Tractor field operation ground type real-time identification method and system

Abstract

The invention discloses a tractor field operation ground type real-time identification method and system, and belongs to the technical field of ground and road surface identification. A sensor is used for collecting multi-dimensional data of wheel vibration, soil characteristics and vehicle dynamics, the ground characteristics are comprehensively reflected, the slip rate is estimated in combination with an extended Kalman filtering algorithm, and the slip rate serves as the judgment capability of a ground attachment characteristic enhancement system on the ground attachment state; introducing variational mode decomposition to carry out feature extraction on the wheel vibration signals, forming a multi-dimensional feature matrix with soil characteristics and slip rate, and then carrying out dimensionality reduction by using a unified manifold approximation and projection algorithm; an extreme learning machine optimized by a genetic algorithm is used as an intelligent ground type identification model, data acquisition and ground type identification are completed in a single rotation period of tractor wheels, good real-time performance and accuracy are achieved, the stability and the automation level of agricultural operation are effectively improved, and the working efficiency is improved. And a technical support is provided for intelligent agricultural machine environment perception and operation parameter adaptive control.

Description

Technical Field

[0001] The present invention relates to the field of ground and road surface recognition, and in particular to a method and system for real-time recognition of ground types used by tractors in field operations. Background Art

[0002] In fields such as agriculture and construction, vehicles and machinery experience significant differences in driving performance and operating strategies on different surface types (e.g., muddy, dry, hard, wet, and saline-alkali land). Accurately identifying surface type is crucial for optimizing vehicle power distribution, braking control, and path planning.

[0003] Current technologies for ground and road surface recognition are primarily applicable to conventional road vehicles and some construction machinery. These approaches fall into two main categories: First, recognition methods based on vehicle kinematic and dynamic models rely on precise mathematical modeling, typically requiring detailed construction of the vehicle's dynamic behavior and analysis of the inherent relationships between its motion parameters. These methods suffer from complex modeling, high computational resource consumption, and poor response timeliness. In particular, they suffer from limited portability and generalizability in practical applications. Second, recognition methods based on the extraction of vehicle operational characteristic parameters extract characteristic data from the vehicle during operation (such as acceleration, vibration, and speed) and construct a mapping between these characteristic parameters and ground surface type. However, these methods have limited adaptability to complex, unstructured ground surfaces. In agricultural operations, ground conditions vary significantly, including soil looseness, moisture content, surface irregularities, and other variables, making them difficult for traditional road vehicle recognition methods to accurately adapt.

[0004] As a core power machine in modern agriculture, tractors are often used to pull a variety of work implements in complex terrain and poor road conditions. Their operating state is highly dependent on surface conditions. The type and condition of the field surface directly affects tire adhesion, significantly impacting the tractor's traction efficiency and driving performance, and is a crucial factor that cannot be ignored in the operating environment. Uneven, soft, or slippery surfaces can weaken the tractor's chassis' adhesion, reducing vehicle stability during operation and even leading to loss of control, potentially creating safety hazards such as skidding and rollover.

[0005] Different soil types exhibit significant differences in physical structure and mechanical properties, which in turn influence vehicle behavior on them, such as wheel speed, vehicle speed, and driving resistance. Therefore, accurately identifying the current ground type is crucial for properly adjusting operational parameters (such as driving speed and implement posture). However, due to the frequent changes and complex distribution of field ground conditions, understanding ground type information during operations often lacks real-time and accuracy.

[0006] Based on the above problems, a method that can quickly and accurately identify ground types is proposed to provide an intelligent control basis for the tractor operation system. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for real-time identification of ground type for tractor field operations, so as to solve the problems in the background technology.

[0008] To achieve the above object, the present invention provides a method for real-time identification of ground type during tractor field operation, comprising the following steps:

[0009] S1. Obtain wheel vibration signals, soil characteristics, wheel speed, and ground speed data through sensors;

[0010] S2. Construct state variables based on wheel speed and ground speed, and recursively estimate the slip rate using an extended Kalman filter as a ground adhesion feature.

