Ground penetrating radar antenna integrated on plant protection robot and real-time imaging method

Abstract

The invention discloses a ground penetrating radar antenna integrated on a plant protection robot and a real-time imaging method, and relates to the technical field of ground penetrating radar inversion, antenna design and deep learning. The method comprises the following steps: controlling a ground penetrating radar through two UR5e mechanical arms to obtain environment actual measurement data, and obtaining environment simulation data through a simulation model to obtain a training data set; based on the dynamic adjustable joint loss function, performing model training on the deep learning inversion model according to the training data set to obtain a ground penetrating radar inversion model; acquiring to-be-analyzed data of the to-be-analyzed environment through a multi-band ground penetrating radar antenna at the bottom of the plant protection robot, and performing ground penetrating radar inversion on the to-be-analyzed data according to the ground penetrating radar inversion model to obtain underground soil and root system real-time imaging of the to-be-analyzed environment. In this way, the ground penetrating radar inversion model is obtained through training of automatically collected data, the multi-band ground penetrating radar antenna is integrated in the center of the bottom of the plant protection robot, and it is ensured that underground detection and ground operation are conducted synchronously.

Description

Technical Field

[0001] The present invention relates to the fields of ground penetrating radar inversion, antenna design and deep learning technology, and in particular to a ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method. Background Art

[0002] In the traditional ground penetrating radar inversion process, the equipment that can realize real-time data processing is often difficult to effectively integrate with the plant protection robot due to its large size and weight. In contrast, although the lighter ground penetrating radar equipment is easy to integrate, it cannot realize real-time data upload and analysis due to the limitation of processing power. Therefore, the operation efficiency and intelligence level of the combination of ground penetrating radar and plant protection robot are greatly limited. Summary of the invention

[0003] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method, especially a real-time inversion imaging method for a plant protection robot integrated with a ground penetrating radar antenna. In view of the difficulties in ground penetrating radar data acquisition, the limitations of traditional inversion algorithms, and the difficulties in integrating ground penetrating radar equipment with plant protection robots, a variety of high-quality data are acquired in a high-precision and automated manner; a large number of training and test samples are provided for deep learning inversion models to improve the accuracy and computational efficiency of underground target recognition in complex environments; a variety of Bow-Tie antennas with different frequencies are designed, and their integration method in the center of the bottom of the plant protection robot is optimized to achieve miniaturization and flexibility of the equipment, ensuring the synchronization of underground detection and ground operations.

[0004] The present invention provides the following technical solutions:

[0005] In a first aspect, the present invention provides a ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method, the method comprising:

[0006] Using two UR5e robotic arms to control the ground penetrating radar to obtain environmental measured data of underground targets in a preset indoor experimental environment, where the preset indoor experimental environment is determined according to preset materials and preset space dimensions;

[0007] Simulating an underground soil scene through a simulation model and obtaining environmental simulation data of the underground soil scene;

[0008] Constructing a training data set according to the environmental measured data and the environmental simulation data;

[0009] Based on a dynamically adjustable joint loss function, a deep learning inversion model is trained according to the training data set to obtain a ground penetrating radar inversion model, wherein the dynamically adjustable joint loss function is constructed according to a preset morphological inversion loss, a preset electrical parameter inversion loss and a preset adjustable weight factor;

[0010] The data to be analyzed of the environment to be analyzed is obtained through the multi-band ground penetrating radar antenna at the bottom of the plant protection robot, and the data to be analyzed is subjected to ground penetrating radar inversion according to the ground penetrating radar inversion model to obtain real-time imaging of the underground soil and root system in the environment to be analyzed.

[0011] In one embodiment, the UR5e manipulator includes a data measurement manipulator and a scatterer operation manipulator, and the two UR5e manipulators are used to control the ground penetrating radar to obtain the measured environmental data of underground targets in a preset indoor experimental environment, including:

[0012] Constructing the preset indoor experimental environment by operating the scatterer as a robotic arm;

[0013] Determine the scanning path of the robot arm according to the preset indoor experimental environment structure and preset task requirements;

[0014] The data measurement robot arm controls the ground penetrating radar to scan the underground target in the preset indoor experimental environment according to the robot arm scanning path to obtain the environmental measured data.

