Method and related device for estimating grassland grazing intensity based on BMogLSTM-KAN model

Abstract

The present application discloses a method and related device for estimating grassland grazing intensity based on the BMogLSTM-KAN model, which relates to the technical field of grassland management. In this method, the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated are obtained, and then based on the trained grassland grazing intensity estimation model, the grazing intensity estimation result of the area to be estimated is obtained according to the above data of the area to be estimated; specifically, the grassland grazing intensity estimation model of the present application is constructed based on the BMogLSTM-KAN model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism. By using this improved model to fully learn the relationship between the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, grazing activity factors and grazing intensity, it can be used to accurately estimate the grassland grazing intensity.

Description

Technical Field

[0001] The present application relates to the technical field of grassland management, and in particular, to a method and related device for estimating grassland grazing intensity based on the BMogLSTM-KAN model. Background Art

[0002] Grasslands are one of the most widely distributed ecosystems on Earth, covering approximately 40% of the land surface area. Grassland ecosystems are an important part of natural ecosystems, providing a wide range of biodiversity and ecosystem services for humans, and playing an important role in aspects such as pasture feed, water conservation, windbreak and sand fixation, climate regulation, soil and water conservation, and food security. Grazing is the main utilization method of grasslands, and its output accounts for approximately 30% of the global meat supply. With the development of grassland animal husbandry, overgrazing has occurred in some areas of Inner Mongolia, and the number of livestock such as cattle and sheep far exceeds the carrying capacity of the grasslands. The contradiction between the grasslands and livestock has become increasingly prominent. The grassland degradation caused by grazing not only affects the sustainable development of animal husbandry but also leads to the weakening of ecosystem functions, posing a huge threat to the local and surrounding ecological security. Therefore, scientifically understanding grazing intensity is crucial for the sustainable development of grassland ecosystems.

[0003] Grazing intensity (GI) refers to the intensity of grazing utilization, that is, the number of grazing livestock per unit area within a certain period of time, and is an important indicator reflecting the degree of grassland utilization. Moderate grazing can improve grassland productivity, biodiversity, and ecosystem stability, while overgrazing will have a negative impact on biodiversity and service functions such as carbon and nitrogen nutrient cycling, leading to ecological problems such as grassland degradation. Quantifying grassland grazing intensity is of great significance for adjusting and improving grassland ecosystem management, improving grassland ecological quality, and ensuring China's ecological security. How to accurately estimate grassland grazing intensity is an urgent problem to be solved currently. Summary of the Invention

[0004] The purpose of the present application is to provide a method and related device for estimating grassland grazing intensity based on the BMogLSTM-KAN model, which can accurately estimate the grassland grazing intensity.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In a first aspect, the present application provides a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model, including the following steps:

[0007] Obtain the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated.

[0008] Based on the trained grassland grazing intensity estimation model, according to the retrieved result of aboveground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factor of the area to be estimated, the grazing intensity estimation result of the area to be estimated is obtained; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism.

[0009] Optionally, before obtaining the retrieved result of aboveground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factor of the area to be estimated, the grassland grazing intensity estimation method based on the BMogLSTM-KAN model further includes the following steps:

[0010] Construct a grazing intensity estimation sample data set; the grazing intensity estimation sample data set includes a number of grazing intensity estimation samples; the grazing intensity estimation samples include the retrieved result of aboveground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factor as independent variables, and the grassland grazing intensity as the dependent variable.

[0011] Based on the BMogLSTM-KAN model, an initial grassland grazing intensity estimation model is constructed.

[0012] Based on the grazing intensity estimation sample data set, the initial grassland grazing intensity estimation model is trained to obtain a trained grassland grazing intensity estimation model.

[0013] Optionally, constructing the grazing intensity estimation sample data set specifically includes the following steps:

[0014] Conduct on-site investigations and records in the study area to obtain the grassland grazing situation in the study area; the grassland grazing situation includes the number of grazing livestock, the types of grazing livestock, the pasture area, and the grazing time; the study area is a grassland area of the same type as the area to be estimated.

[0015] According to the grassland grazing situation in the study area, calculate the grassland grazing intensity in the study area.

[0016] Conduct on-site measurements in the study area to obtain the retrieved result of aboveground biomass in the study area.

[0017] Obtain the remote sensing data, the terrain data, the meteorological data, and the grazing activity factor of the study area.

[0018] Take the retrieved result of aboveground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factor of the study area as independent variables, and the grassland grazing situation in the study area as the dependent variable to obtain a grazing intensity estimation sample.