[0011] S3. Perform variational modal decomposition on the wheel vibration signal to obtain the intrinsic mode function, extract the statistical characteristics of each mode, and combine the soil characteristics and slip rate to form a multidimensional feature matrix;

[0012] S4, using UMAP to perform nonlinear dimensionality reduction on the multidimensional feature matrix to obtain a low-dimensional feature matrix;

[0013] S5. Input the low-dimensional feature matrix into the extreme learning machine model optimized by the genetic algorithm for classification reasoning, and output the ground type recognition result.

[0014] Preferably, in S1, the soil properties include soil moisture and electrical conductivity.

[0015] Preferably, in S2, the state variable is expressed as:

[0016] x=[v,s] T ;

[0017] Where v is the ground speed and s is the slip rate, which can be expressed as:

[0018]

[0019] Where r is the wheel diameter and w is the wheel angular velocity.

[0020] Preferably, in S2, the state transition model of the extended Kalman filter is:

[0021] X k =f(X k-1 ,u k-1 )+W k ;

[0022] Where, X kis the current state, f(·) is the evolution function of the previous moment, W k is the process noise;

[0023] The observation model is:

[0024] Z k =h(X k )+V k ;

[0025] Where Z k is the observed value, h(X k ) is the state-to-observation mapping function, V k is the observation noise;

[0026] The slip rate is estimated through the state transfer update and observation update process of the extended Kalman filter.

[0027] Preferably, the specific steps of S3 are as follows:

[0028] 1) The collected wheel vibration signal is subjected to variational mode decomposition using VMD to obtain the intrinsic mode function {u k (t)}; The goal of variational mode decomposition is to minimize the sum of the bandwidths of each mode, which can be expressed as:

[0029]

[0030] In the formula, {u k (t)} is the kth eigenmode function, ω k are k modal center frequencies, is the derivative with respect to time t, j is the imaginary unit;

[0031] 2) Through ADMM, the statistical characteristics of each mode are extracted to form a complete feature vector and construct the ground response feature description space;

[0032] The eigenvector is represented as:

[0033] F=[f1,f2,...,f n ] T ;

[0034] Where, f i is the eigenvalue of the i-th modal statistics;

[0035] 3) Combine the obtained eigenvector with soil moisture, electrical conductivity, and slip rate to form a multidimensional feature matrix, which is expressed as:

[0036] X∈R N×D ;

[0037] Where X is the multidimensional feature matrix, N is the number of samples, and D is the number of feature bits for each sample.

[0038] Preferably, in S4, the optimization goal of UMAP is to minimize the KL divergence of the high- and low-dimensional neighborhood distributions to achieve mapping of high-dimensional features to low-dimensional features, which is expressed as:

[0039]

[0040] Where, P ij , Q ij Represents the proximity probability between samples in high-dimensional and low-dimensional spaces, is the loss function (the optimization goal is to maximize the consistency of neighborhood relationships in high-dimensional and low-dimensional spaces).

[0041] Preferably, in S5, the extreme learning machine is a single hidden layer feedforward network, and the network structure includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer corresponds to the low-dimensional feature dimension, and the number of neurons in the output layer corresponds to the number of ground type categories.

[0042] Preferably, in S5, the extreme learning machine constructs a hidden layer output matrix by randomly initializing input weights and biases, and outputs weights, which are expressed as:

[0043] β=H + T;

[0044] Among them, β is the output weight, H is the output matrix, H + is the Moore-Penrose pseudo-inverse, and T is the target output matrix.

[0045] Preferably, in S5, the input weights and biases are globally optimized by a genetic algorithm, with the root mean square error RESM as the fitness function, which is expressed as:

[0046]

[0047] Where y i is the true value of the i-th sample, is the predicted value of the i-th sample, and N is the number of samples;

[0048] Genetic algorithms generate optimal parameter individuals through encoding, selection, crossover, and mutation mechanisms, replacing the random initialization process, enhancing the prediction ability of the extreme learning machine and avoiding overfitting.

[0049] The present invention also provides a real-time identification system for the ground type of a tractor operating in the field, which is used to implement a real-time identification method for the ground type of a tractor operating in the field, comprising a multi-source sensor acquisition unit, a wireless communication transmission module, and an intelligent decision-making and recognition module. The multi-source sensor acquisition unit comprises a wheel multi-axis acceleration sensor, a soil moisture sensor, a soil conductivity sensor, a wheel speed sensor, and a GNSS receiver.

