Intelligent fertilization control method for improving specific characters of pasture based on pasture growth characteristics

Abstract

The invention discloses an intelligent fertilization control method for improving specific characters of pasture based on pasture growth characteristics, and the method comprises the steps: carrying out the semantic feature extraction of a pasture growth state image through employing an image processing technology based on deep learning, so as to excavate key semantic features which can reflect the growth states of the pasture in different growth stages; and the expression ability of the features is enhanced through multi-level feature semantic enhancement, so that the growth state of the pasture can be identified more accurately, the specific growth characteristic category label of the pasture can be intelligently determined, and the appropriate fertilization amount can be provided for the pasture by combining the corresponding relation table of the pasture growth characteristics and the soil fertilizer amount obtained through a large amount of experimental data. Thus, the optimal fertilization strategy can be automatically recommended according to the specific growth state of the pasture, excessive fertilization or insufficient fertilization is avoided, it is ensured that the pasture obtains appropriate nutrients in all growth stages, intelligent fertilization control is achieved, and therefore the yield and quality of the pasture are further improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics. Background Art

[0002] The growth of forage grass is affected by many factors, among which soil fertility is one of the key factors. The right amount of fertilizer can significantly improve the yield and quality of forage grass, but excessive fertilization will not only cause resource waste, but also cause environmental pollution. Therefore, precision fertilization is an important topic in modern agriculture.

[0003] Traditional fertilization methods mainly rely on farmers' experience and simple soil tests, lack real-time monitoring and accurate evaluation of forage growth status, and are often unable to adjust fertilization strategies in a timely manner, resulting in low fertilizer utilization, uneven forage growth, and unstable yield and quality.

[0004] Therefore, an intelligent fertilization control method based on forage growth characteristics to improve specific traits of forage is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an intelligent fertilization control method for improving the specific traits of forage based on forage growth characteristics, which extracts semantic features from the forage growth state image by using deep learning-based image processing technology to dig out key semantic features that can reflect the growth state of forage at different growth stages, and enhances the expression ability of features through multi-level feature semantic enhancement, so as to more accurately identify the growth state of forage, so as to intelligently determine its specific growth characteristic category label, and then, combined with the correspondence table of forage growth characteristics and soil fertilizer obtained through a large amount of experimental data, provide appropriate fertilizer amount for forage. In this way, the best fertilization strategy can be automatically recommended according to the specific growth state of forage, avoiding over-fertilization or under-fertilization, ensuring that forage obtains appropriate nutrients at each growth stage, so as to realize intelligent fertilization control, thereby further improving the yield and quality of forage.

[0006] According to one aspect of the present application, there is provided an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics, which comprises:

[0007] Construct growth characteristics-soil fertilizer recommendation table;

[0008] Acquire the grass growth status image collected by the camera;

[0009] Performing grass growth state feature extraction and semantic enhancement processing on the grass growth state image to obtain a grass growth state semantic enhancement feature map;

[0010] Performing feature enhancement processing on the semantic enhancement feature map of the grass growth state to obtain a multi-dimensional enhancement feature map of the grass growth state;

[0011] Determining a category label of growth characteristics based on the multi-dimensional enhanced feature map of forage growth status;

[0012] A fertilization strategy is determined based on the category label of the growth characteristics and the growth characteristics-soil fertilizer amount recommendation table.

[0013] Compared with the prior art, the present application provides an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics. It uses deep learning-based image processing technology to extract semantic features from images of forage growth status, so as to mine key semantic features that can reflect the growth status of forage at different growth stages, and enhance the expressiveness of features through multi-level feature semantic enhancement, so as to more accurately identify the growth status of forage, thereby intelligently determining its specific growth characteristic category label, and then, combined with the correspondence table between forage growth characteristics and soil fertilizer obtained through a large amount of experimental data, to provide the appropriate amount of fertilizer for forage. In this way, the best fertilization strategy can be automatically recommended according to the specific growth status of forage, to avoid over-fertilization or under-fertilization, to ensure that forage obtains appropriate nutrients at each growth stage, to achieve intelligent fertilization control, and thus further improve the yield and quality of forage. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 It is a flow chart of an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics according to an embodiment of the present application;

[0016] Figure 2 A data flow diagram of an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics according to an embodiment of the present application;

[0017] Figure 3 This is a flowchart of sub-step S4 of the intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0019] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0020] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0021] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] Traditional fertilization methods mainly rely on farmers' experience and simple soil tests, lack real-time monitoring and accurate evaluation of grass growth status, and often fail to adjust fertilization strategies in time, resulting in low fertilizer utilization, uneven grass growth, and unstable yield and quality. Therefore, an intelligent fertilization control method based on grass growth characteristics to improve specific traits of grass is expected.

