A method and system for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar

Abstract

The invention discloses a method and system for detecting bulk density of farmland cultivated layer soil based on a multi-channel ground penetrating radar, and relates to the field of soil bulk density detection. The invention can directly measure a large area of ​​soil through the multi-channel ground penetrating radar, and obtain the corresponding inverse of the average amplitude envelope of the cultivated layer electromagnetic wave signal using antennas with different spacings, predict the volume moisture content through a model based on the inverse of the average amplitude envelope, and then use the volume moisture content and the inverse of the average amplitude envelope as inputs of a trained bulk density inversion machine learning model, so as to achieve high-resolution soil bulk density measurement, and be compatible with loose cultivated layer soil, effectively improve the accuracy and efficiency of measurement, and solve the problems that the electromagnetic wave information of the cultivated layer of the existing single-channel ground penetrating radar system is insufficient, the traditional soil bulk density detection method is difficult to achieve large-area rapid and continuous detection, and is not suitable for detecting loose cultivated layer soil.

Description

Technical Field

[0001] The present invention relates to the field of soil bulk density detection, and in particular to a method and system for detecting soil bulk density of a farmland cultivated layer based on a multi-channel ground penetrating radar. Background Art

[0002] Farmland soil is generally divided into the tillage layer, plow bottom layer, subsoil layer and subsoil layer from top to bottom. The tillage layer plays a vital role in agricultural production due to its fast soil material transformation, good permeability and high nutrient content. Its role is mainly reflected in providing nutrients required for plant growth, good physical properties, regulating soil temperature, and buffering properties of physical and chemical properties. Soil bulk density is the mass of unit original soil solids and is an indicator to measure the looseness and tightness of soil. Bulk density is a comprehensive reflection of soil texture, structure, porosity and other physical properties. Appropriate tillage layer soil bulk density can better retain water and nutrients and provide a stable environment for crop growth. In the process of agricultural production, unreasonable tillage methods, fertilization management and irrigation management will lead to an increase in the bulk density of the tillage layer soil, an increase in soil compaction, and poor permeability and air permeability, which is not conducive to the further rooting and growth of crop roots, and may also lead to the loss of water and nutrients. Therefore, obtaining the bulk density of the tillage layer soil in farmland is of great significance for evaluating soil physical properties, fertility status, and formulating agricultural production measures.

[0003] There are currently a variety of methods for measuring bulk density of cultivated soil based on point scale, among which the most commonly used and widespread is the ring knife method, which has the advantages of simple equipment, easy operation, and accurate results. Similar methods for measuring soil bulk density based on the relationship between mass and volume include the wax sealing method, mercury discharge method, and sand filling method. In addition, there are methods for indirectly measuring soil bulk density by measuring other physical properties of the soil, such as the ray method and electromagnetic-stable probe measurement. However, although these methods can provide relatively accurate bulk density detection data at a point scale, they have the problems of low efficiency and high cost, making it difficult to achieve rapid and continuous detection over a large area, and most methods are not suitable for detecting loose cultivated soil. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method and system for detecting bulk density of farmland cultivated layer soil based on multi-channel ground penetrating radar, which solves the problem that the existing soil bulk density detection method is difficult to achieve rapid and continuous detection over a large area and is not suitable for detecting loose cultivated layer soil.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A method for detecting soil bulk density of a cultivated layer of a farmland based on a multi-channel ground penetrating radar is provided, which comprises the following steps:

[0007] S1. Construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the cultivated layer of farmland soil;

[0008] S2, constructing a trained bulk density inversion machine learning model;

[0009] S3, detecting the tillage layer of the target farmland by using a ground penetrating radar to obtain an initial detection signal;

[0010] S4, preprocessing the initial detection signal, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal;

[0011] S5. Based on the inverse of the average amplitude envelope obtained in step S4, a corresponding volume moisture content prediction value is obtained according to a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer soil of the farmland;

[0012] S6. Use the inverse of the average amplitude envelope of the electromagnetic wave signal of the cultivated layer obtained by antennas with different spacings obtained in step S4 and the predicted value of volumetric moisture content obtained in step S5 as inputs of the trained bulk density inversion machine learning model, and predict the bulk density of the cultivated layer of the farmland soil through the trained bulk density inversion machine learning model to obtain the bulk density test result of the cultivated layer of the farmland soil.

