hemispherical canopy images by the fish-eye camera (4) from below the crop canopy upwards. To realize the collection of hemispherical canopy images from above the crop canopy downwards, the stepping motor (1) controls the fish-eye camera support cross arm (3) to move upwards on the lead screw assembly until it is above the crop canopy. Then, through the up-and-down flipping structure of the fish-eye camera, it receives the control signal from the central processing unit and drives the flipping stepping motor (2) to flip downwards, finally realizing the collection of hemispherical canopy images by the fish-eye camera from above the crop canopy downwards.

Images