[0024] In the technical solution of the present application, an intelligent fertilization control method is proposed for improving specific traits of forage based on forage growth characteristics. Figure 1 The present invention is a flowchart of an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of an intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to an embodiment of the present application, an intelligent fertilization control method 300 for improving specific traits of forage based on forage growth characteristics includes the following steps: S1, constructing a growth characteristic-soil fertilizer amount recommendation table; S2, acquiring a forage growth status image captured by a camera; S3, performing forage growth status feature extraction and semantic enhancement processing on the forage growth status image to obtain a forage growth status semantic enhancement feature map; S4, performing feature enhancement processing on the forage growth status semantic enhancement feature map to obtain a forage growth status multi-dimensional enhancement feature map; S5, determining a category label of growth characteristics based on the multi-dimensional enhancement feature map of forage growth status; S6, determining a fertilization strategy based on the category label of the growth characteristics and the growth characteristics-soil fertilizer amount recommendation table.

[0025] In particular, S1 constructs a growth characteristics-soil fertilizer recommendation table. The growth characteristics-soil fertilizer recommendation table is established based on a large amount of experimental data, and records in detail the demand for soil fertilizer of forage grasses under different growth characteristics. It is a reference data for guiding the type and amount of fertilizer required for forage grasses at different growth stages. In other words, based on the growth characteristics-soil fertilizer recommendation table, the fertilization strategy can be dynamically adjusted according to the growth status of forage grasses and soil fertility, providing a scientific basis for fertilization, thereby improving the growth quality and yield of forage grasses, and helping to save resources.

[0026] In particular, the S2 obtains a grass growth status image captured by a camera. It should be understood that installing a camera to capture grass growth status images in real time can provide immediate growth feedback. In the grass growth status image, the color and texture of the grass blades can reflect its growth status and health status; the height of the grass plant can reflect its growth stage and growth rate. Therefore, in the technical solution of the present application, it is expected to use grass growth status information to classify the grass growth status, so as to determine its specific growth characteristic category label, so as to determine the corresponding fertilization strategy in combination with the specific growth characteristic category label of the grass and the growth characteristics-soil fertilizer recommendation table.

[0027] In particular, the S3 performs grass growth state feature extraction and semantic enhancement processing on the grass growth state image to obtain a grass growth state semantic enhancement feature map. In a specific example of the present application, first, in order to extract the key features that best reflect the growth state of grass, efficient feature extraction and deep semantic understanding are achieved. In the technical solution of the present application, first, the grass growth state image is input into a growth state feature extractor based on a deep separable convolutional neural network model to obtain a grass growth state feature map; wherein the deep separable convolutional neural network model includes deep convolution and point-by-point convolution, and the deep convolution performs convolution operations on each channel of the grass growth state image separately to capture the spatial information of the input data, which is very useful for identifying local features such as color and texture of grass leaves; point-by-point convolution performs information fusion between multiple channels through a 1x1 convolution kernel to help the model learn higher-level abstract features. By extracting features of different scales, it helps the model to describe the growth state of grass from multiple angles from the global to the local, and improve the richness and diversity of features. Next, considering that the deep separable convolutional neural network model mainly focuses on the extraction of local features, it is impossible to fully consider the contextual relationship between features, which makes it difficult to accurately distinguish and identify in complex scenes. Compared with traditional convolutional neural networks, converters can better capture global information in images, thereby providing a more comprehensive semantic description. Therefore, in the technical solution of the present application, the grass growth state feature map is further input into a growth state context encoder based on a converter structure to obtain the grass growth state semantic enhancement feature map. Among them, the converter can establish associations between features at different scales through a self-attention mechanism, process long-distance dependencies and contextual information, and extract higher-level semantic information from the feature map. This helps to better understand the overall growth state of the grass. By effectively combining local and global information, a comprehensive analysis of the grass growth state is further achieved. In one example, the grass growth state feature map is semantically encoded based on context using a vit model to obtain the grass growth state semantic enhancement feature map.