[0013] Furthermore, the specific method of constructing the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the cultivated layer of the farmland through experiments in step S1 includes the following steps and sub-steps:

[0014] S1-1. Select the target farmland cultivated layer soil for air-drying and sieving;

[0015] S1-2. Based on the soil processed in step S1-1, several simulated samples with different bulk densities and volumetric moisture contents are constructed in a wooden box; wherein the thickness of the soil in the wooden box is greater than or equal to 20 cm;

[0016] S1-3, placing a ground penetrating radar on the surface of each simulated sample constructed in step S1-2 with more than three antenna spacings, and collecting initial detection signals corresponding to the simulated samples;

[0017] S1-4, preprocessing the initial detection signal corresponding to the analog sample collected in step S1-3, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the detection signal corresponding to the preprocessed analog sample;

[0018] S1-5, using the ring knife method to collect each simulated sample and dry it, and calculate the actual volume moisture content and bulk density of the simulated sample;

[0019] S1-6. Use the inverse of the average amplitude envelope obtained in step S1-4 as a training sample of the shallow neural network model, and the actual volume moisture content of the simulated sample as a label of the shallow neural network model, and train the shallow neural network model to obtain a trained shallow neural network model, that is, a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer of farmland soil.

[0020] Furthermore, during the training process of the shallow neural network model, the stability of the shallow neural network model is evaluated through cross-validation, and the prediction accuracy of the shallow neural network model is evaluated through the root mean square error indicator, until the stability and prediction accuracy of the shallow neural network model reach the set conditions, the training of the shallow neural network model is terminated.

[0021] Furthermore, the specific method for constructing the trained bulk density inversion machine learning model in step S2 is:

[0022] The inverse of the average amplitude envelope of the electromagnetic wave signal of the plow layer and the true volume moisture content obtained by antennas with different spacings corresponding to the simulated samples are used as training samples of the convolutional neural network model. The true bulk density of the simulated samples is used as labels to train the convolutional neural network model to obtain the trained convolutional neural network model, that is, the trained bulk density inversion machine learning model is obtained.

[0023] Furthermore, during the training process of the convolutional neural network model, the stability of the convolutional neural network model is evaluated by cross-validation, and the prediction accuracy of the convolutional neural network model is evaluated by the root mean square error indicator. The training of the convolutional neural network model is terminated when the stability and prediction accuracy of the convolutional neural network model meet the set conditions.

[0024] Furthermore, the specific method for preprocessing the initial detection signal is:

[0025] The initial detection signal is corrected at time 0 by measuring the travel time of WARR signals with different spacings in the air to obtain the signal after correction at time 0;

[0026] The low-frequency noise in the signal corrected at time 0 is eliminated by using a De-Wow filter to obtain a low-frequency filtered signal;

[0027] For the signal after low-frequency filtering, by comparing the differences in the amplitude energies of multiple groups of fixed-pitch air waves measured N times, the amplitude energy of the fixed-pitch air waves measured in a single time is selected as the benchmark, and the amplitude energies of the fixed-pitch air waves measured in the remaining N-1 times are corrected to the same level to obtain the amplitude-corrected signal, that is, the initial detection signal after preprocessing.

[0028] A system for realizing a method for detecting soil bulk density of a cultivated layer of a farmland based on a multi-channel ground penetrating radar is provided, which comprises a data collection module, a multi-channel ground penetrating radar measurement module, a data transmission module and a host computer;

[0029] The data collection module includes a camera and an RTK positioning device; the camera is used to collect the surface image of the cultivated layer of the target farmland; the RTK positioning device is used to monitor the position information of the points detected by the ground penetrating radar in real time through the RTK positioning technology;

[0030] Multi-channel ground penetrating radar measurement module, used to detect the tillage layer of the target farmland and obtain the initial detection signal;

[0031] A data transmission module, used to transmit the data obtained by the data collection module and the multi-channel ground penetrating radar measurement module to the host computer through a wireless communication network;

[0032] The host computer includes a feature extraction module and a data integration intelligent processing module; wherein the feature extraction module is used to preprocess the initial detection signal and extract the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal;

[0033] The data integration intelligent processing module is used to construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; to construct a trained bulk density inversion machine learning model; to obtain the corresponding volumetric moisture content prediction value based on the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; based on the inverse of the average amplitude envelope and the volumetric moisture content prediction value, the bulk density of the farmland cultivated layer soil is predicted through the trained bulk density inversion machine learning model to obtain the bulk density test results of the farmland cultivated layer soil; to store and display the data obtained by the data collection module and the corresponding bulk density test results of the farmland cultivated layer soil.