[0019] Repeat the above process in several study areas to obtain a grazing intensity estimation sample dataset including several grazing intensity estimation samples.

[0020] Optionally, calculate the grassland grazing intensity of the study area during the grazing time according to the following formula:

[0021]

[0022] Wherein, GI is the grassland grazing intensity of the study area, SU is the sheep unit value of the study area, and the sheep unit is based on sheep, converting different grazing livestock species in the study area into sheep units, and GA is the pasture area.

[0023] Optionally, the BMogLSTM-KAN model includes a batch attention module, a MogLSTM model, and a KAN mechanism; the batch attention module includes multi-head attention, layer normalization, and a feed-forward network; the MogLSTM model is an improved model obtained by adding an interactive mechanism on the basis of the traditional LSTM model.

[0024] Optionally, the grassland grazing intensity estimation method based on the BMogLSTM-KAN model further includes the following steps:

[0025] According to the grazing intensity estimation result of the area to be estimated, draw a spatial distribution map of the grazing intensity of the area to be estimated, and determine the overall trend of the grazing intensity within the area to be estimated.

[0026] In a second aspect, the present application provides a grassland grazing intensity estimation system based on the BMogLSTM-KAN model, including:

[0027] An in-region data acquisition module for acquiring the inversion result of above-ground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated.

[0028] A regional grazing intensity estimation module for obtaining the grazing intensity estimation result of the area to be estimated based on the trained grassland grazing intensity estimation model according to the inversion result of above-ground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factors of the area to be estimated; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism.

[0029] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the grassland grazing intensity estimation method based on the BMogLSTM-KAN model described above.

[0030] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the grassland grazing intensity estimation method based on the BMogLSTM-KAN model described above.

[0031] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the grassland grazing intensity estimation method based on the BMogLSTM-KAN model described above.

[0032] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0033] The present application provides a grassland grazing intensity estimation method and related devices based on the BMogLSTM-KAN model. In this method, the inversion result of above-ground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated are obtained, and then based on the trained grassland grazing intensity estimation model, according to the inversion result of above-ground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated, the grazing intensity estimation result of the area to be estimated is obtained; specifically, the grassland grazing intensity estimation model of the present application is a model constructed based on the BMogLSTM-KAN model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism. This fusion not only retains the high efficiency of MogLSTM in sequence learning, but also strengthens the model's learning ability of the relationship between different samples within a small batch through the batch attention mechanism. In addition, the introduction of the KAN mechanism also provides a new expression way of the activation function for the model, enabling the weight parameters themselves to participate in the learning process, further enhancing the model's approximation ability for complex functions. Using the improved model to fully learn the relationship between the inversion result of above-ground biomass, remote sensing data, terrain data, meteorological data, grazing activity factors and grazing intensity can accurately estimate the grassland grazing intensity, which helps to provide an important reference basis for the scientific protection and sustainable management of grasslands. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0035] Figure 1 It is a flowchart of a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0036] Figure 2 It is a flowchart of constructing and training a grassland grazing intensity estimation model in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0037] Figure 3 It is a schematic structural diagram of a traditional LSTM model in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0038] Figure 4 It is a schematic structural diagram of the MogLSTM model adopted in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0039] Figure 5 It is a schematic structural diagram of the BMogLSTM-KAN model in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0040] Figure 6 It is a schematic structural diagram of the batch attention module in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0041] Figure 7 It is a schematic network structure diagram of the KAN mechanism in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0042] Figure 8 It is a schematic diagram of the result of residual analysis of the BMogLSTM-KAN model in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0043] Figure 9 It is a box plot of the regression performance of three algorithms on the same dataset in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0044] Figure 10 This is a schematic diagram of the fitting effect of the evaluation index of the BMogLSTM-KAN model in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0045] Figure 11 This is a spatial distribution map of the grazing intensity of the Inner Mongolia temperate typical grassland in a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0046] Figure 12 This is a schematic diagram of the functional modules of a grassland grazing intensity estimation system based on the BMogLSTM-KAN model provided by an embodiment of the present application.

[0047] Figure 13 This is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0049] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0050] A method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided by an embodiment of the present application can be executed by a computer device, specifically, it can be executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server, and includes the following steps:

[0051] A1. Obtain the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated.

[0052] A2. Based on the trained grassland grazing intensity estimation model, obtain the grazing intensity estimation result of the area to be estimated according to the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism.