[0028] In particular, the S4 performs feature enhancement processing on the semantically enhanced feature map of the grass growth status to obtain a multi-dimensional enhanced feature map of the grass growth status. Considering that traditional convolutional neural networks often find it difficult to fully explore the complex semantic associations within the image when processing visual tasks, especially when faced with areas with subtle differences but closely related semantics, in order to further refine and enhance the feature expression capabilities of each pixel in the feature map, so as to more accurately capture the subtle changes in the grass growth status and improve the feature resolution and description capabilities. In the technical solution of the present application, pixel granularity feature enhancement processing is performed on the semantically enhanced feature map of the grass growth status to obtain the multi-dimensional enhanced feature map of the grass growth status. In a specific example of the present application, such as Figure 3 As shown, the S4 includes: S41, calculating the pixel-level semantic association score of the forage growth status semantic enhancement feature map to obtain the forage growth status pixel-granularity semantic association score topological matrix; S42, based on the pixel spatial position information of the forage growth status semantic enhancement feature map, performing spatial attention attenuation modulation on the forage growth status pixel-granularity semantic association score topological matrix to obtain the forage growth status semantic association space soft constraint score topological feature matrix; S43, based on the forage growth status semantic association space soft constraint score topological feature matrix, performing pixel-granularity spatial enhancement processing on the forage growth status semantic enhancement feature map to obtain the forage growth status multi-dimensional enhanced feature map.

[0029] Specifically, the S41 calculates the pixel-level semantic association score of the grass growth state semantic enhancement feature map to obtain a grass growth state pixel granularity semantic association score topological matrix. By calculating the semantic association score between any two pixel granularity feature vectors in the set, a pixel granularity semantic association score topological matrix is ​​constructed to reveal the intrinsic semantic association of the image. Specifically, first, the feature of the grass growth state semantic enhancement feature map is dispersed along the channel dimension to modulate the feature shape of the grass growth state semantic enhancement feature map into a feature distribution set of pixel granularity to obtain a set of grass growth state pixel granularity feature vectors; then, the semantic association score between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors is calculated to obtain the grass growth state pixel granularity semantic association score topological matrix. In particular, in this process, considering that compared with Euclidean space, hyperbolic space can more effectively capture the nonlinear relationship between data, the present application first maps each pixel granularity feature vector from Euclidean space to Poincare space, so that it adapts to the geometric characteristics of hyperbolic space while maintaining the original semantic information, and calculates the Poincare distance between any two mapped pixel granularity feature vectors as the semantic association score between the two, so as to utilize the geometric characteristics of Poincare space to enhance the nonlinear association between features, improve the model's ability to distinguish features of different categories, and thus more accurately simulate the intrinsic semantic association structure of the image.

[0030] Specifically, in S42, based on the pixel spatial position information of the forage growth state semantic enhancement feature map, the forage growth state pixel granularity semantic association score topological matrix is ​​subjected to spatial attention attenuation modulation to obtain the forage growth state semantic association space soft constraint score topological feature matrix. In particular, considering that adjacent areas in reality often have stronger correlation than distant areas, it is necessary to perform spatial attention attenuation modulation on the forage growth state pixel granularity semantic association score topological matrix. Here, a distance attenuation function is defined and applied to the score to obtain the forage growth state semantic association space soft constraint score topological feature matrix that is more in line with the actual situation. Specifically, in a specific example of the present application, the forage growth state pixel granularity semantic association score topology matrix is ​​subjected to spatial attention attenuation modulation based on the spatial distance between any two forage growth state pixel granularity feature vectors in the set of the forage growth state pixel granularity feature vectors to obtain the forage growth state semantic association space soft constraint score topology matrix; further, the forage growth state semantic association space soft constraint score topology matrix is ​​subjected to hole convolution coding to obtain the forage growth state semantic association space soft constraint score topology feature matrix. That is, the forage growth state semantic association space soft constraint score topology matrix is ​​subjected to hole convolution coding using the hole convolution technology to capture the semantic association topology features between each forage growth state pixel granularity feature vector of the forage growth state semantic association space soft constraint score topology matrix. Among them, the process of performing spatial attention attenuation modulation on the grass growth status pixel granularity semantic association score topological matrix based on the spatial distance between any two grass growth status pixel granularity feature vectors in the set of grass growth status pixel granularity feature vectors to obtain the grass growth status semantic association spatial soft constraint score topological matrix includes: calculating the natural exponential function value of the Euclidean distance between any two grass growth status pixel granularity feature vectors in the set of grass growth status pixel granularity feature vectors to obtain the grass growth status pixel granularity spatial distance topological matrix; multiplying the grass growth status pixel granularity spatial distance topological matrix by a preset attenuation modulation constant to obtain a spatial attention attenuation modulation matrix, and calculating the position point-by-position division of the grass growth status pixel granularity semantic association score topological matrix and the spatial attention attenuation modulation matrix to obtain the grass growth status semantic association spatial soft constraint score topological matrix.