[0015] In one embodiment, the step of constructing a training data set based on the environmental measured data and the environmental simulation data includes:

[0016] Comparing the measured data of the environment with the underground target parameters in the preset indoor experimental environment to obtain measured tag data;

[0017] Acquire simulation tag data according to the environmental simulation data;

[0018] The training data set is constructed according to the environmental measured data, the measured label data, the environmental simulation data and the simulation label data.

[0019] In one embodiment, constructing the training data set according to the environmental measured data, the measured label data, the environmental simulation data and the simulation label data includes:

[0020] Acquire a first corresponding relationship between the environmental measured data and the measured tag data;

[0021] Acquire a second corresponding relationship between the environment simulation data and the simulation label data;

[0022] Integrate the environmental measured data with the measured label data according to the first corresponding relationship to obtain a first data set;

[0023] Integrating the environment simulation data with the simulation label data according to the second corresponding relationship to obtain a second data set;

[0024] The training data set is obtained according to the first data set and the second data set.

[0025] In one embodiment, the method of training the deep learning inversion model based on the training data set based on the dynamically adjustable joint loss function to obtain the ground penetrating radar inversion model includes:

[0026] Preprocessing the training data set to obtain a preprocessed training set;

[0027] Based on the dynamically adjustable joint loss function, the deep learning inversion model is trained according to the preprocessed training set to obtain the ground penetrating radar inversion model;

[0028] The expression of the dynamically adjustable joint loss function is: L = λ 1 L shape +λ 2 L electrical ;

[0029] Where L is the value of the joint loss function, L shape is the morphological inversion loss value, L electrical is the electrical parameter inversion loss value, λ 1 and λ 2 is the preset adjustable weight factor.

[0030] In one embodiment, the deep learning inversion model is trained based on the joint loss function and the preprocessed training set to obtain the ground penetrating radar inversion model, including:

[0031] Based on the dynamically adjustable joint loss function, the deep learning inversion model is preliminarily trained according to the environmental simulation data in the preprocessed training set to obtain an initial inversion model;

[0032] Determining a model initial layer and a model fine-tuning layer according to the initial inversion model;

[0033] Based on a preset learning rate and the preset adjustable weight factor, the model fine-tuning layer is fine-tuned according to the measured environmental data in the preprocessed training set to obtain a model target layer;

[0034] The ground penetrating radar inversion model is obtained according to the model initial layer and the model target layer.

[0035] In one embodiment, preprocessing the training data set to obtain a preprocessed training set includes:

[0036] Denoising the training data set to obtain a denoised training set;

[0037] Normalizing the denoised training set to obtain a normalized training set;

[0038] The normalized training set is enhanced to obtain the preprocessed training set.

[0039] In one embodiment, the multi-band ground penetrating radar antenna includes a first frequency band antenna, a second frequency band antenna, a third frequency band antenna and a plurality of reflector structures; the frequency band of the first frequency band antenna is 900 MHz, the frequency band of the second frequency band antenna is 1.6 GHz, and the frequency band of the third frequency band antenna is 2.6 GHz;

[0040] The method further comprises: calculating the bow-tie arm lengths of the first frequency band antenna, the second frequency band antenna, and the third frequency band antenna according to the low frequency band wavelength and characteristic impedance;

[0041] The calculation formula of the bow-tie arm length is:

[0042] In the formula, l i is the bow-tie arm length of the ith frequency band antenna, is the characteristic impedance of the antenna in the ith frequency band, is the opening angle of the antenna in the i-th frequency band;

[0043] The first frequency band antenna, the second frequency band antenna and the third frequency band antenna are each provided with a corresponding reflector structure;

[0044] The method also includes: determining the wavelength of the i-th frequency band antenna according to the frequency band of the i-th frequency band antenna; determining the size of the reflector structure corresponding to the i-th frequency band antenna according to a first preset ratio and the wavelength of the i-th frequency band antenna; and determining the distance between the reflector structure corresponding to the i-th frequency band antenna and the vibrator of the i-th frequency band antenna according to a second preset ratio and the vibrator length of the i-th frequency band antenna.