[0050] The wireless communication transmission module uses Bluetooth wireless communication to transmit multi-source data of the multi-source sensor acquisition unit, and cooperates with the STM32 receiving module and the CAN bus converter to realize data conversion and synchronization;

[0051] The intelligent decision-making and recognition module is set as a notebook computing platform, which integrates the UMAP-GA-ELM model to perform feature dimensionality reduction and classification reasoning on the multi-source data of the multi-source sensor acquisition unit and output the ground type recognition result;

[0052] The above hardware structure adopts magnetic metal packaging to achieve high-strength and impact-resistant fixation of the sensor, and has good rapid deployment and adaptability to highly dynamic operating environments.

[0053] Therefore, the present invention provides a method and system for real-time identification of ground type during tractor field operations, which has the following beneficial effects:

[0054] (1) By integrating multiple types of sensors such as acceleration, wheel speed, GNSS, soil moisture and conductivity, the extended Kalman filter (EKF) algorithm is used to achieve real-time dynamic estimation of slip rate, thereby enhancing the system's comprehensive perception of ground adhesion status and disturbance conditions.

[0055] (2) The variational mode decomposition (VMD) method is used to perform adaptive multi-band decomposition of the wheel vibration signal, extract statistical features such as kurtosis factor, impulse factor, and waveform factor, and construct a high-dimensional working condition feature vector with physical interpretability and statistical discriminability, significantly improving the system's perception and analysis accuracy of complex ground conditions.

[0056] (3) To address the problems of high dimensionality and strong redundancy in feature space, the unified manifold approximation and projection algorithm (UMAP) is introduced to achieve low-dimensional embedding of feature vector space. While retaining key discriminant information, it reduces the complexity of model training and inference, and improves the response speed and computational stability of the overall system.

[0057] (4) A lightweight ground recognition model based on the extreme learning machine (ELM) is constructed, and the genetic algorithm (GA) is integrated to perform global parameter optimization of the network input weights and hidden layer biases, effectively solving the performance uncertainty caused by random initialization. The model has high-speed training, strong generalization ability and real-time deployability, and can complete ground type recognition within a single wheel cycle, meeting the requirements of field operations for high real-time performance and robustness.

[0058] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1is a flow chart of an identification method according to an embodiment of the present invention;

[0060] Figure 2 This is a diagram of a multi-sensor parameter fusion process based on Kalman aluminum foil according to an embodiment of the present invention;

[0061] Figure 3 This is an exploded view of a VMD according to an embodiment of the present invention;

[0062] Figure 4 This is a diagram of the neural network structure of an extreme learning machine according to an embodiment of the present invention;

[0063] Figure 5 This is a diagram of the UMAP dimension reduction process according to an embodiment of the present invention:

[0064] Figure 6 This is a flow chart of the genetic algorithm optimization ELM according to an embodiment of the present invention;

[0065] Figure 7 This is a graph showing the error of an embodiment of the present invention changing with genetic generation:

[0066] Figure 8 This is a test set identification result diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0068] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0069] Example

[0070] The present invention also provides a real-time identification system for the ground type of a tractor operating in the field, which is used to implement a real-time identification method for the ground type of a tractor operating in the field, including a multi-source sensor acquisition unit, a wireless communication transmission module and an intelligent decision-making and recognition module. The multi-source sensor acquisition unit includes a wheel multi-axis acceleration sensor, a soil moisture sensor, a soil conductivity sensor, a wheel speed sensor and a GNSS receiver; and is used to obtain ground disturbance response, moisture status, conductivity characteristics and vehicle dynamics parameters.

[0071] The wireless communication transmission module uses Bluetooth wireless communication to transmit multi-source data from the multi-source sensor acquisition unit to avoid the risk of wheel entanglement. It cooperates with the STM32 receiving module to complete TTL signal conversion and transfer. The CAN bus converter synchronously reads wheel speed information for dynamic adjustment of the acceleration signal sampling period.

[0072] The intelligent decision-making and recognition module is set as a notebook computing platform. The notebook computing platform integrates the UMAP-GA-ELM model to perform feature dimensionality reduction and classification reasoning on the multi-source data of the multi-source sensor acquisition unit, and output the ground type recognition results.