[0031] Specifically, the S43, based on the soft constraint score topological feature matrix of the semantic association space of the forage growth state, performs pixel granularity spatial enhancement processing on the semantic enhancement feature map of the forage growth state to obtain the multi-dimensional enhanced feature map of the forage growth state. In a specific example of the present application, first, the set of pixel granularity feature vectors of the forage growth state and the soft constraint score topological feature matrix of the semantic association space of the forage growth state are input into the graph convolution encoding module to obtain a set of context-related enhanced forage growth state pixel granularity feature vectors; by utilizing the characteristics of graph structure data, that is, the connection relationship between nodes (pixels), the local and global context information between pixels of the forage growth state is captured. By taking the set of pixel granularity feature vectors of the forage growth state and the soft constraint score topological feature matrix of the semantic association space as input, the graph convolution can effectively enhance the context-related features of each pixel point of the forage growth state to obtain the set of context-related enhanced forage growth state pixel granularity feature vectors. It should be understood that although the set of the context-related enhanced grass growth state pixel granularity feature vectors contains the feature information of each grass growth state pixel point, this information exists in a one-dimensional form, and the spatial position relationship between the pixels is lost. Therefore, further, the set of the context-related enhanced grass growth state pixel granularity feature vectors is reshaped to obtain the multi-dimensional enhanced feature map of the grass growth state. Here, the feature shape reshaping further enhances the spatial distribution characteristics between the grass growth state pixels, making the association between adjacent grass growth state pixels more obvious, which helps the model to better understand and utilize context information and improve the recognition accuracy of the grass growth state.

[0032] In summary, in the above embodiment, the feature enhancement processing is performed on the semantic enhancement feature map of the grass growth state to obtain the multi-dimensional enhanced feature map of the grass growth state, including: processing the semantic enhancement feature map of the grass growth state with the following feature enhancement formula to obtain the multi-dimensional enhanced feature map of the grass growth state, wherein the feature enhancement formula is:

[0033]

[0034] h i =W1v i W2

[0035]

[0036]

[0037] T = AtrousConv(S′)

[0038] V′=GCN(V,T)

[0039] F = reshape[V′,(H,W,C)]

[0040] Where X is the semantic enhancement feature map of the grass growth state, H, W and C represent the height, width and number of channels of the semantic enhancement feature map of the grass growth state, reshape[·] represents the feature shape reshaping, V represents the set of pixel granularity feature vectors of the grass growth state, v i and v j represent the i-th and j-th grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors respectively, the value of HW is the number of feature vectors in the set of grass growth state pixel granularity feature vectors, and The value of C is the length of the pixel granularity feature vector of the grass growth state, W1 is the weight transformation matrix, W2 is the weight transformation vector, and h i and h j denote the i-th and j-th post-mapping grass growth state pixel granularity feature vectors in the set of post-mapping grass growth state pixel granularity feature vectors, arccosh(·) denotes the hyperbolic space inverse hyperbolic cosine function, ||·|| denotes the norm of the feature vector, d P (h i ,h j ) represents the Poincare distance between the i-th mapped grass growth state pixel granularity feature vector and the j-th mapped grass growth state pixel granularity feature vector, S ij represents the semantic association score value of the position (i, j) in the topological matrix of the semantic association score of the pixel granularity of the grass growth state, α is the preset attenuation modulation constant, d(v i ,v j ) represents the Euclidean distance between the i-th grass growth state pixel granularity feature vector and the j-th grass growth state pixel granularity feature vector, S' is the soft constraint score topological matrix of the grass growth state semantic association space, S ij ' is the eigenvalue of the position (i, j) in the soft constraint score topological matrix of the semantic association space of the grass growth status, AtrousConv(·) represents the void convolution coding, T is the soft constraint score topological feature matrix of the semantic association space of the grass growth status, GCN(·) represents the graph convolution processing, V' represents the set of pixel granularity feature vectors of the context-related enhanced grass growth status, and F represents the multi-dimensional enhanced feature map of the grass growth status.