[0034] The beneficial effects of the present invention are:

[0035] 1. The present invention can directly measure a large area of ​​soil through a ground penetrating radar, and obtain the corresponding inverse of the average amplitude envelope. The volume moisture content is predicted through a model based on the inverse of the average amplitude envelope. The volume moisture content and the inverse of the average amplitude envelope are used as inputs of the trained bulk density inversion machine learning model. High-resolution soil bulk density measurement can be achieved, and it is compatible with loose cultivated layer soil, effectively improving the accuracy and efficiency of the measurement.

[0036] 2. The present invention adopts the target farmland cultivated layer soil to construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil. By testing the volumetric moisture content and bulk density of the target farmland cultivated layer soil in the experimental process, a trained bulk density inversion machine learning model can be constructed for the target farmland cultivated layer soil. This allows the soil to construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil and the trained bulk density inversion machine learning model to obtain more accurate prediction results when predicting the bulk density of the target farmland cultivated layer soil. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flowchart of the training and testing process of this method;

[0038] Figure 2 This is a schematic diagram of the structure of this system. DETAILED DESCRIPTION

[0039] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0040] like Figure 1 As shown, the method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar includes the following steps:

[0041] S1. Construct a relationship model between the inverse of the average amplitude envelope of shallow electromagnetic wave signals obtained by antennas with different spacing and the volumetric moisture content of the cultivated layer of farmland soil (i.e., moisture inversion model);

[0042] S2, constructing a trained bulk density inversion machine learning model (i.e., bulk density inversion model);

[0043] S3, detecting the tillage layer of the target farmland by using a ground penetrating radar to obtain an initial detection signal;

[0044] S4, preprocessing the initial detection signal, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal;

[0045] S5. Based on the inverse of the average amplitude envelope obtained in step S4, a corresponding volume moisture content prediction value is obtained according to a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer soil of the farmland;

[0046] S6. Use the inverse of the average amplitude envelope obtained in step S4 and the predicted value of volumetric moisture content obtained in step S5 as inputs to the trained bulk density inversion machine learning model, and predict the soil bulk density of the cultivated layer of the farmland through the trained bulk density inversion machine learning model to obtain the soil bulk density test results of the cultivated layer of the farmland.

[0047] The specific method of constructing the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the cultivated layer of the farmland through experiments in step S1 includes the following steps and sub-steps:

[0048] S1-1. Select the target farmland cultivated layer soil (depth within 20 cm) and air-dry and sieve it in the laboratory;

[0049] S1-2, based on the soil processed in step S1-1, set the volume moisture content range and bulk density detection range (1.0 g / cm 3 ~1.5 g / cm 3 ) with different bulk density and volume moisture content conditions (more than 5 for each), prepare the corresponding simulated samples of cultivated soil and fill them evenly in the wooden box for testing; the thickness of the soil in the wooden box is greater than or equal to 20cm;

[0050] S1-3, placing a ground penetrating radar (with a center frequency between 400 MHz and 900 MHz) on the surface of each simulated sample constructed in step S1-2 with more than three antenna spacings, and collecting initial detection signals corresponding to the simulated samples;

[0051] S1-4, preprocessing the initial detection signal corresponding to the analog sample collected in step S1-3, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the detection signal corresponding to the preprocessed analog sample;

[0052] S1-5. Collect each simulated sample using the ring knife method and dry it, and calculate the actual volume moisture content and bulk density of the simulated sample; the moisture content after drying is The calculation formula is:

[0053]

[0054] is the mass of water (the mass of the wet soil sample minus the mass of the dry soil sample), is the mass of dry soil sample;

[0055] The calculation formula for the bulk density after drying is:

[0056]

[0057] Where V is the volume of the soil sample (determined by the volume of the ring cutter);

[0058] S1-6. Use the inverse of the average amplitude envelope obtained in step S1-4 as a training sample of the shallow neural network model, and the actual volume moisture content of the simulated sample as a label of the shallow neural network model, and train the shallow neural network model to obtain a trained shallow neural network model, that is, a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer of farmland soil.