[0045] In one embodiment, the method of obtaining the data to be analyzed of the environment to be analyzed by using the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot includes:

[0046] Based on the segmented acquisition mechanism, the scanning data of the preset road section is obtained through the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot;

[0047] The scanned data of the preset road section is determined as the data to be analyzed.

[0048] The ground-penetrating radar antenna and real-time imaging method integrated in the plant protection robot disclosed in the present invention can adapt to the detection needs of plant roots at different depths by using a multi-band ground-penetrating radar antenna; the ground-penetrating radar is controlled by a robotic arm to automatically collect indoor data, and can accurately collect multi-angle and multi-directional data according to a preset path, thereby reducing human errors and providing high-quality, standardized underground data, laying a foundation for subsequent inversion model training and testing; in addition, the ground-penetrating radar module is installed in the center of the bottom of the plant protection robot, and an integrated quick-release design is used to support the rapid replacement of antennas of different frequency bands, and the segmented acquisition mechanism can be used to transmit data in real time for inversion processing; at the same time, the ground-penetrating radar inversion model based on deep learning is combined with the underground target morphological loss and electrical parameter loss for inversion processing, thereby improving the model inversion accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope of protection of the present invention. In each of the drawings, similar components are numbered similarly.

[0050] Figure 1 A schematic diagram of a process of a ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method proposed in this embodiment is shown;

[0051] Figure 2 A schematic diagram of the structure of the UR5e robotic arm proposed in this embodiment is shown;

[0052] Figure 3 A schematic structural diagram of a tracked ground penetrating radar device proposed in this embodiment is shown;

[0053] Figure 4 Another schematic diagram of the process of the ground penetrating radar antenna integrated in the plant protection robot and the real-time imaging method proposed in this embodiment is shown;

[0054] Figure 5 Another schematic diagram of the process of the ground penetrating radar antenna integrated in the plant protection robot and the real-time imaging method proposed in this embodiment is shown;

[0055] Figure 6 A schematic diagram of a preset indoor experimental environment proposed in this embodiment is shown;

[0056] Figure 7 A schematic diagram of the 900 MHz (left), 1.6 GHz (middle) and 2.6 GHz (right) bow-tie antennas proposed in this embodiment is shown;

[0057] Figure 8A schematic diagram of the bow-tie antenna reflector proposed in this embodiment is shown;

[0058] Fig. 9 A schematic diagram of the structure of the multi-band ground penetrating radar antenna proposed in this embodiment is shown;

[0059] Fig.10 Another schematic diagram of the process of the ground penetrating radar antenna integrated in the plant protection robot and the real-time imaging method proposed in this embodiment is shown;

[0060] Fig.11 Another flow chart of the ground penetrating radar antenna integrated in the plant protection robot and the real-time imaging method proposed in this embodiment is shown. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0062] The components of the embodiments of the present invention generally described and shown in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0063] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present invention, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0064] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0065] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present invention belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present invention.

[0066] Example 1

[0067] Plant protection robots face many challenges in underground structure detection. Most of the current plant protection robots can only process surface information and have difficulty perceiving complex underground structures, such as the growth status of crop roots, which is directly related to the absorption capacity and health of plants. At the same time, changes in soil structure also have an important impact on crop growth. In addition, due to the dynamics and complexity of the agricultural environment (such as differences in soil texture, undulating terrain, uneven crop distribution, etc.), the perception ability of plant protection robots for underground targets is severely limited. Current sensing technologies and algorithms are mainly optimized for surface information, and lack system integration and application specifically for underground detection. This leads to a clear gap in the acquisition of underground structure information by plant protection robots, limiting their comprehensive application potential in precision agriculture.

[0068] However, the current ground-penetrating radar equipment that can achieve real-time data processing is usually large in size and heavy in weight, making it difficult for small or medium-sized plant protection robots to carry it; while lighter equipment, although easy to integrate, cannot achieve real-time data upload and processing, limiting its practicality in precision agriculture. Plant protection robots are usually designed to adapt to the soft soil of farmland and to move flexibly between crops. Their load capacity is limited, and overweight equipment may cause the robot to lose its center of gravity, sink into the soil, or damage crops. The installation height and angle of the antenna also need to be strictly controlled, but the limited space makes it difficult to meet its performance optimization requirements. These problems greatly limit the efficient integration of ground-penetrating radar and plant protection robots and the level of intelligent operation.