[0041] In particular, the S5 determines the category label of the growth characteristics based on the multi-dimensional enhanced feature map of the grass growth state. In a specific example of the present application, the multi-dimensional enhanced feature map of the grass growth state is input into a growth characteristic identifier based on a classifier to obtain the category label of the growth characteristics. That is, by utilizing the powerful classification mapping ability of the classifier model, by learning the grass growth state information contained in the multi-dimensional enhanced feature map of the grass growth state, an accurate recognition result is intelligently output, thereby obtaining the category label of the growth characteristics.

[0042] That is, the applicant of the present application takes into account that the grass growth state feature map represents the image semantic features of the grass growth state image. Inputting the grass growth state feature map into a growth state context encoder based on a converter structure is to use the growth state context encoder based on a converter structure to perform fine-grained global context association encoding on the grass growth state feature map to expand the feature receptive field of the grass growth state feature map to obtain the grass growth state semantic enhancement feature map. Inputting the grass growth state semantic enhancement feature map into a feature enhancement module based on spatial soft constraints is to use the feature enhancement module based on spatial soft constraints to optimize the spatial dimension feature expression significance of the grass growth state semantic enhancement feature map to obtain the grass growth state multi-dimensional enhancement feature map. However, the enhancement of feature distribution in different dimensions will also make the complex spatial structure of the multi-dimensional enhancement feature map of the grass growth state based on the enhancement of feature distribution in different dimensions. Therefore, it is expected to improve the semantic consistency aggregation expression effect of the multi-dimensional enhancement feature map of the grass growth state.

[0043] Preferably, in an example of the present application, inputting the multi-dimensional enhanced feature map of the forage growth state into a classifier-based growth characteristic identifier to obtain a category label of the growth characteristic comprises:

[0044] The feature set of the multi-dimensional enhanced feature map of the forage growth state is subjected to feature clustering to obtain a multi-dimensional enhanced in-class feature set of the forage growth state and a multi-dimensional enhanced out-class feature set of the forage growth state, namely:

[0045]

[0046] in, represents the multi-dimensional enhanced intra-class feature set of forage growth status, f 1i represents the i-th eigenvalue of the multi-dimensional enhanced intra-class feature set of forage growth status, f 2j Represents the jth eigenvalue of the multi-dimensional enhanced out-of-class feature set of forage growth status.

[0047] The ratio of the number of eigenvalues ​​in the feature set of the multi-dimensional enhanced class of the forage growth state to the number of eigenvalues ​​in the feature set of the multi-dimensional enhanced feature map of the forage growth state is calculated, that is:

[0048]

[0049] Among them, k represents the number of eigenvalues ​​in the intra-class feature set of the multi-dimensional enhanced grass growth state, n represents the number of eigenvalues ​​in the feature set of the multi-dimensional enhanced feature map of the grass growth state, and λ represents the ratio.

[0050] The ratio of the λ-th power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the forage growth state multidimensional enhancement class to the λ-th power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the forage growth state multidimensional enhancement feature map is calculated to obtain the forage growth state multidimensional enhancement modulation weight, that is:

[0051]

[0052] Among them, f p represents each feature value in the feature set of the multi-dimensional enhanced feature map of the forage growth state, w1 represents the multi-dimensional enhanced modulation weight of the forage growth state;

[0053] The ratio of the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the multidimensional enhancement class of the forage growth state to the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the multidimensional enhancement feature map of the forage growth state is calculated to obtain the multidimensional enhancement harmonic weight of the forage growth state, that is:

[0054]

[0055] Among them, w2 represents the multi-dimensional enhanced harmonic weight of forage growth status;

[0056] For each feature value in the forage growth state multidimensional enhancement intra-class feature set, calculate the product of the feature value and the forage growth state multidimensional enhancement harmonic weight value plus the forage growth state multidimensional enhancement modulation weight value to obtain the optimized feature value, that is:

[0057] f′ 1i =w2×f 1i +w1

[0058] Among them, f' 1i represents the i-th optimized eigenvalue;

[0059] For each feature value in the forage growth state multi-dimensional enhanced out-of-class feature set, the product of the feature value and the forage growth state multi-dimensional enhanced value adjustment weight is calculated to obtain the optimized feature value, that is:

[0060] f 2j '=w1×f 2j

[0061] Among them, f 2j ' represents the jth optimized eigenvalue;

[0062] The optimized forage growth state multidimensional enhanced feature map based on the optimized feature values ​​of the forage growth state multidimensional enhanced in-class feature set and the forage growth state multidimensional enhanced out-class feature set is input into a classifier-based growth characteristic identifier to obtain the category label of the growth characteristic.