[0059] During the training process of the shallow neural network model, the stability of the shallow neural network model is evaluated through cross-validation, and the prediction accuracy of the shallow neural network model is evaluated by the root mean square error indicator. The training of the shallow neural network model is terminated when the stability and prediction accuracy of the shallow neural network model both meet the set conditions.

[0060] In this embodiment, the shallow neural network model uses a BP neural network with three hidden layers (the number of node neurons is 20, 20, and 30 respectively), the activation function uses ReLu, the training number is 1000 times, and the initial learning rate is set to 10 -3 Cross-validation (10 folds) was used to evaluate the model stability and the root mean square error (RMSE) was used to evaluate the model prediction accuracy. During normalization, it was assumed that a certain feature sample set of the original signal was , where the maximum and minimum values ​​are F max and F min , then the normalized value of the feature is:

[0061]

[0062] The root mean square error calculation formula is as follows:

[0063]

[0064] Where n represents the total number of training samples. It is i The true value of the samples, It is i The model predicts the value of samples.

[0065] The specific method for constructing the trained bulk density inversion machine learning model in step S2 is:

[0066] The convolutional neural network model is trained with the inverse of the average amplitude envelope corresponding to the simulated samples and the true volumetric moisture content as training samples of the convolutional neural network model and the true bulk density of the simulated samples as labels to obtain a trained convolutional neural network model, that is, a trained bulk density inversion machine learning model.

[0067] During the training process of the convolutional neural network model, the stability of the convolutional neural network model is evaluated through cross-validation (10 folds), and the prediction accuracy of the convolutional neural network model is evaluated by the root mean square error indicator. The training of the convolutional neural network model is terminated when the stability and prediction accuracy of the convolutional neural network model meet the set conditions.

[0068] In this embodiment, when the number of antennas of the ground penetrating radar is four, the initial detection signal corresponds to four channels, and the average amplitude envelope inverse also corresponds to four channels. At this time, the average amplitude envelope inverse and the volumetric water content prediction value, when used as the input of the (trained) bulk density inversion machine learning model, can be expressed as: ,in It corresponds to the inverse of the average amplitude envelope of the four channels (amplitude time domain signal characteristics of the four channels), and W is the predicted value of volumetric water content.

[0069] In the specific implementation process, the convolutional neural network model contains 1 convolution layer (number of filters 32, one-dimensional convolution kernel 1*32) and 3 fully connected layers (size 128, 128, 128), the activation function uses ReLu, the number of training times is 1000, and the initial learning rate is set to 10 -5 .

[0070] The specific method for preprocessing the initial detection signal is:

[0071] The initial detection signal is corrected at time 0 by measuring the travel time of WARR signals with different spacings in the air to obtain the signal after correction at time 0;

[0072] The low-frequency noise in the signal corrected at time 0 is eliminated by using a De-Wow filter to obtain a low-frequency filtered signal;

[0073] For the signal after low-frequency filtering, by comparing the differences in the amplitude energies of multiple groups of fixed-pitch air waves measured N times, the amplitude energy of the fixed-pitch air waves measured in a single time is selected as the benchmark, and the amplitude energies of the fixed-pitch air waves measured in the remaining N-1 times are corrected to the same level to obtain the amplitude-corrected signal, that is, the initial detection signal after preprocessing.

[0074] In this embodiment, the air wave correction data measured for the first time can be directly set as the reference matrix, and the average maximum absolute value of each signal amplitude of K intervals of N channels is calculated as the reference energy Amax. Then the reference matrix is:

[0075] ;

[0076] The air wave correction data of each test is used to calculate the amplitude energy matrix Obs using the above method. By comparing it with the reference matrix Base, the correction coefficient matrix Coef can be obtained. The correction coefficient of each channel is obtained by taking the average value of the correction coefficient matrix rows, and then the corrected amplitude is calculated. The formula is as follows:

[0077]

[0078] in, i is the channel number, j is the antenna spacing number, For the i Channel raw amplitude, For the i The corrected amplitude of the channel.