[0053] In an exemplary embodiment, before step A1, the method for estimating grassland grazing intensity based on the BMogLSTM-KAN model further includes a process of constructing and training a grassland grazing intensity estimation model, as Figure 2 shown, and this process includes the following steps:

[0054] B1. Construct a grazing intensity estimation sample data set; the grazing intensity estimation sample data set includes a number of grazing intensity estimation samples; the grazing intensity estimation samples include the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors as independent variables, and the grassland grazing intensity as the dependent variable. In this embodiment, step B1 of constructing the grazing intensity estimation sample data set specifically includes the following steps:

[0055] B11. Conduct on-site investigations and records in the study area to obtain the grassland grazing situation in the study area; the grassland grazing situation includes the number of grazing livestock, types of grazing livestock, pasture area, and grazing time; the study area is a grassland area of the same type as the area to be estimated.

[0056] B12. Calculate the grassland grazing intensity in the study area according to the grassland grazing situation in the study area. In this embodiment, the grassland grazing intensity in the study area during the grazing time is calculated according to the following formula:

[0057]

[0058] where GI is the grassland grazing intensity in the study area, SU is the sheep unit value in the study area, and the sheep unit is based on sheep, converting different types of grazing livestock in the study area into sheep units, and GA is the pasture area.

[0059] B13. Conduct on-site measurements in the study area to obtain the inversion result of aboveground biomass in the study area.

[0060] B14. Obtain the remote sensing data, terrain data, meteorological data, and grazing activity factors of the study area.

[0061] B15. Take the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors in the study area as independent variables, and the grassland grazing situation in the study area as the dependent variable to obtain a grazing intensity estimation sample.

[0062] B16. Repeat the process of steps B11 to B15 in several study areas to obtain a grazing intensity estimation sample data set including a number of grazing intensity estimation samples.

[0063] B2. Based on the BMogLSTM-KAN model, an initial grassland grazing intensity estimation model is constructed. The BMogLSTM-KAN model includes a batch attention module, a MogLSTM model, and a KAN mechanism; the batch attention module includes multi-head attention, layer normalization, and a feed-forward network; the MogLSTM model is an improved model obtained by adding an interactive mechanism to the traditional LSTM model.

[0064] B3. Based on the grazing intensity estimation sample data set, the initial grassland grazing intensity estimation model is trained to obtain a trained grassland grazing intensity estimation model.

[0065] To facilitate understanding of the overall trend of the grazing intensity in the area to be estimated, the grassland grazing intensity estimation method based on the BMogLSTM-KAN model provided in this embodiment further includes the following steps:

[0066] A3. According to the grazing intensity estimation result of the area to be estimated, a spatial distribution map of the grazing intensity of the area to be estimated is drawn to determine the overall trend of the grazing intensity in the area to be estimated.

[0067] The following uses a specific example to illustrate the effectiveness of the grassland grazing intensity estimation method based on the BMogLSTM-KAN model provided in the above embodiments of the present application. First, the study area is selected. The study area in this example is located in the temperate typical grassland of Inner Mongolia, between 43°03′ and 46°73′ north latitude and between 113°46′ and 119°21′ east longitude, with an average altitude of 1076 m. The area is mainly semi-arid, with an average annual temperature of 2.5 °C and an average annual precipitation of about 271 mm, of which 80% is concentrated in the growing season from May to August. According to the Chinese soil classification method, the soil type in this area is sandy loam chestnut. The study area is mainly a typical grassland ecosystem type, with a vegetation type of herbaceous plants, and the dominant species include Allium mongolicum, Anemarrhena asphodeloides, Festuca ovina, Stipa grandis, Leymus chinensis, etc. The growing season is from April to August. The main types of grazing livestock are Simmental cattle and large-tailed sheep.

[0068] After the study area is determined, the known data of the study area need to be collected first for training the deep learning model, including the following steps:

[0069] (1) The AGB data and GI data are obtained through on-site measurement in the study area:

[0070] Field data collection work in the study area was carried out from July 1 to August 1, 2023, and two types of measured data, GI and grassland AGB, were collected. The GI data was investigated and recorded by means of a GPS toolbox for the number, type, pasture area, grazing time, etc. of grazing livestock.

[0071] The grassland grazing intensity refers to the number of livestock per unit area within a certain period. Based on the field-collected data, the measured grassland GI is calculated. To make GI comparable among different enclosures, in this example, the numbers of various livestock are uniformly converted into sheep units (SU). Among them, 1 sheep is equal to 1 SU, 1 cow is equal to 4 SU, and 1 horse is equal to 5 SU. The calculation method of GI is as follows:

[0072]

[0073] where GI is the grazing intensity (head / ha); SU is the sheep unit (head); GA is the grazing area (ha). The measured GI data is used for the training and testing of the subsequent estimation model.

