Soil moisture content monitoring method based on crop image multi-feature fusion

Abstract

The invention belongs to the technical field of soil moisture content monitoring, and particularly relates to a soil moisture content monitoring method based on crop image multi-feature fusion, and the method comprises the steps: 1, collecting images of crops at all growth stages; step 2, extracting multiple features, including color features, morphological features and texture features, of the crop images from the images of the crops in each growth stage; step 3, fusing the multiple features of the crop image in each growth stage of the crop to obtain the fused multiple features of the crop image; 4, detecting the soil moisture content of the crops at different depths in each growth stage; 5, constructing a soil moisture content calculation model of different depths in each growth stage of the crop by using a neural network; and step 6, collecting images of crops, extracting multiple features of the images of the crops, performing fusion to obtain fused multiple features of the images of the crops, and utilizing the soil moisture content calculation model of the corresponding depth of the corresponding growth stage to obtain the soil moisture content.

Description

Technical Field

[0001] The present application belongs to the technical field of soil moisture monitoring, and specifically relates to a soil moisture monitoring method based on multi-feature fusion of crop images. Background Art

[0002] Accurately obtaining the moisture content in the soil is of great significance for large-scale agricultural production. Currently, the moisture content is mostly monitored by deploying contact sensors in the soil, or the moisture content in the soil is obtained through manual sampling and laboratory analysis. This technical solution is costly and has limited coverage, making it difficult to quickly and efficiently obtain the distribution of soil moisture content over a large area.

[0003] This application is proposed in view of the above-mentioned technical defects.

[0004] It should be noted that the disclosure of the above background technology content is only used to assist in understanding the inventive concept and technical solution of the present invention, and it does not necessarily belong to the prior art of this patent application. In the absence of clear evidence that the above content has been disclosed on the filing date of this application, the above background technology should not be used to evaluate the novelty and creativity of the present application. Summary of the invention

[0005] The purpose of this application is to provide a soil moisture monitoring method based on multi-feature fusion of crop images to overcome or alleviate at least one of the known technical defects.

[0006] The technical solution of this application is:

[0007] A soil moisture monitoring method based on multi-feature fusion of crop images, characterized by comprising:

[0008] Step 1: Collect images of crops at various growth stages;

[0009] Step 2: Extract multiple features of crop images from images of crops at various growth stages, including color features, morphological features, and texture features;

[0010] Step 3: Fusing multiple features of crop images at different growth stages of crops to obtain fused multiple features of crop images;

[0011] Step 4: Detect soil moisture content at different depths during each growth stage of the crop;

[0012] Step 5: Using the fused multi-features of crop images at each growth stage of the crop as the independent variable and the soil moisture content at different depths at each growth stage of the crop as the dependent variable, a neural network is used to construct a soil moisture content calculation model at different depths at each growth stage of the crop;

[0013] Step 6: Collect crop images, extract multiple features of the crop images, fuse them, and obtain fused multiple features of the crop images. Use the soil moisture calculation model at the corresponding depth of the corresponding growth stage to obtain the soil moisture content.

[0014] According to at least one embodiment of the present application, in the above-mentioned soil moisture monitoring method based on multi-feature fusion of crop images, the growth stages of crops include jointing stage, tasseling-silking stage, and milky-mature stage.

[0015] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, the color features of the crop images include green dominance and color moment.

[0016] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, the green dominance includes absolute green dominance and relative green dominance, wherein:

[0017] The absolute green dominance is the ratio of the G component of all pixels to the sum of the R, G, and B components under the RGB color model;

[0018] The relative green dominance is the ratio of the G component to the R component and the B component of all pixels under the RGB color model.

[0019] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, the color moment of the crop image includes a first-order color moment, a second-order color moment, and a third-order color moment.

[0020] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images,

[0021]

[0022]

[0023] in,

[0024] μ i is the first-order color moment, σ i is the second-order color moment, ζ i is the third-order color moment, P ij is the i-th color component of the j-th pixel, and N is the total number of pixels.

[0025] According to at least one embodiment of the present application, in the above-mentioned soil moisture monitoring method based on multi-feature fusion of crop images, the morphological features of the crop images include leaf curl, plant height, plant width, leaf inclination angle value, and crop projection area.

[0026] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, the curling degree of the leaves is measured by the number of curling circles of the leaves.