[0063] Therefore, while performing feature clustering on the multidimensional enhanced feature map of forage growth status, the interactive description of the key feature information of the multidimensional enhanced feature map of forage growth status in the clustering process is carried out, and the geometric equivariant topology of the features is constructed by low-rank harmonic modulation of the features on the equivariant basis of the clustering features of the multidimensional enhanced feature map of forage growth status and the features as a whole, so as to obtain the translation and rotational symmetry of the schematic distribution of the clustering features of the multidimensional enhanced feature map of forage growth status relative to the features as a whole, thereby introducing geometric message passing into the feature expression of the multidimensional enhanced feature map of forage growth status and realizing the clustering mapping symmetry of the multidimensional enhanced feature map of forage growth status through manipulation of irreducible low-rank order coefficients, thereby improving the consistency of the clustering-based feature representation of the multidimensional enhanced feature map of forage growth status, thereby improving the accuracy of the category label of the growth characteristic obtained by inputting the multidimensional enhanced feature map of forage growth status into the classifier-based growth characteristic identifier.

[0064] In particular, the S6 determines the fertilization strategy based on the category label of the growth characteristics and the growth characteristics-soil fertilizer recommendation table. In a specific example of the present application, based on the category label of the growth characteristics, the soil fertilizer recommendation result is extracted from the growth characteristics-soil fertilizer recommendation table; then, based on the comparison between the soil fertilizer recommendation result and the current soil fertilizer status, the fertilization strategy is determined. In one example, if the category label of the growth characteristics is "nitrogen deficiency", the category label "nitrogen deficiency" is used to find the corresponding soil fertilizer recommendation value from the growth characteristics-soil fertilizer recommendation table; then, the fertilization strategy is determined based on the recommended soil fertilizer and the current soil fertilizer status. In this way, precise fertilization management based on the type label of growth characteristics and the current soil fertilizer status can be achieved, which can improve the growth quality and yield of forage and help save resources.

[0065] In summary, according to the embodiment of the present application, the intelligent fertilization control method for improving the specific traits of forage based on the growth characteristics of forage is explained, which extracts semantic features from the forage growth state image by using deep learning-based image processing technology to dig out key semantic features that can reflect the growth state of forage at different growth stages, and enhances the expressive ability of features through multi-level feature semantic enhancement, so as to more accurately identify the growth state of forage, so as to intelligently determine its specific growth characteristic category label, and then, combined with the correspondence table of forage growth characteristics and soil fertilizer obtained through a large amount of experimental data, provide the appropriate amount of fertilizer for forage. In this way, the best fertilization strategy can be automatically recommended according to the specific growth state of forage, avoiding over-fertilization or under-fertilization, ensuring that forage obtains appropriate nutrients at each growth stage, so as to realize intelligent fertilization control, thereby further improving the yield and quality of forage.

[0066] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent fertilization control method for improving specific traits of forage based on forage growth characteristics, characterized in that: include: Construct growth characteristics-soil fertilizer recommendation table; Acquire the grass growth status image collected by the camera; Performing grass growth state feature extraction and semantic enhancement processing on the grass growth state image to obtain a grass growth state semantic enhancement feature map; Performing feature enhancement processing on the semantic enhancement feature map of the grass growth state to obtain a multi-dimensional enhancement feature map of the grass growth state; Determining a category label of growth characteristics based on the multi-dimensional enhanced feature map of forage growth status; Determining a fertilization strategy based on the category label of the growth characteristics and the growth characteristics-soil fertilizer amount recommendation table; Among them, the feature enhancement processing is performed on the semantic enhancement feature map of the grass growth status, including: calculating the pixel-level semantic association score of the semantic enhancement feature map of the grass growth status to obtain the pixel-granularity semantic association score topological matrix of the grass growth status; based on the pixel spatial position information of the semantic enhancement feature map of the grass growth status, performing spatial attention attenuation modulation on the pixel-granularity semantic association score topological matrix of the grass growth status to obtain the grass growth status semantic association space soft constraint score topological feature matrix; based on the grass growth status semantic association space soft constraint score topological feature matrix, performing pixel-granularity spatial enhancement processing on the semantic enhancement feature map of the grass growth status to obtain the multi-dimensional enhanced feature map of the grass growth status.

2. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 1, characterized in that: Performing grass growth state feature extraction and semantic enhancement processing on the grass growth state image to obtain a grass growth state semantic enhancement feature map, including: Inputting the forage growth state image into a growth state feature extractor based on a deep separable convolutional neural network model to obtain a forage growth state feature map; The grass growth state feature map is input into a growth state context encoder based on a converter structure to obtain the grass growth state semantic enhancement feature map.

3. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 2, characterized in that: Calculating the pixel-level semantic association score of the forage growth state semantic enhancement feature map to obtain a forage growth state pixel granularity semantic association score topological matrix includes: Performing feature dispersion on the grass growth state semantic enhancement feature map along the channel dimension to obtain a set of grass growth state pixel granularity feature vectors; The semantic association score between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors is calculated to obtain the grass growth state pixel granularity semantic association score topological matrix.

4. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 3, characterized in that: Calculating the semantic association score between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors to obtain the grass growth state pixel granularity semantic association score topological matrix, including: Performing nonlinear mapping on each of the forage growth state pixel granularity feature vectors in the set of forage growth state pixel granularity feature vectors to obtain a set of mapped forage growth state pixel granularity feature vectors; The Poincare distance between any two post-mapping grass growth state pixel granularity feature vectors in the set of post-mapping grass growth state pixel granularity feature vectors is calculated as the semantic association score to obtain the grass growth state pixel granularity semantic association score topological matrix.

5. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 4, characterized in that: Based on the pixel spatial position information of the forage growth state semantic enhancement feature map, the forage growth state pixel granularity semantic association score topology matrix is ​​subjected to spatial attention attenuation modulation to obtain the forage growth state semantic association space soft constraint score topology feature matrix, including: Based on the spatial distance between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors, the grass growth state pixel granularity semantic association score topological matrix is ​​subjected to spatial attention attenuation modulation to obtain a grass growth state semantic association spatial soft constraint score topological matrix; The soft constraint score topology matrix of the semantic association space of the forage growth state is subjected to dilated convolution coding to obtain the soft constraint score topology feature matrix of the semantic association space of the forage growth state.

6. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 5, characterized in that: Based on the spatial distance between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors, the grass growth state pixel granularity semantic association score topology matrix is ​​subjected to spatial attention attenuation modulation to obtain a grass growth state semantic association space soft constraint score topology matrix, including: Calculating the natural exponential function value of the Euclidean distance between any two grass growth state pixel granularity feature vectors in the set of grass growth state pixel granularity feature vectors to obtain a grass growth state pixel granularity spatial distance topological matrix; The grass growth status pixel granularity spatial distance topological matrix is ​​multiplied by a preset attenuation modulation constant to obtain a spatial attention attenuation modulation matrix, and the grass growth status pixel granularity semantic association score topological matrix is ​​calculated and divided by the spatial attention attenuation modulation matrix by position points to obtain the grass growth status semantic association spatial soft constraint score topological matrix.

7. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 6, characterized in that: Based on the forage growth state semantic association space soft constraint score topological feature matrix, the forage growth state semantic enhancement feature map is subjected to pixel granularity spatial enhancement processing to obtain the forage growth state multi-dimensional enhancement feature map, including: Inputting the set of the grass growth state pixel granularity feature vectors and the grass growth state semantic association space soft constraint score topological feature matrix into a graph convolutional coding module to obtain a set of context-related enhanced grass growth state pixel granularity feature vectors; The set of the context-related enhanced grass growth state pixel granularity feature vectors is reshaped to obtain the grass growth state multi-dimensional enhanced feature map.

8. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 7, characterized in that: Based on the multi-dimensional enhanced feature map of the forage growth state, the category label of the growth characteristic is determined, including: The multi-dimensional enhanced feature map of the forage growth state is input into a classifier-based growth characteristic identifier to obtain a category label of the growth characteristic.

9. The intelligent fertilization control method for improving specific traits of forage grass based on forage grass growth characteristics according to claim 8, characterized in that: Based on the category label of the growth characteristics and the growth characteristics-soil fertilizer amount recommendation table, a fertilization strategy is determined, including: Based on the category label of the growth characteristic, extracting the soil fertilizer amount recommendation result from the growth characteristic-soil fertilizer amount recommendation table; The fertilization strategy is determined based on the comparison between the soil fertilizer recommendation result and the current soil fertilizer status.

Images