[0074] Regarding the sampling of above-ground biomass (AGB), typical temperate grassland communities with flat terrain and relatively uniform growth conditions were selected to set up plots (a total of 75 plots and 225 sample AGB data were collected). The distance between plots (30m×30m) is not less than 5km. Three quadrats (1m×1m) were randomly selected in each plot. The grassland AGB was cut at ground level, and the longitude, latitude, and sampling time of the quadrat were recorded. The samples were bagged, numbered, and taken back to the laboratory and dried to a constant weight in an oven at 65°C to obtain the dry weight of AGB for each quadrat. Finally, the average value of AGB of the three quadrats was taken as the plot data to obtain the measured ground data of grassland AGB.

[0075] AGB can usually reflect the grass potential of the pasture and is an important indicator for evaluating grassland GI. The inversion of grassland AGB is very mature. In this study, the random forest model was used to invert AGB with the field-measured AGB data. The accuracy of the AGB inversion model, R 2 = 0.94, RMSE = 7.50. The AGB inversion results, as one of the important influencing factors of the grazing intensity estimation model, are used to study the grassland grazing intensity.

[0076] (2) The network obtains the remote sensing data, topographic data, meteorological data, and grazing activity factors of the study area:

[0077] Sentinel-2 consists of two complementary satellites (Sentinel-2A / B) launched by the European Space Agency (ESA). The Multi-Spectral Instrument (MSI) on Sentinel-2 has multiple spectral bands in the visible to short-wave infrared (443 - 2190 nm), with high spatial resolution and a revisit period of 5 days. The Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are the most widely used vegetation indices in vegetation monitoring, which can reflect the vegetation condition of pastures and have a negative correlation with GI. In this example, Sentinel-2 MSI image products are called on the GEE platform, and preprocessing such as QA cloud removal, image fusion, and calculation of vegetation indices is performed on the products.

[0078] The Digital Elevation Model (DEM) data is from the European Space Agency (ESA), with a resolution of 30 meters and a spatial reference of GCS_WGS_1984. The livestock grazing process varies with terrain, and it is easier to graze on gentler slopes. The ArcGIS software is used to process the DEM data to obtain slope and aspect data.

[0079] Meteorological data includes air temperature and precipitation data, which come from the National Oceanic and Atmospheric Administration (NOAA) of the United States. Air temperature and precipitation affect the growth of grassland vegetation and have a certain correlation with grazing intensity. Kriging interpolation is used for meteorological data with the air temperature and precipitation station data in cooperation with the DEM data to obtain the monthly average air temperature data and monthly average precipitation data of the study area.

[0080] Considering that the high-coverage grasslands near water sources are more convenient for herders to graze and for livestock to drink water (livestock need to drink water in addition to grazing during the grazing process), at the same time, there is a strong correlation between livestock grazing and the location of settlements. Therefore, the distance to the river and the distance to the settlement are selected as grazing activity factors. The river data comes from Open StreetMap, and the settlement data comes from the National Tibetan Plateau Data Center of Sciences.

[0081] Before the specific use of the known data obtained through the above steps, corresponding preprocessing should be done on the obtained data to generate measured GI data and AGB, NDVI, EVI, DEM, slope, aspect, monthly average air temperature, monthly average precipitation, distance to the river DTR, and distance to the settlement DTS. The resolution and data format of each data are shown in Table 1.

[0082] Table 1 Resolution and data format of each data obtained in this example

[0083]

[0084] Next, the model is constructed, and the BMogLSTM-KAN regression model is established. Using the GI data of the measured points as the dependent variable, and the corresponding AGB, NDVI, EVI, climate, terrain, and grazing activity data as the independent variables, which serve as the input data for the model. The model parameters are adjusted, and the model is trained and tested. The specific steps are as follows:

[0085] (1) Construction and training of the BMogLSTM-KAN model:

[0086] As a type of recurrent neural network (RNN), LSTM can capture long-term dependencies in data. LSTM can effectively avoid the problem of gradient explosion or disappearance in RNNs. Figure 3 Shows the node connection method of a typical LSTM cell, where c t-1 and h t-1 represent the cell state and output of the previous time step, respectively. In the LSTM model, there are three gates, including the forget gate f, the input gate i, the output gate o, and the internal cell state c. Denote the weight matrix, recurrent weight, and bias as W1 = [W f , W i , W c , W o , W2 = [W hf , W hi , W hc , W ho , and b1 = [b f , b i , b c , b o , then the feedforward process of LSTM at time t is described as:

[0087]

[0088] where σ and tanh are activation functions, and ⊙ represents multiplication.