[0027] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, n=L 0 / (π·d);

[0028] Where n is the number of curling circles of the blade, L 0 is the maximum leaf width when the leaf is stretched, and d is the maximum leaf width of the leaf in the crop image.

[0029] According to at least one embodiment of the present application, in the soil moisture monitoring method based on multi-feature fusion of crop images, the texture features of the crop image are reflected by the grayscale change characteristics within the quantized area.

[0030] This application has at least the following beneficial technical effects:

[0031] A soil moisture monitoring method based on multi-feature fusion of crop images is provided. The fused multi-features of crop images at various growth stages of crops are used as independent variables, and the soil moisture content at different depths at various growth stages of crops is used as a dependent variable. A soil moisture calculation model at different depths at various growth stages of crops is constructed using a neural network. Then, crop images can be collected to extract multi-features of the crop images, which are fused to obtain the fused multi-features of the crop images. The soil moisture calculation model at the corresponding depth of the corresponding growth stage is used to obtain the soil moisture content. The method has high accuracy and can quickly and efficiently obtain the distribution of soil moisture content over a large range. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a schematic diagram of a soil moisture monitoring method based on crop image multi-feature fusion provided in an embodiment of the present application;

[0033] Figure 2 is a schematic diagram of sequentially marking a connected domain pixel set of each leaf provided by an embodiment of the present application;

[0034] Figure 3 It is a schematic diagram of sequentially marking and extracting a single stretched leaf and a curled leaf, and detecting the minimum circumscribed rectangle of the leaf, provided by an embodiment of the present application;

[0035] Figure 4 It is a schematic diagram of converting a crop image from pixel level to region level and dividing it into several sub-domains after superpixel processing provided by an embodiment of the present application;

[0036] Figure 5It is a schematic diagram of performing feature block selection, principal component extraction, weighting and fusion on multiple features of crop images at various growth stages of crops provided by an embodiment of the present application;

[0037] Figure 6 It is a schematic diagram of a multi-scale regression algorithm model based on the YOLO series framework provided in an embodiment of the present application.

[0038] In order to better illustrate this embodiment, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product. In addition, the drawings are only used for illustrative purposes and cannot be understood as limitations on this patent. DETAILED DESCRIPTION

[0039] In order to make the technical solution and advantages of the present application clearer, the technical solution of the present application will be further described in detail in detail and in detail in conjunction with the accompanying drawings. It can be understood that the specific embodiments described here are only partial embodiments of the present application, which are only used to explain the present application, not to limit the present application. It should be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, and other related parts can refer to the general design. In the absence of conflict, the embodiments in the present application and the technical features in the embodiments can be combined with each other to obtain new embodiments.

[0040] In addition, unless otherwise defined, the technical terms or scientific terms used in the description of this application shall be the common meanings understood by the general technicians in the field to which this application belongs. The words "upper", "lower", "left", "right", "center", "vertical", "horizontal", "inside", "outside" and other words indicating orientation used in the description of this application are only used to indicate the relative direction or positional relationship, and do not imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. When the absolute position of the described object changes, its relative positional relationship may also change accordingly, so it cannot be understood as a limitation on this application. The "first", "second", "third" and similar terms used in the description of this application are only used for descriptive purposes to distinguish different components, and cannot be understood as indicating or implying relative importance. The similar words "one", "one" or "the" used in the description of this application should not be understood as an absolute limitation on quantity, but should be understood as the existence of at least one. The similar words "including" or "comprising" used in the description of this application mean that the elements or objects appearing in front of the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0041] In addition, it should be noted that, unless otherwise clearly specified and limited, the words "installed", "connected", "connected" and similar terms used in the description of this application should be understood in a broad sense. For example, the connection can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, an indirect connection through an intermediate medium, or the internal connection of two components. Technical personnel in the field can understand their specific meanings in this application according to the specific circumstances.

[0042] The moisture content in the soil is usually reflected in the appearance characteristics of the crop. For example, yellow and severely curled leaves often indicate low moisture content. Currently, there are some technical methods for analyzing soil moisture content through crop leaf image features, but most of them are to identify and analyze a single feature of the crop leaf surface, and then estimate the soil moisture content or only judge the drought level of the crop. The reliability is low and lacks real-time performance. Based on this, the present application provides a soil moisture monitoring method based on multi-feature fusion of crop images, such as Figure 1 shown.