[0073] In this embodiment, a multi-axis wheel acceleration sensor (MPU6050) is fixed to the wheel hub, a soil moisture and conductivity sensor is buried 5 cm above the ground at the front of the vehicle, a wheel speed sensor is mounted on the axle, and the GNSS receiver antenna is placed in an unobstructed position on the roof. The wheel-end sensor transmits data via a Bluetooth module (such as HC-05), an STM32 receiver module (STM32F103) completes the TTL to USB signal conversion, and a CAN bus converter (such as USBCAN-II) connects the wheel speed sensor to the laptop. The sensor sampling frequency is set to 100Hz, the Bluetooth communication baud rate is 115200bps, the STM32 processor has a main frequency of 72MHz, and the laptop computing platform is equipped with an i7 processor and 16GB of memory.

[0074] In this system, all data is centrally processed by a laptop computing platform. Feature dimensionality reduction and classification inference are performed based on the UMAP-GA-ELM model, enabling rapid identification of ground types (such as muddy, dry, hard, wet, and saline-alkali) within a single wheel rotation. The system utilizes a magnetic metal package, making it suitable for agricultural operations subject to strong vibration and high-frequency impact, and possesses excellent environmental adaptability and engineering deployment capabilities.

[0075] like Figure 1 As shown, the present invention provides a method for real-time identification of the ground type during tractor field operation, comprising the following steps:

[0076] S1. Obtain wheel vibration signals, soil properties, wheel speed, and ground speed data through sensors; soil properties include soil moisture and electrical conductivity;

[0077] S2. Construct state variables based on wheel speed and ground speed, expressed as:

[0078] x=[v,s] T ;

[0079] Where v is the ground speed and s is the slip rate, which can be expressed as:

[0080]

[0081] Where r is the wheel diameter and w is the wheel angular velocity.

[0082] Then the slip rate is estimated recursively by using the extended Kalman filter. First, the state transition model of the extended Kalman filter is defined as:

[0083] X k=f(X k-1 ,u k-1 )+W k ;

[0084] Where, X k is the current state, f(·) is the evolution function of the previous moment, W k is the process noise.

[0085] The observation model is:

[0086] Z k =h(X k )+V k ;

[0087] Where Z k is the observed value, h(X k ) is the state-to-observation mapping function, V k is the observation noise.

[0088] In the prediction stage, the state estimation of the previous moment is used Predict the current state through the state transition function:

[0089]

[0090] At the same time, the prediction covariance matrix is ​​updated and expressed as:

[0091]

[0092] Among them, F k is the first-order Jacobian matrix of the state transfer function, f(·), Q k is the observation noise covariance matrix, the prediction covariance Characterizes the uncertainty of the current estimated state and will be used as a weight factor in observation updates.

[0093] The predicted state is corrected with the observed value, which is expressed as:

[0094]

[0095] Where K k is the Kalman gain matrix,

[0096] And update the state covariance matrix, expressed as:

[0097]

[0098] Where I is the identity matrix, H k is the observation function h(X k ) for the state variable, X k The Jacobian matrix of ;

[0099] The prediction error is adjusted through the covariance update process, thereby closing the state recursive loop and improving the stability and accuracy of the estimation.

[0100] Use Figure 2 The process shown in the figure performs multi-sensor parameter fusion and estimates the slip rate through the state transition update and observation update process of the extended Kalman filter, effectively suppressing multi-source measurement errors and obtaining key parameters of ground contact characteristics. Specifically:

[0101] The system first initializes, setting an initial estimate of the vehicle state and its covariance matrix. It then concurrently acquires measurements from wheel speed sensors and observations from the navigation satellite system (GNSS). Based on the previous state estimate and the current control input, the system predicts the vehicle state and covariance. Next, it calculates the Kalman gain to balance the error between the predicted and observed values, and updates the state estimate and covariance matrix. Finally, the vehicle's slip is calculated based on the updated state variables, achieving dynamic estimation and fusion of the ground adhesion state.