[0089] The Mogrifier LSTM model (abbreviated as MogLSTM) is based on LSTM and significantly improves performance in processing long sequence data by adding an interactive mechanism to fully interact the input unit and the hidden unit of the current state. In the storage unit of the LSTM network, there is a lack of deep interaction between the previous hidden state h t-1 and the current input x t . They only have a simple interaction when entering the unit. Melis et al. believe that the independence of h t-1 and x t may limit the generalization ability of the LSTM network. Melis et al. improved the LSTM model by introducing a gating mechanism called "interaction gate". This mechanism ensures the hidden state ht-1 and the input x t fully interact before entering the storage unit, thus improving the learning efficiency of the model. For example Figure 4 shows the MogLSTM model with a 2-round interaction mechanism. If the process of updating data in the LSTM model is represented as a function:

[0090] LSTM(x t , h t-1 , c t-1 ) = (h t , c t ).

[0091] Then the mechanism for MogLSTM to update information can be expressed as:

[0092]

[0093] where represents the output of x t and h t-1 after the interaction gate.

[0094] In this example, the MogLSTM model is built in the Python 3.8 environment. Further, on the basis of MogLSTM, a batch attention module (Batch Attention Module) and the weight activation function of KAN (Kolmogorov - Arnold Networks) are introduced, as Figure 5 shown. This fusion not only retains the high efficiency of MogLSTM in sequence learning, but also enhances the model's learning ability of the relationships between different samples within a mini - batch through the batch attention mechanism. Specifically, the batch attention module allows the model to not only focus on the features of individual samples when processing data, but also capture the mutual relationships between samples, thus achieving a more comprehensive feature representation. In addition, the introduction of KAN also provides a new way of expressing the activation function (spline function as the weight activation function) for the model, enabling the weight parameters themselves to participate in the learning process and further improving the model's approximation ability for complex functions. This BMogLSTM - KAN that combines batch attention and the weight activation function is not only innovative in theory, but also shows excellent performance in practical applications.

[0095] 1) Batch attention module

[0096] During the information propagation process of MogLSTM, the possible information propagation between different training samples within each mini-batch is ignored. The relationships between different data samples are diverse and complex, and effectively extracting the relationships between samples is crucial for improving the learning efficiency of the model. In the prior art, by leveraging the ability of the Transformer encoder to automatically learn sample relationships, a simple and effective batch attention module is proposed to promote robust representation learning. The batch attention module consists of multi-head attention, layer normalization, and a feed-forward network. In the batch attention module, the Transformer encoder is used to process the batch dimension of the input data. The specific process is as Figure 6 shown. Suppose is the sequence of input features, where B is the length of the sequence, i.e., the batch size, and d model is the dimension of the features.

[0097] The framework of the batch attention module is specifically manifested as introducing the Transformer network and leveraging the structural advantages of the Transformer network to explore the relationships between samples. Through the proposed batch attention module, we can transmit messages along different samples within each mini-batch. From an optimization perspective, all samples can contribute to the learning of any sample. Therefore, the batch attention module implicitly enriches the current training data based on the samples of the entire mini-batch. In this way, the attention mechanism in the batch attention module becomes the cross-attention between different samples in each mini-batch.

[0098] 2) KAN mechanism

[0099] Inspired by the Kolmogorov-Arnold representation theorem, a neural network structure, Kolmogorov-Arnold Networks (KANs), is proposed in the prior art as an alternative to multi-layer perceptrons (MLPs). The core idea of KANs is to replace the fixed activation function at the nodes (neurons) in traditional neural networks with a learnable activation function on the edges (weights), and place these functions on the edges (weights) of the network instead of at the nodes (neurons) in traditional MLPs. KAN does not have a linear weight matrix, and each of its weight parameters is replaced by a learnable one-dimensional function parameterized as a spline function.