[0043] Step 1: Collect images of crops at various growth stages.

[0044] The growth stages of crops may specifically include the jointing stage, the tasseling-silking stage, and the milky-maturity stage.

[0045] Specifically, a camera that can rotate freely or a drone with a camera can be used to collect crop images. After collecting crop images, they can be preprocessed and compressed to a size of 1920×1280. The image annotation tool LabelMe software is used to annotate the leaves of the crop and generate a JSON format file containing the image's location coordinates, category labels, image size and other annotation information. The open source computer vision library OpenCV can be used to annotate the annotated images and draw the coordinate images of the leaves respectively.

[0046] Step 2: Extract multiple features of crop images from images of crops at various growth stages, including color features, morphological features, and texture features.

[0047] 1) The color features of crop images include green dominance and color moment.

[0048] Green dominance includes absolute green dominance and relative green dominance. The absolute green dominance is the ratio of the G component of all pixels to the sum of the R, G, and B components under the RGB color model; the relative green dominance is the ratio of the G component of all pixels to the R component and the B component respectively under the RGB color model.

[0049] To extract the green dominance of crop images, specifically, the R, G, and B channels of images of crops at various growth stages can be separated to obtain grayscale images of three channels, and the grayscale values ​​of the leaf surface pixels of each channel can be read to further calculate the absolute green dominance and relative green dominance of the crop images.

[0050] The color moments of crop images include first-order color moments, second-order color moments, and third-order color moments. The first-order color moment reflects the average intensity of the color; the second-order color moment reflects the unevenness of the color distribution; and the third-order color moment reflects the asymmetry of the color.

[0051]

[0052]

[0053]

[0054] in,

[0055] μ i is the first-order color moment, σ i is the second-order color moment, ζ i is the third-order color moment, P ij is the i-th color component of the j-th pixel, and N is the total number of pixels.

[0056] To extract the color moment of the crop image, specifically, the calculation formulas of the first-order color moment, second-order color moment, and third-order color moment of the crop images at each growth stage can be embedded in the MATLAB code, and the first-order color moment, second-order color moment, and third-order color moment of each RGB and HSV component of the image can be calculated respectively, as well as the first-order color moment, second-order color moment, and third-order color moment after the RGB image is converted into a grayscale image.

[0057] 2) The morphological characteristics of crop images include leaf curl, plant height, plant width, leaf inclination angle, and crop projection area.

[0058] When a leaf curls, its cross section becomes approximately circular, and the degree of curl of the leaf is measured by the number of curling circles of the leaf.

[0059] n=L 0 / (π·d)…………(4)

[0060] Where n is the number of curling circles of the blade, L 0 is the maximum leaf width when the leaf is stretched, and d is the maximum leaf width of the leaf in the crop image.

[0061] Extract the curl of the leaves. Specifically, the script code can be compiled using Matlab software to separate the leaves of the segmented crop image and sequentially mark the connected domain pixel set of each leaf, such as Figure 2As shown, single stretched leaves and curled leaves are extracted in sequence, and the minimum enclosing rectangle of the leaves is detected, as shown in Figure 3 As shown, the maximum leaf width is calculated, and then the number of leaf curling circles is calculated to obtain the curling degree of each leaf.

[0062] To extract plant height and plant width, the im2bw function in the MATLAB system can be used to convert the segmented crop image from an RGB image to a binary image, and the smaller connected domains in the image can be cleared. The binary image of the crop can be pixel retrieved to detect the four edge vertices of the crop on the upper, lower, left and right sides. The row coordinate difference between the upper and lower edge vertices of the crop is calculated as the pixel length of the plant height, and the column coordinate difference between the left and right edge vertices is calculated as the pixel length of the plant width. The crop image before segmentation is retrieved, and the preset ruler in the image is segmented according to the color threshold. The pixel length corresponding to the actual unit length on the ruler is calculated to obtain the plant height and plant width.

[0063] Extract the leaf inclination angle value. Specifically, the superpixel clustering method can be used to calculate the mean and variance of the leaf inclination angle. After the crop image is segmented and binarized, two types of pixels, plant and black background, are retained. After superpixel processing, the crop image is converted from pixel level to region level and segmented into several subdomains, such as Figure 4 As shown, by detecting each subdomain, the minimum circumscribed rectangle of the leaves in each subdomain is determined, and then the deflection angle of each leaf, that is, the leaf inclination angle value, is calculated, and finally the mean and variance of the leaf inclination angles of all leaves of the crop can be calculated.