[0069] The disclosed embodiments provide a ground penetrating radar antenna and a real-time imaging method integrated in a plant protection robot, in particular, a ground penetrating radar antenna and a real-time imaging method integrated in a plant protection robot. The method aims to obtain diversified high-quality data in a high-precision and automated manner to address the difficulties in collecting ground penetrating radar data, the limitations of traditional inversion algorithms, and the difficulties in integrating ground penetrating radar equipment with plant protection robots. The method provides a large number of training and test samples for deep learning inversion models to improve the accuracy and computational efficiency of underground target recognition in complex environments. The method designs Bow-Tie antennas of multiple frequencies and optimizes their integration method at the bottom of the plant protection robot to achieve miniaturization and flexibility of the equipment and ensure the synchronization of underground detection and ground operations.

[0070] See also Figure 1 , a ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method, including steps S101 to S105, each step is described in detail below.

[0071] Step S101, using two UR5e robotic arms to control a ground penetrating radar to obtain environmental measured data of underground targets in a preset indoor experimental environment, wherein the preset indoor experimental environment is determined according to a preset material and a preset space size.

[0072] In this embodiment, two UR5e robotic arms are used to control the ground penetrating radar to obtain environmental measured data of underground targets in a preset indoor experimental environment. The multi-band ground penetrating radar antenna integrates multiple antennas of different frequency bands to cope with signal collection of different root depths, densities and distribution characteristics in the environment.

[0073] The environmental measured data are authentic and reliable, reflecting the noise, interference and soil conditions in the actual environment. Data acquisition is limited by experimental conditions (site size, soil type, etc.), but the diversity and complexity of the scenes are limited.

[0074] In order to fully simulate the real underground environment and reduce external interference, this embodiment designs and constructs a high-precision wooden test box as a preset indoor experimental environment, aiming to provide a controllable test platform for underground target detection experiments, while facilitating the robot arm to collect accurate data during the experiment. To achieve this goal, the test box uses dry wood, a material with an electrical conductivity of 10 -13 to 10 -12 S / m (Siemens / meter), the reflection and interference of electromagnetic waves are small, so the influence of the wooden box itself on the ground penetrating radar transmission signal can be reduced, ensuring the purity of the measurement signal, thereby ensuring more accurate signal collection and data analysis, especially when the robotic arm performs precision measurement, it can minimize the impact of external interference on the data.

[0075] The dimensions of the experimental box are 2 meters long, 1 meter wide, 0.8 meters high, and 0.6 meters deep in the soil, providing ample space to accommodate multiple scatterers and support multi-target measurements in complex scenarios. Due to the large space design of the experimental box, experiments can simultaneously collect multi-target detection data to meet the needs of different experimental conditions.

[0076] It should be noted that this embodiment uses the high-performance collaborative robot arm UR5e developed by Universal Robots. Figure 2 As shown in the figure. UR5e can achieve a repeatable positioning accuracy of ±0.03 mm, and the robot arm can accurately locate the target position to ensure the consistency of each data collection point, meeting the high requirements of ground penetrating radar for data accuracy; each joint can achieve ±360° rotation, covering multi-angle and multi-directional collection areas; the maximum working radius is 850 mm, which can cover the collection area with a diameter of more than 1.5 meters, meeting the data collection needs of indoor experimental sites; the load capacity is 5 kg, which can easily carry the ground penetrating radar antenna and related additional equipment; the whole machine weighs 20.6 kg. Compared with traditional industrial robot arms, UR5e is lighter and easier to install, debug and move in indoor experimental environments.

[0077] Step S102: simulating an underground soil scene through a simulation model, and obtaining environmental simulation data of the underground soil scene.