[0100] Since there is an MLP outside KAN and a spline curve inside, KAN can not only learn the combinatorial structure (external degrees of freedom), but also approximate univariate functions well (internal degrees of freedom), and learn functions more accurately, such as Figure 7As shown, KAN replaces the fixed activation function with a learnable activation function (“weight”) on the edge. The unique framework of LAN enables KAN to approximate non - linear functions more effectively while maintaining a smaller network scale. KAN addresses these limitations by combining a learnable spline function with each weight, thus providing greater flexibility and accuracy in learning non - linear relationships. Compared with MLP, this unique architecture offers a faster neural scaling law and requires fewer parameters.

[0101] The Kolmogorov - Arnold theorem is specifically manifested as that a multivariate continuous on a bounded domain can be represented by a finite combination of univariate continuous functions and additive binary operations. The Kolmogorov - Arnold theorem is defined by a KAN of the form [n, 2n + 1, 1]. Here, a general deeper KAN architecture is defined using [n1,..., nL+1], where L represents the number of layers of the KAN.

[0102]

[0103] All operations in the formula are differentiable. Therefore, KAN can be trained by backpropagation. The B MogLSTM - KAN model is constructed in a Python 3.8 environment, and the training process and GI estimation process of this model are the same as those of the MogLSTM model.

[0104] (2) Model performance evaluation and comparison:

[0105] In this example, the coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute error (MAE) are used as evaluation indicators for the accuracy of the grassland grazing intensity estimation model. R 2 reflects the fitting ability between the estimated value and the measured value, RMSE reflects the degree of dispersion between the estimated value and the measured value, and MAE reflects the average level of the difference between the estimated value and the measured value. The unit of RMSE is head / ha -1 , which is the same as the unit of the measured value of grassland grazing intensity. The calculation formulas are as follows:

[0106]

[0107] where n is the number of samples, y i represents the measured value of GI, is the average value of GI, is the estimated value of GI.

[0108] In this example, it is further improved based on the MogLSTM model, and the BMogLSTM-KAN model is developed. For the residual analysis of the BMogLSTM-KAN model, a total of 197 sample points are used during the analysis process, and 80% of the data is divided into the training set, and the remaining 20% is used as the test set. As Figure 8 shown, the distance between the measured value and the estimated value represents the residual, and the larger the distance between two points, the larger the residual. The improved BMogLSTM-KAN performs excellently in estimating the grazing intensity. The residual of BMogLSTM-KAN is overall the smallest, and the estimated GI curve is highly consistent with the actual GI measured value, showing excellent fitting performance.

[0109] To verify the improvement effect of the BMogLSTM-KAN model in this embodiment, the fitting effects of BMogLSTM, MogLSTM-KAN, and BMogLSTM-KAN on the same data set are compared. To further evaluate the stability of the model and reduce the random fluctuations during the training process, we adopt the method of 10-cycle training to evaluate the model performance. Record the performance of the model on the test set, including R 2 , RMSE, MAE, and the loss value. By comparing the regression losses of different algorithms, it can be found that the average regression loss of the BMogLSTM-KAN algorithm is the lowest, specifically BMogLSTM-KAN(0.80) < BMogLSTM(1.17) < MogLSTM-KAN(1.32). Figure 9 The box plot of the regression performance of the three algorithms on the same data set is shown. As shown in the figure, the upper and lower limits of the four indicators of BMogLSTM and MogLSTM-KAN have a large gap, and the gap between the upper and lower limits of RMSE exceeds 0.55, indicating that the model is not stable enough. The four indicators of the BMogLSTM-KAN algorithm are relatively evenly distributed, and the gap between the first quartile and the third quartile is also relatively small, so the stability is better.

[0110] Through the comparison between different models, in this example, the task of estimating the grassland grazing intensity is completed using the BMogLSTM-KAN model. The input factors for model establishment include a total of 10 indicators such as aboveground biomass, vegetation index, meteorological factors, terrain factors, and grazing activity factors, specifically AGB, NDVI, EVI, monthly average temperature, monthly average precipitation, DEM, slope, aspect, and DTR and DTS. The input data is divided into a training data set and a test data set according to 8:2, and R 2 , RMSE, and MAE are used as evaluation indicators for the accuracy of the grazing intensity estimation model. As Figure 10As shown in the figure, the BMogLSTM-KAN model with the added batch attention module and KAN mechanism has the best fitting effect, with a coefficient of determination of 0.93 and a root mean square error of 0.68.