[0064] Extract the crop projection area. Specifically, you can set the area threshold, clear the small connected domains in the non-crop area, traverse the segmented crop image, calculate the total number of pixels of the crop part in the image, and for the crop projection area of ​​the front view and side view, calculate the number of pixels corresponding to the unit area according to the size of the preset scale in the crop image, so as to convert the projection area of ​​the crop plant; for the crop projection area of ​​the top view, extract the processed square edges, calculate the number of pixels of the enclosed area, and combine the processed actual size to calculate the number of pixels corresponding to the unit actual area, so as to convert the projection area of ​​the crop.

[0065] 3) The texture characteristics of crops can be reflected by the grayscale change characteristics within the quantized area, and the texture characteristics of crop images can be extracted. Specifically, the grayscale difference statistics method can be used to extract the texture characteristics of crop images. The RGB image of the crop is converted into a grayscale image, and the grayscale difference value of each crop pixel in the grayscale image and its lower right adjacent pixel is calculated to obtain a difference matrix. The frequency of occurrence of different difference values ​​in the difference matrix is ​​counted, and the average value, contrast, angular second-order moment, entropy and other statistics of the difference values ​​are calculated.

[0066] Gray=(R*299+G*587+B*114+500) / 1000…………(5)

[0067]

[0068] Con=∑ i [i 2 p(i)]…………(7)

[0069] Asm=∑ i [p(i) 2 ]…………(8)

[0070] Ent=-∑ i {[p(i)]log 2 [p(i)]}…………(9)

[0071] Among them, Gray is the gray value, R, G, B are the three-channel values ​​of the RGB image, Mean, Con, Asm, Ent are the average value, contrast, angular second moment, and entropy of the difference value, i is the gray difference value, and p(i) is the frequency of the gray difference value i.

[0072] Step 3: Fuse the multiple features of crop images at each growth stage of the crop to obtain fused multiple features of the crop images.

[0073] The ratio of the multi-features of the crop image at each growth stage of the crop to the multi-features of the crop image corresponding to the expected soil moisture content can be summed up to serve as the fused multi-features of the crop image.

[0074] The FWFAAPCA algorithm can also be used to perform feature block selection, principal component extraction, weighting and fusion on the multiple features of crop images at various growth stages to obtain the fused multiple features of the crop images, such as Figure 5 As shown, please refer to the following for details:

[0075] 1) Feature block selection

[0076] The multiple features of crop images at different growth stages are identified in blocks, and feature blocks are selected based on the block feature dimension.

[0077] The color feature, morphological feature, and texture feature blocks are selected to have the same dimensions, as follows:

[0078]

[0079]

[0080]

[0081] in,

[0082] R(Ci), R(Sj), and R(Tk) are the recognition rates of the feature blocks of color features, morphological features, and texture features; R(C), R(S), and R(t) are the recognition rates of the final selected feature blocks of color features, morphological features, and texture features; D(t) and D(T) are the dimensions of the candidate and final selected feature blocks of texture features; and τ is the feature block tolerance.

[0083] 2) Extract principal components

[0084] Assume that the D-dimensional sample feature is x=(x 1 , x 2 , …, x D )T, and obtain the P-dimensional feature y=(y 1 ,y 2 , …, y p )T, (P≤D), and from it the optimal o-dimensional principal component feature yo=(y 1 ,y 2 , …, y o )T is extracted to satisfy:

[0085]

[0086] Where: C(y o )、R(y o )、V(y o ) is the contribution rate, recognition rate and rising speed of the main component; T is C(y o ) is 80%, η is the tolerance (the principal component is 10-12%), R(y s )、R(y r ) is the recognition rate of the first s and first r principal components.

[0087] 3) Weighting and fusion

[0088] First, the PCA principal components are identified, and the three types of principal components are weighted using the recognition rate. Then, the principal component dimensions of color features, morphological features, and texture features are used to weight the individual components. Finally, the weighted components are fused.