[0078] In the present embodiment, the simulation model is constructed using the simulation software gprMax, which uses the Finite-Difference Time-Domain (FDTD) method to solve Maxwell's equations. This calculation method can accurately simulate the propagation characteristics of electromagnetic waves in complex medium environments. Through computer simulation of electromagnetic wave propagation and reflection in underground scenes, virtual underground soil scenes can be created, including different strata, objects and structures, and underground targets of different shapes and dielectric constants, and the reflection, refraction and attenuation of electromagnetic waves can be observed. Environmental simulation data can simulate complex conditions that are difficult to obtain in a variety of actual scenes, providing a rich source of data for initial model training, while environmental simulation data is diversified, and scenes under complex and extreme conditions can be generated, providing a wide range of training data, which helps to improve the adaptability and generalization ability of the model to actual complex scenes, but lacks the real environmental noise and physical errors in the experimental site, and there is a certain distribution difference with the actual scene.

[0079] Step S103: construct a training data set according to the environmental measured data and the environmental simulation data.

[0080] In this embodiment, a training data set is constructed using environmental measured data and environmental simulation data for training a deep learning inversion model.

[0081] Step S104, based on the dynamically adjustable joint loss function, the deep learning inversion model is trained according to the training data set to obtain a ground penetrating radar inversion model. The dynamically adjustable joint loss function is constructed according to a preset morphological inversion loss, a preset electrical parameter inversion loss and a preset adjustable weight factor.

[0082] In this embodiment, based on the dynamically adjustable joint loss function, the deep learning inversion model is trained using the training data set to obtain the ground penetrating radar inversion model. Among them, the environmental simulation data in the training data set is mainly used for the initial training of the model, providing a large amount of basic data to help the model learn the basic characteristics and laws of underground targets; and the environmental measured data in the training data set is used for fine-tuning and verification of the model to ensure the adaptability and accuracy of the model in actual scenarios. As part of the training data, it is mixed with the simulation data to improve the generalization ability of the model to real scenarios.

[0083] It should be noted that the morphological inversion loss is usually based on the error calculation of the geometric morphology of the target (such as the difference between the predicted root depth, distribution density, etc. and the actual label), which is measured by cross entropy loss; the electrical parameter inversion loss mainly focuses on the error between the predicted physical properties such as dielectric constant and soil moisture and the actual label, which is measured by mean square error (MSE).

[0084] Furthermore, the morphological inversion loss and the electrical parameter inversion loss are combined, and a weighted sum is formed using preset adjustable weight factors to obtain a dynamically adjustable joint loss function. Through the dynamically adjustable joint loss function, the model can simultaneously optimize the predictions of these two aspects during the training process, ensuring that the final inversion result can accurately reflect the morphology of the target and accurately estimate its electrical parameters.

[0085] Step S105, obtaining the data to be analyzed of the environment to be analyzed through the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot, performing ground penetrating radar inversion on the data to be analyzed according to the ground penetrating radar inversion model, and obtaining real-time imaging of the underground soil and root system in the environment to be analyzed.

[0086] In this embodiment, when the plant protection robot is used to perform real-time ground penetrating inversion, the data to be analyzed of the environment to be analyzed is obtained through the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot.

[0087] Furthermore, ground penetrating radar inversion is performed on the data to be analyzed according to the ground penetrating radar inversion model to obtain underground distribution data in the environment to be analyzed. The underground distribution data mainly includes the morphological characteristics of plant roots, the depth and distribution of different soil layers, and the electrical parameters of soil and roots.

[0088] Furthermore, the predicted underground distribution data is visualized to obtain real-time imaging of underground soil and root systems, so that the robot can adjust its path, apply pesticides, irrigate and other operations to ensure precision agricultural operations. The plant protection robot continuously collects data and continuously adjusts its operation strategy based on real-time feedback to optimize the operation effect.

[0089] It should be noted that the multi-band ground-penetrating radar antenna is installed at the bottom center of the plant protection robot and uses a quick-release design to facilitate the replacement of antennas with different frequencies to meet the needs of underground plant root detection in various agricultural scenarios. This not only achieves efficient integration of ground-penetrating radar, but also greatly improves the maintenance and expansion capabilities of the equipment.

[0090] The multi-band ground penetrating radar antenna is fixed at the center of the bottom of the plant protection robot to ensure that the signal propagates vertically downward and covers the central area of ​​the operation path. The central position helps stabilize the center of gravity of the robot and avoid equipment imbalance caused by antenna bias. The installation height of the antenna is set at 5-10cm from the ground, taking into account the signal penetration depth and resolution to ensure the accuracy of data collection. Figure 3 This is a schematic diagram of the structure of a tracked ground penetrating radar device.