[0111] Finally, using the trained grassland grazing intensity estimation model obtained from the above steps, based on the aboveground biomass inversion result AGB, remote sensing data (NDVI, EVI), terrain data (DEM, slope, aspect), meteorological data (monthly average temperature, monthly average precipitation), and grazing activity factors (DTR, DTS) of the area to be estimated, the grazing intensity estimation result of the area to be estimated is obtained. As Figure 11 shown in the figure, it is the grazing intensity estimation result of the Inner Mongolia temperate typical steppe by the BMogLSTM-KAN model. The overall trend of the drawn grazing intensity spatial distribution map of the Inner Mongolia temperate typical steppe is as follows: from the northeast to the southwest of the study area, GI shows a gradually increasing trend. And the GI range estimated by BMogLSTM-KAN is 0 - 8.51 head / ha. In addition, the GIs estimated by the BMogLSTM-KAN model all show that the places with high GI are greater than those with low GI.

[0112] Using the estimation result of the grassland grazing intensity by the BMogLSTM-KAN model, the results are statistically analyzed by county (banner) boundaries. As shown in Table 2, the maximum grazing intensity of each county (banner) in the Inner Mongolia temperate typical steppe exceeds 8.2 head / ha, indicating that overgrazing exists in each county. Especially in Xilingol City and Abag Banner, the average grazing intensities are 5.34 head / ha and 5.92 head / ha respectively, and they have a small standard deviation, reflecting that these two areas not only have a high grazing intensity but also a wide distribution range.

[0113] Table 2 Grazing intensity of each county (banner) in the Inner Mongolia temperate typical steppe

[0114] Name Minimum value Maximum value Average value Standard deviation Dong Ujimqin Banner 0 8.337016106 4.45583 2.2152212 Xi Ujimqin Banner 0 8.289620399 3.981979 2.441014 Abag Banner 0 8.505903244 5.921941 1.5667299 Xilinhot City 0 8.301514626 5.339965 1.909221 Linxi County 0 8.202166557 2.361669 2.6007649 Keshiketeng Banner 0 8.341110229 2.837717 2.1454055

[0115] In this example, using the improved BMogLSTM-KAN model, based on the measured grazing intensity data and grazing impact factors (AGB, NDVI, EVI, monthly average temperature, monthly average precipitation, DEM, slope, aspect, DTR, DTS), the grazing intensity of the Inner Mongolia temperate typical grassland is estimated, and the grassland grazing intensity is quantitatively studied, realizing the estimation of the grazing intensity of the Inner Mongolia temperate typical steppe. We analyzed the performance of the model in grazing intensity estimation and found that the deep learning model BMogLSTM-KAN showed good results in estimating GI, R 2= 0.93, RMSE = 0.68. This model is the result of optimizing and improving the MogLSTM model by introducing an advanced batch attention module and the KAN mechanism. In the comparative evaluation of the model optimization process, this model demonstrated the accuracy of its estimation and the robustness of its performance. Finally, using 10 key grazing impact factors as input data, four models were respectively used to estimate the grazing intensity of the typical temperate steppe in Inner Mongolia, depicting the spatial distribution map of the grazing intensity of the typical temperate steppe in Inner Mongolia and quantitatively evaluating the distribution of the grazing intensity in the study area. The research results can provide an important reference for the scientific protection and sustainable management of grasslands in Inner Mongolia.

[0116] Based on the same inventive concept, an embodiment of the present application also provides a system for implementing the above-mentioned method for estimating the grazing intensity of grasslands based on the BMogLSTM-KAN model. The implementation solutions provided by this system to solve problems are similar to those recorded in the above method. Therefore, the specific limitations in one or more system embodiments provided below can refer to the limitations on the method for estimating the grazing intensity of grasslands based on the BMogLSTM-KAN model in the above text and will not be elaborated here.

[0117] In an exemplary embodiment, as Figure 12 shown, a system for estimating the grazing intensity of grasslands based on the BMogLSTM-KAN model is provided, including the following modules:

[0118] An in-region data acquisition module, configured to acquire the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated.

[0119] A regional grazing intensity estimation module, configured to obtain an estimation result of the grazing intensity of the area to be estimated based on the trained grassland grazing intensity estimation model according to the inversion result of aboveground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factors of the area to be estimated; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism.

[0120] Of course, Figure 12 the architecture shown Figure 12 is only exemplary. When implementing different functions, one or at least two components in the system shown

[0121] In an exemplary embodiment, a computer device is provided. This computer device can be a server or a terminal, and its internal structure diagram can be as Figure 13As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it can implement a method for estimating grassland grazing intensity based on the BMogLSTM-KAN model provided in the above method embodiments.