[0089] Assume that the main component of OC dimension color is y C =(y C 1 ,y C 2 , …, y C OC )T, the OS-dimensional morphological principal component is y S =(y S 1 ,y S 2 , …, y SOS )T, the principal component of the OT-dimensional texture is y T =(y T 1 ,y T 2 , …, y T OS )T, the weights of the principal components of color, morphology, and texture are calculated as follows:

[0090]

[0091] Among them, Oθ is the feature dimension, is the recognition rate of color features, morphological features, and texture features, θ∈(C, S, T).

[0092] The weight distribution formula is used to calculate the weight of each individual component, as follows:

[0093]

[0094] Among them, d is the weight step size, ω θ (ρ) is the weight of the ρth component in the principal components.

[0095] Feature fusion consists of principal components and single component weights, calculated as follows:

[0096]

[0097] Step 4: Detect the soil moisture content at different depths during each growth stage of the crop.

[0098] Specifically, a soil moisture monitoring device can be used to detect the soil moisture content at different depths at each growth stage of the crop.

[0099] Three soil moisture monitoring devices can be pre-buried at soil depths of 20cm, 40cm, and 60cm, and deployed in the crop root zone with a monitoring frequency of 15 minutes for all-weather monitoring. The signal collector calculates the average value of the data collected by the three soil moisture monitoring devices within 10 minutes every 10 minutes as the soil moisture content.

[0100] Step 5. Using the fused multi-features of crop images at various growth stages of crops as independent variables and the soil moisture content at different depths at various growth stages of crops as dependent variables, a neural network is used to construct a soil moisture content calculation model at different depths at various growth stages of crops.

[0101] The fused multi-features of crop images at various growth stages and the soil moisture content at different depths at the corresponding moments are combined into a long sequence matrix. The training set and validation set are screened using the SPXY algorithm, with 80% of the data used for modeling and 20% for validation.

[0102] The GNN graph neural network uses the fused multiple features of crop images at various growth stages of crops as independent variables, and the soil moisture content at different depths at various growth stages of crops as the dependent variable to input the model to obtain the coupling results.

[0103] The GNN graph neural network graph learning layer can be used to automatically learn the dependencies between variables in multivariate time series and represent them as the adjacency matrix of the graph. The graph structure is constructed by learning node embedding and unidirectional relationships are extracted. The graph convolution module processes the spatial dependencies in the graph. The Mix-hop propagation layer processes the spatial dependencies through two steps of information propagation and information selection, and recursively propagates node information along the given graph structure. The parameter matrix is ​​used as a feature selector to filter out unimportant information generated in each propagation step.

[0104] The multivariate time series forecasting task can be reformulated as a forecast on a hypervariate graph, and the operation can be expressed as:

[0105]

[0106] The convolution equation can be expressed as:

[0107]

[0108] The time convolution module is used to capture the time pattern in the time series data, and finally standardizes the information jumping to the output module to make it have the same sequence length. The mean average precision, precision, recall and F1score are used to evaluate the detection model and machine learning model. The calculation formula is as follows:

[0109]

[0110]

[0111]

[0112]

[0113] Among them, TP is the positive example with correct classification, FP is the positive example with incorrect classification, FN and TN are the negative examples with incorrect classification and the negative examples with correct classification respectively, N is the number of image samples, Precision is the recognition accuracy of the model, Recall is the recall rate of the model, F1score is the harmonic mean of precision and recall rate, and MAP is the mean average precision of the model for detection.

[0114] The initial parameters of the soil moisture calculation model at different depths in each growth stage of crops can be set, and the input layer parameters, graph convolution layer parameters, readout layer parameters and output layer parameters can be determined. The training rounds are 100 rounds, the initial learning rate is 0.01, and the regularization parameter is set to 0.0001 to prevent overfitting. Set up an ablation experiment, change only one parameter each time, and keep other parameters unchanged. Train and evaluate each modified model, tune the model parameters according to the evaluation data, and establish a soil moisture calculation model based on different depths in each growth stage of crops.

[0115] Using deep learning algorithms, we can further analyze and mine the fused multiple features, discover the potential laws and patterns in the crop growth process, monitor and warn of crop growth conditions in real time, and further analyze and predict the fused multiple features, providing a scientific basis for the precise management and optimization of crops.