[0102] S3. Perform variational modal decomposition on the wheel vibration signal to obtain the intrinsic mode function, extract the statistical characteristics of each mode, and combine the soil characteristics and slip rate to form a multi-dimensional feature matrix; Figure 3 The specific steps are as follows:

[0103] 1) The collected wheel vibration signal is subjected to variational mode decomposition using VMD, and the wheel vibration signal is input as the original signal to obtain the intrinsic mode function {u k (t)}; The goal of variational mode decomposition is to minimize the sum of the bandwidths of each mode, which can be expressed as:

[0104]

[0105] In the formula, {u k (t)} is the kth eigenmode function, ω k are k modal center frequencies, is the derivative with respect to time t, and j is the imaginary unit.

[0106] In this process, the penalty factor and Lagrange multiplier are introduced to construct the augmented Lagrangian objective function as follows:

[0107]

[0108] in, is the augmented Lagrangian objective function; u k (t) is the kth modal component; ω kis the center frequency of the kth mode; λ(t) is the Lagrange multiplier; K is the number of modes; α is the penalty factor.

[0109] Initialize the eigenfunction u of each mode k (0), center frequency w k (t), Lagrange multiplier λ(t).

[0110] 2) The optimization objective function is solved iteratively by using the Direction Alternation Multiplier Method (ADMM), and the modal function u is continuously updated in the process. k and its corresponding center frequency w k After each update, it is determined whether the convergence conditions are met. If not, the iteration is continued. When the convergence conditions are met, K modal components with different frequency characteristics are obtained. Finally, the corresponding statistical characteristic parameters are extracted from each modal component to form a complete feature vector, and the ground response feature description space is constructed as the input of the subsequent ground type recognition model.

[0111] The eigenvector is represented as:

[0112] F=[f1,f2,...,f n ] T ;

[0113] Where, f i is the ith modal statistical eigenvalue.

[0114] 3) Combine the obtained eigenvector with soil moisture, electrical conductivity, and slip rate to form a multidimensional feature matrix, which is expressed as:

[0115] X=[f1,f2,…,f n , humidity, conductivity, v,s]∈R N×D ;

[0116] Where X is the multidimensional feature matrix, N is the number of samples, and D is the number of feature bits for each sample.

[0117] S4. Considering the problem of high feature dimension and strong redundancy, UMAP is further used to perform nonlinear dimensionality reduction on the multidimensional feature matrix. Figure 4 The process shown in the figure converts the high-dimensional feature matrix X∈R N×D Mapped to a low-dimensional feature matrix X∈R N×d First, the input high-dimensional feature matrix (modal statistics, soil properties, and dynamic characteristics) X is used as the initial data, and a high-dimensional neighborhood graph is constructed by calculating the distance relationship between samples. Then, based on this neighborhood structure, UMAP embeds the high-dimensional features into a low-dimensional space while maintaining the local data structure, obtaining a dense and clearly distributed low-dimensional feature representation.

[0118] The optimization goal of UMAP is to minimize the KL divergence of the high- and low-dimensional neighborhood distributions and realize the mapping of high-dimensional features to low-dimensional features, which can be expressed as:

[0119]

[0120] Where, P ij , Q ij Represents the proximity probability between samples in high-dimensional and low-dimensional spaces, is the loss function.

[0121] S5. Input the low-dimensional feature matrix into the extreme learning machine model optimized by the genetic algorithm for classification reasoning, and output the ground type recognition result.

[0122] The extreme learning machine is a single hidden layer feedforward network, such as Figure 5 As shown in the figure, the network structure includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer corresponds to the low-dimensional feature dimension, and the number of neurons in the output layer corresponds to the number of ground type categories. The extreme learning machine constructs the hidden layer output matrix and output weights by randomly initializing the input weights and biases, which are expressed as:

[0123] β=H + T;

[0124] Among them, β is the output weight, H is the output matrix, H + is the Moore-Penrose pseudo-inverse, and T is the target output matrix.

[0125] In order to overcome the instability caused by random initialization, a genetic algorithm is introduced to perform global optimization of input weights and biases. The specific process is as follows: Figure 6 As shown, the root mean square error RESM is used as the fitness function, which is expressed as:

[0126]

[0127] Where y i is the true value of the i-th sample, is the predicted value of the i-th sample, and N is the number of samples.

[0128] Genetic algorithms generate optimal parameter individuals through encoding, selection, crossover, and mutation mechanisms, replacing the random initialization process, enhancing the prediction ability of the extreme learning machine and avoiding overfitting.