[0079] When processing the initial detection signal after preprocessing, it is necessary to take the amplitude envelope E(t) of the time domain signal A(t) and perform Hilbert transform on the signal:

[0080]

[0081] in is the kernel function of the Hilbert transform, used for convolution operation. The amplitude envelope is obtained:

[0082] ;

[0083] The average amplitude envelope inverse (AEA) of the time interval from time 0 to the maximum amplitude envelope value of each channel can be extracted. -1 ).

[0084] like Figure 2 As shown, the system for realizing the method for detecting soil bulk density of farmland cultivated layer based on multi-channel ground penetrating radar includes a data collection module, a multi-channel ground penetrating radar measurement module, a data transmission module and a host computer;

[0085] The data collection module includes a camera and an RTK positioning device; the camera is used to collect the surface image of the cultivated layer of the target farmland; the RTK positioning device is used to monitor the position information of the points detected by the ground penetrating radar in real time through the RTK positioning technology;

[0086] Multi-channel ground penetrating radar measurement module, used to detect the tillage layer of the target farmland and obtain the initial detection signal;

[0087] A data transmission module, used to transmit the data obtained by the data collection module and the multi-channel ground penetrating radar measurement module to the host computer through a wireless communication network;

[0088] The host computer includes a feature extraction module and a data integration intelligent processing module; wherein the feature extraction module is used to preprocess the initial detection signal and extract the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal;

[0089] The data integration intelligent processing module is used to construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; to construct a trained bulk density inversion machine learning model; to obtain the corresponding volumetric moisture content prediction value based on the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; based on the inverse of the average amplitude envelope and the volumetric moisture content prediction value, the bulk density of the farmland cultivated layer soil is predicted through the trained bulk density inversion machine learning model to obtain the bulk density test results of the farmland cultivated layer soil; to store and display the data obtained by the data collection module and the corresponding bulk density test results of the farmland cultivated layer soil.

[0090] In one embodiment of the present invention, when performing soil bulk density detection in the field cultivated layer, the parameter settings and operation mode of the ground penetrating radar must be consistent with the trained bulk density inversion machine learning model construction experiment. The data integration intelligent processing module can mark and remove abnormal points such as surface unevenness, rutting, etc. based on the surface photos and terrain change information obtained by the data collection module.

[0091] In summary, the present invention collects ground penetrating radar and other detection data, adopts multiple data processing methods, integrates multiple data features, and combines machine learning algorithms to perform large-area tillage layer soil bulk density detection, which can improve the measurement efficiency and accuracy, and solves the problem that existing soil bulk density detection methods are difficult to achieve large-area rapid and continuous detection and are not suitable for detecting loose tillage layer soil.

Claims

1. A method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar, characterized in that: The following steps are involved: S1. Construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the cultivated layer of farmland soil; S2, constructing a trained bulk density inversion machine learning model; S3, detecting the tillage layer of the target farmland by using a ground penetrating radar to obtain an initial detection signal; S4, preprocessing the initial detection signal, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal; S5. Based on the inverse of the average amplitude envelope obtained in step S4, a corresponding volume moisture content prediction value is obtained according to a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer soil of the farmland; S6, taking the inverse of the average amplitude envelope of the electromagnetic wave signal of the cultivated layer obtained by antennas with different spacings obtained in step S4 and the predicted value of the volumetric moisture content obtained in step S5 as the input of the trained bulk density inversion machine learning model, predicting the bulk density of the cultivated layer of the farmland soil through the trained bulk density inversion machine learning model, and obtaining the bulk density test result of the cultivated layer soil of the farmland; The specific method for constructing the trained bulk density inversion machine learning model in step S2 is: The convolutional neural network model is trained with the inverse of the average amplitude envelope corresponding to the simulated samples and the true volumetric moisture content as training samples of the convolutional neural network model and the true bulk density of the simulated samples as labels to obtain a trained convolutional neural network model, that is, a trained bulk density inversion machine learning model.