[0122] Those skilled in the art can understand that Figure 13 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0123] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0124] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0125] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0126] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0127] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0128] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0129] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0130] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for estimating the grassland grazing intensity based on the BMogLSTM-KAN model, characterized in that, Including: Obtaining the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated; Based on the trained grassland grazing intensity estimation model, obtaining the grazing intensity estimation result of the area to be estimated according to the aboveground biomass inversion result, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factors of the area to be estimated; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism; Among them, the batch attention module includes multi-head attention, layer normalization, and a feed-forward network; the MogLSTM model is an improved model obtained by adding an interactive mechanism on the basis of the traditional LSTM model; the KAN mechanism is a neural network mechanism based on the Kolmogorov-Arnold representation theorem and using a spline function as an activation function.

2. The method for estimating the grassland grazing intensity based on the BMogLSTM-KAN model according to claim 1, wherein, Before obtaining the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the area to be estimated, the grassland grazing intensity estimation method based on the BMogLSTM-KAN model further includes: Constructing a grazing intensity estimation sample data set; the grazing intensity estimation sample data set includes a number of grazing intensity estimation samples; the grazing intensity estimation sample includes the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors as independent variables, and the grassland grazing intensity as the dependent variable; Based on the BMogLSTM-KAN model, constructing an initial grassland grazing intensity estimation model; Based on the grazing intensity estimation sample data set, training the initial grassland grazing intensity estimation model to obtain a trained grassland grazing intensity estimation model.

3. The method for estimating the grassland grazing intensity based on the BMogLSTM-KAN model according to claim 2, characterized in that Constructing a grazing intensity estimation sample data set specifically includes: Conducting on-site investigations and records in the study area to obtain the grassland grazing situation in the study area; the grassland grazing situation includes the number of grazing livestock, the types of grazing livestock, the pasture area, and the grazing time; the study area is a grassland area of the same type as the area to be estimated; Calculating the grassland grazing intensity in the study area according to the grassland grazing situation in the study area; Conducting on-site measurements in the study area to obtain the inversion result of aboveground biomass in the study area; Obtaining the remote sensing data, terrain data, meteorological data, and grazing activity factors of the study area; Taking the inversion result of aboveground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors of the study area as independent variables, and taking the grassland grazing situation in the study area as the dependent variable, to obtain a grazing intensity estimation sample; Repeating the above process in several study areas to obtain a grazing intensity estimation sample data set including a number of grazing intensity estimation samples.

4. The method for estimating grassland grazing intensity based on the BMogLSTM-KAN model according to claim 3, characterized in that Calculating the grassland grazing intensity in the study area during the grazing time according to the following formula: Among them, GI is the grassland grazing intensity of the study area, SU is the sheep unit value of the study area, and the sheep unit is based on sheep, converting different grazing livestock species in the study area into sheep units, and GA is the pasture area.

5. The method for estimating the grassland grazing intensity based on the BMogLSTM-KAN model according to claim 1, characterized in that, The method for estimating grassland grazing intensity based on the BMogLSTM-KAN model further includes: According to the grazing intensity estimation result of the area to be estimated, a spatial distribution map of the grazing intensity of the area to be estimated is drawn to determine the overall trend of the grazing intensity within the area to be estimated.

6. A grassland grazing intensity estimation system based on the BMogLSTM-KAN model, characterized in that, It includes: The data acquisition module in the area is used to obtain the inversion result of above-ground biomass, remote sensing data, terrain data, meteorological data, and grazing activity factors in the area to be estimated; The regional grazing intensity estimation module is used to obtain the grazing intensity estimation result of the area to be estimated based on the trained grassland grazing intensity estimation model according to the inversion result of above-ground biomass, the remote sensing data, the terrain data, the meteorological data, and the grazing activity factors in the area to be estimated; the grassland grazing intensity estimation model is a model constructed based on the BMogLSTM-KAN model; the BMogLSTM-KAN model is a model obtained by improving the MogLSTM model based on the batch attention module and the KAN mechanism; Among them, the batch attention module includes multi-head attention, layer normalization, and a feed-forward network; the MogLSTM model is an improved model obtained by adding an interactive mechanism on the basis of the traditional LSTM model; the KAN mechanism is a neural network mechanism based on the Kolmogorov-Arnold representation theorem and using a spline function as an activation function.

7. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the method for estimating grassland grazing intensity based on the BMogLSTM-KAN model according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for estimating grassland grazing intensity based on the BMogLSTM-KAN model according to any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for estimating grassland grazing intensity based on the BMogLSTM-KAN model according to any one of claims 1-5.

Images