[0129] The above identification system and identification method are used to perform performance verification. The specific steps are as follows:

[0130] To conduct field data collection, we first wrote a serial communication script based on Python to read the sensor data in real time and store it as a .csv file, which contains fields such as timestamp, acceleration, humidity, conductivity, wheel speed, GNSS coordinates and speed. Then, after processing according to the steps described in the present invention, a data set of a multidimensional feature matrix was constructed. After dimensionality reduction processing using the unified manifold approximation and projection algorithm, 4320 sets of valid data containing 19 principal components were obtained. The sample data was divided into a training set and a test set, of which 3320 sets were used for model training and 1000 sets were used for model verification. The training data and the test data were completely independent. The distribution of samples of each ground type in the test set was roughly balanced, and the labeling method was: cement ground was 1, sand ground was 2, and clay ground was 3.

[0131] In the modeling process, the extreme learning machine (GA-ELM) optimized by genetic algorithm is used as the ground classification model. The number of input layer nodes is set to 19, corresponding to the principal component characteristics after dimensionality reduction; the output layer is 3, corresponding to three types of ground types. The genetic algorithm parameters are set as follows: population size is 20, maximum evolutionary generations is 100, crossover probability is 0.7, mutation probability is 0.01, and fitness inheritance ratio is 0.95. The activation function of the extreme learning machine is the Sigmoid function, and the number of hidden layer neurons is set to 40. During the model training process, the input weights and bias parameters of the neural network are iteratively optimized through the genetic algorithm. Figure 7 As shown in the figure, at the 57th generation, the root mean square error reaches the minimum and the model converges well.

[0132] The test set data was then used to verify the accuracy of the trained GA-ELM model. The recognition results are as follows: Figure 8 The overall performance shows that the model is most accurate in identifying cement floor samples. The recognition results for other types of floors also have good discrimination and low error rates, indicating that the operating parameters vary significantly under different floor conditions and have good classification characteristics.

[0133] In general, for the 1,000 samples in the test set, the model correctly identified 956 of them, with an accuracy rate of 95.6%, which is a good accuracy rate.

[0134] The recognition results of the three types of ground in the test set, including Precision, Recall, and F1, are shown in Table 1.

[0135] Table 1 Precision, Recall and F1 of the recognition model

[0136] Ground type Precision Recall F1 cement 99.7％ 94.6％ 97.1％ sand 95.1％ 93.1％ 94.1％ clay 92.8％ 93.4％ 93.1％

[0137] Analysis of Table 1 shows that the recognition model performs best for cement surfaces in the test data set, while clay surfaces perform relatively poorly. Overall, the average Precision of the samples is 95.9%, the average Recall is 93.7%, and the average F1 is 94.8%, demonstrating that the recognition method presented here has high accuracy and stability, making it particularly suitable for quickly determining surface type in agricultural scenarios. Furthermore, the system relies on only 1 second of tractor operating parameter data to identify the working surface, offering high efficiency and adaptability, meeting the requirements for rapid environmental perception and real-time control in agricultural operations.

[0138] Therefore, the present invention provides a real-time identification method and system for the ground type of tractor field operations. The system relies on the extended Kalman filter (EKF) to estimate the slip rate as the basis of the working condition parameter, combines the variational mode decomposition (VMD) to extract multi-frequency feature information, adopts the unified manifold approximation and projection (UMAP) to achieve dimensionality reduction of high-dimensional features, and then uses the extreme learning machine (ELM) neural network to construct a classification model. The parameters are optimized through the genetic algorithm (GA) to achieve efficient and intelligent identification under complex agricultural ground conditions.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A real-time identification method for the ground type of a tractor in field operation, characterized in that: The following steps are involved: S1. Obtain wheel vibration signals, soil characteristics, wheel speed, and ground speed data through sensors; S2. Construct state variables based on wheel speed and ground speed, and recursively estimate the slip rate using an extended Kalman filter as a ground adhesion feature. S3. Perform variational modal decomposition on the wheel vibration signal to obtain the intrinsic mode function, extract the statistical characteristics of each mode, and combine the soil characteristics and slip rate to form a multidimensional feature matrix; S4, using UMAP to perform nonlinear dimensionality reduction on the multidimensional feature matrix to obtain a low-dimensional feature matrix; S5. Input the low-dimensional feature matrix into the extreme learning machine model optimized by the genetic algorithm for classification reasoning, and output the ground type recognition result.

2. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S1, soil properties include soil moisture and electrical conductivity.

3. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S2, the state variables are expressed as: x=[v,s] T ; Where v is the ground speed and s is the slip rate, which can be expressed as: Where r is the wheel diameter and w is the wheel angular velocity.

4. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S2, the state transition model of the extended Kalman filter is: X k =f(X k-1 ,u k-1 )+W k ; Where, X k is the current state, f(·) is the evolution function of the previous moment, W k is the process noise; The observation model is: Z k =h(X k )+V k ; Where Z k is the observed value, h(X k ) is the state-to-observation mapping function, V k is the observation noise; The slip rate is estimated through the state transfer update and observation update process of the extended Kalman filter.

5. The method for real-time identification of ground type during tractor field operations according to claim 2, characterized in that: The specific steps of S3 are as follows: 1) The acquired wheel vibration signal is subjected to variational mode decomposition using VMD to obtain the intrinsic mode function. The goal of variational mode decomposition is to minimize the sum of the bandwidths of each mode, which can be expressed as: In the formula, {u k (t)} is the kth eigenmode function, ω k are k modal center frequencies, is the derivative with respect to time t, j is the imaginary unit; 2) Through ADMM, the statistical characteristics of each mode are extracted to form a complete feature vector and construct the ground response feature description space; The eigenvector is represented as: F=[f1,f2,...,f n ] T ; Where, f i is the eigenvalue of the i-th modal statistics; 3) Combining soil moisture, electrical conductivity and slip rate, a multidimensional feature matrix is ​​formed, which is expressed as: X∈R N×D ; Where X is the multidimensional feature matrix, N is the number of samples, and D is the number of feature bits for each sample.

6. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S4, the optimization goal of UMAP is to minimize the KL divergence of the high- and low-dimensional neighborhood distributions and realize the mapping of high-dimensional features to low-dimensional features, which is expressed as: Where, P ij , Q ij Represents the proximity probability between samples in high-dimensional and low-dimensional spaces, is the loss function.

7. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S5, the extreme learning machine is a single hidden layer feedforward network, and the network structure includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer corresponds to the low-dimensional feature dimension, and the number of neurons in the output layer corresponds to the number of ground type categories.

8. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S5, the extreme learning machine constructs the hidden layer output matrix and output weights by randomly initializing the input weights and biases, which are expressed as: β=H + T; Among them, β is the output weight, H is the output matrix, H + is the Moore-Penrose pseudo-inverse, and T is the target output matrix.

9. The method for real-time identification of ground type during tractor field operations according to claim 1, characterized in that: In S5, the input weights and biases are globally optimized by a genetic algorithm, with the root mean square error RESM as the fitness function, which is expressed as: Where y i is the true value of the i-th sample, is the predicted value of the i-th sample, and N is the number of samples; Genetic algorithms generate optimal parameter individuals through encoding, selection, crossover, and mutation mechanisms, replacing the random initialization process, enhancing the prediction ability of the extreme learning machine and avoiding overfitting.

10. A system for real-time identification of ground type for tractor field operations, for implementing the method for real-time identification of ground type for tractor field operations according to any one of claims 1 to 9, characterized in that: It includes a multi-source sensor acquisition unit, a wireless communication transmission module and an intelligent decision-making recognition module. The multi-source sensor acquisition unit includes a wheel multi-axis acceleration sensor, a soil moisture sensor, a soil conductivity sensor, a wheel speed sensor and a GNSS receiver; The wireless communication transmission module uses Bluetooth wireless communication to transmit multi-source data of the multi-source sensor acquisition unit, and cooperates with the STM32 receiving module and the CAN bus converter to realize data conversion and synchronization; The intelligent decision-making and recognition module is set as a notebook computing platform, which integrates the UMAP-GA-ELM model, performs feature dimension reduction and classification reasoning on the multi-source data of the multi-source sensor acquisition unit, and outputs the ground type recognition result.

Images