[0116] According to the fusion of multiple features and soil moisture data, the multi-scale regression algorithm based on the YOLO series framework can optimize sample overlap and occlusion problems and improve target detection efficiency. The multi-scale regression algorithm model based on the YOLO series framework consists of four parts: input end, backbone network, connection network and prediction end. Figure 6 As shown in the figure, conv represents convolution and concat represents the connection function. The input module performs data enhancement, image adaptive scaling, anchor box calculation and other operations; the backbone network module introduces a pixel reconstruction network and a gradient integration structure to reduce the amount of calculation and increase the speed of sample information extraction; the connection network module uses the feature pyramid network to perform up and down sampling, strengthen feature fusion, and improve sample diversity and robustness; the prediction module adds regression training of the GoU Loss function to optimize sample overlap and occlusion problems and improve target detection efficiency. O U optimizes the position regression of the target sequence by combining I O The global information of U is added to the boundary sequence of the non-overlapping part to measure the similarity between the predicted boundary sequence and the true boundary sequence.

[0117]

[0118]

[0119] Among them, I O U is the intersection of the predicted sequence and the true sequence, A c is the minimum closure range of the predicted and true sequences, and U is the range in the closure region that does not belong to the two sequences.

[0120] Step 6: Collect crop images in real time, extract multiple features of the crop images, fuse them, and obtain fused multiple features of the crop images. Use the soil moisture calculation model at the corresponding depth of the corresponding growth stage to obtain the soil moisture content.

[0121] The soil moisture monitoring method based on multi-feature fusion of crop images disclosed in the above embodiment uses the fused multi-features of crop images at various growth stages of crops as independent variables, and the soil moisture content at different depths at various growth stages of crops as dependent variables, and uses a neural network to construct a soil moisture calculation model at different depths at various growth stages of crops. Then, by collecting crop images, extracting multiple features of the crop images, and fusing them, the fused multi-features of the crop images are obtained. The soil moisture content is obtained by using the soil moisture calculation model at the corresponding depth of the corresponding growth stage. The method has high accuracy and can quickly and efficiently obtain the distribution of soil moisture content over a large area.

[0122] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. Those skilled in the art should understand that the scope of protection of the present application is obviously not limited to these specific embodiments. Without departing from the principles of the present application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the scope of protection of the present application.

Claims

1. A soil moisture monitoring method based on multi-feature fusion of crop images, characterized in that: include: Step 1: Collect images of crops at various growth stages; Step 2: Extract multiple features of crop images from images of crops at various growth stages, including color features, morphological features, and texture features; Step 3: Fusing multiple features of crop images at different growth stages of crops to obtain fused multiple features of crop images; Step 4: Detect soil moisture content at different depths during each growth stage of the crop; Step 5: Using the fused multi-features of crop images at each growth stage of the crop as the independent variable and the soil moisture content at different depths at each growth stage of the crop as the dependent variable, a neural network is used to construct a soil moisture content calculation model at different depths at each growth stage of the crop; Step 6: Collect crop images, extract multiple features of the crop images, fuse them, and obtain fused multiple features of the crop images. Use the soil moisture calculation model at the corresponding depth of the corresponding growth stage to obtain the soil moisture content.

2. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 1 is characterized in that: The growth stages of crops include jointing stage, tasseling-silking stage, and milky-maturity stage.

3. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 2 is characterized in that: The color features of crop images include green dominance and color moment.

4. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 3 is characterized in that: Green dominance includes absolute green dominance and relative green dominance. The absolute green dominance is the ratio of the G component of all pixels to the sum of the R, G, and B components under the RGB color model; The relative green dominance is the ratio of the G component to the R component and the B component of all pixels under the RGB color model.

5. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 4 is characterized in that: The color moments of crop images include first-order color moments, second-order color moments, and third-order color moments.

6. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 5 is characterized in that: in, μ i is the first-order color moment, σ i is the second-order color moment, ζ i is the third-order color moment, P ij is the i-th color component of the j-th pixel, and N is the total number of pixels.

7. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 6 is characterized in that: The morphological features of crop images include leaf curl, plant height, plant width, leaf inclination angle, and crop projection area.

8. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 7 is characterized in that: The curl of the blade is measured by the number of curl turns of the blade.

9. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 8 is characterized in that: n=L0 / (π·d); Among them, n is the number of leaf curling circles, L0 is the maximum leaf width when the leaf is stretched, and d is the maximum leaf width of the leaf in the crop image.

10. The soil moisture monitoring method based on crop image multi-feature fusion according to claim 9 is characterized in that: The texture characteristics of crop images are reflected by the grayscale variation characteristics within the quantized area.

Images