2. The method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar according to claim 1, characterized in that: The specific method of constructing the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the cultivated layer of the farmland through experiments in step S1 includes the following steps and sub-steps: S1-1. Select the target farmland cultivated layer soil for air-drying and sieving; S1-2, based on the soil treated in step S1-1, construct several simulated samples with different bulk densities and volumetric moisture contents in a wooden box; The soil thickness in the wooden box is greater than or equal to 20cm; S1-3, placing a ground penetrating radar on the surface of each simulated sample constructed in step S1-2 with more than three antenna spacings, and collecting initial detection signals corresponding to the simulated samples; S1-4, preprocessing the initial detection signal corresponding to the analog sample collected in step S1-3, and extracting the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the detection signal corresponding to the preprocessed analog sample; S1-5, using the ring knife method to collect each simulated sample and dry it, and calculate the actual volume moisture content and bulk density of the simulated sample; S1-6. Use the inverse of the average amplitude envelope obtained in step S1-4 as a training sample of the shallow neural network model, and the actual volume moisture content of the simulated sample as a label of the shallow neural network model, and train the shallow neural network model to obtain a trained shallow neural network model, that is, a relationship model between the inverse of the average amplitude envelope and the volume moisture content of the cultivated layer of farmland soil.

3. The method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar according to claim 2 is characterized in that: During the training process of the shallow neural network model, the stability of the shallow neural network model is evaluated through cross-validation, and the prediction accuracy of the shallow neural network model is evaluated by the root mean square error indicator. The training of the shallow neural network model is terminated when the stability and prediction accuracy of the shallow neural network model both meet the set conditions.

4. The method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar according to claim 1, characterized in that: During the training process of the convolutional neural network model, the stability of the convolutional neural network model is evaluated through cross-validation, and the prediction accuracy of the convolutional neural network model is evaluated by the root mean square error indicator. The training of the convolutional neural network model is terminated when the stability and prediction accuracy of the convolutional neural network model meet the set conditions.

5. The method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar according to claim 1, characterized in that: The specific method for preprocessing the initial detection signal is: The initial detection signal is corrected at time 0 by measuring the travel time of WARR signals with different spacings in the air to obtain the signal after correction at time 0; The low-frequency noise in the signal corrected at time 0 is eliminated by using a De-Wow filter to obtain a low-frequency filtered signal; For the signal after low-frequency filtering, by comparing the differences in the amplitude energies of multiple groups of fixed-pitch air waves measured N times, the amplitude energy of the fixed-pitch air waves measured in a single time is selected as the benchmark, and the amplitude energies of the fixed-pitch air waves measured in the remaining N-1 times are corrected to the same level to obtain the amplitude-corrected signal, that is, the initial detection signal after preprocessing.

6. A system for implementing the method for detecting soil bulk density of cultivated layer of farmland based on multi-channel ground penetrating radar according to any one of claims 1 to 5, characterized in that: It includes data collection module, multi-channel ground penetrating radar measurement module, data transmission module and host computer; Data collection module, including camera and RTK positioning equipment; The camera is used to collect the surface image of the cultivated layer of the target farmland; the RTK positioning equipment is used to monitor the position information of the points detected by the ground penetrating radar in real time through the RTK positioning technology; Multi-channel ground penetrating radar measurement module, used to detect the tillage layer of the target farmland and obtain the initial detection signal; A data transmission module, used to transmit the data obtained by the data collection module and the multi-channel ground penetrating radar measurement module to the host computer through a wireless communication network; Host computer, Including feature extraction module and data integration intelligent processing module; The feature extraction module is used to preprocess the initial detection signal and extract the average amplitude envelope inverse of the time interval from time 0 to the maximum value of the amplitude envelope of the preprocessed initial detection signal; The data integration intelligent processing module is used to construct a relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; to construct a trained bulk density inversion machine learning model; and to obtain the corresponding volumetric moisture content prediction value based on the relationship model between the inverse of the average amplitude envelope and the volumetric moisture content of the farmland cultivated layer soil; Based on the inverse of the average amplitude envelope and the predicted value of volumetric moisture content, the soil bulk density of the cultivated layer of farmland is predicted through the trained bulk density inversion machine learning model to obtain the soil bulk density test results of the cultivated layer of farmland; the data obtained by the data collection module and the corresponding soil bulk density test results of the cultivated layer of farmland are stored and displayed.

Images