[0091] In addition, the multi-band ground penetrating radar antenna adopts a quick-release mounting design, using a snap-on or quick-plug interface to support quick installation and removal. The quick-release design makes the maintenance, replacement or frequency switching of the antenna more efficient, without the need for additional tools, greatly reducing maintenance time.

[0092] See also Figure 4 , the following is a comparison of the figures from multiple specific embodiments Figure 4 Each process is explained in detail.

[0093] See also Figure 5 In a specific embodiment, the UR5e manipulator includes a data measurement manipulator and a scatterer operation manipulator, and the data measurement manipulator, the scatterer operation manipulator and the multi-band ground penetrating radar antenna are integrated into a ground penetrating radar indoor measurement device. Step S101 includes steps S1011 to S1013, and each step is described in detail below.

[0094] Step S1011, constructing the preset indoor experimental environment by using the scatterer manipulation robot.

[0095] In this example, see Figure 6 The scatterer manipulation robot is responsible for the placement, burial and excavation of scatterers in the preset indoor experimental environment. It is equipped with a camera for real-time positioning and manipulation. Through the camera, the robot can observe the position of the scatterer in real time to ensure its accurate placement or recovery in the experimental box. The camera can provide real-time feedback to assist the robot in performing precise excavation operations to avoid damage to the scatterer or misoperation.

[0096] Step S1012, determining the robot arm scanning path according to the preset indoor experimental environment structure and preset task requirements.

[0097] In this embodiment, the scanning path of the robot arm is planned according to the environmental structure of the preset indoor experimental environment and the acquisition task requirements.

[0098] Step S1013, controlling the ground penetrating radar to scan underground targets in the preset indoor experimental environment according to the scanning path of the data measurement robot arm to obtain the environmental measured data.

[0099] In this embodiment, the data measurement robot is responsible for controlling the movement and positioning of the ground penetrating radar. The end effector of the robot maintains a stable height and angle, consistent with the ground or target object, pre-defines the starting point position and the target point position, ensures that the robot accurately reaches the target point according to the determined scanning path, and realizes accurate data collection of the preset indoor experimental environment to obtain the actual measured data of the environment. The collected data can be processed into npy format by software. The ground penetrating radar here can be a common type of radar.

[0100] The data measurement robot is equipped with a high-definition camera and a high-precision control system to ensure the consistency of each measurement point and generate standardized high-quality B-Scan data. The main function of the camera is to provide real-time visual feedback to the robot arm to help confirm the precise position of the antenna and ensure that it collects data accurately along the predetermined path. With the assistance of the camera, the robot arm can dynamically adjust its posture to avoid deviations or errors during the measurement process and ensure the accuracy of each measurement point. In addition, the camera can also be used to monitor the experimental environment, detect possible obstacles or interference sources in real time, and ensure the purity and high quality of the ground penetrating radar data.

[0101] It should be noted that the camera of the scatterer manipulation arm can also work in conjunction with the system of the data measurement arm to ensure the consistency of the target position and measurement task during the entire experiment. At the front end of the scatterer manipulation arm, a Kantman tool is used as an excavation tool, which can accurately perform underground target burial and excavation tasks, ensuring high precision and efficiency of the operation.

[0102] In order to simulate environmental conditions of different complexity, the test box constructs three different types of soil structures. The first is uniform soil, which provides a basic test environment and verifies the performance of the equipment. The second is non-uniform soil, which simulates the complex medium distribution in actual farmland. The third is layered soil, which tests the system's ability to detect changes in underground structures. The robotic arm is installed on a high-strength adjustable base to ensure stability and flexibility. The base position can be moved to adapt to different test scenarios. The collaborative work of the two robotic arms not only improves the efficiency of the experiment, but also ensures the repeatability of the experimental results and the high accuracy of the data.

[0103] In a specific embodiment, the multi-band ground penetrating radar antenna includes a first frequency band antenna, a second frequency band antenna, a third frequency band antenna and a plurality of reflector structures; the frequency band of the first frequency band antenna is 900 MHz, the frequency band of the second frequency band antenna is 1.6 GHz, and the frequency band of the third frequency band antenna is 2.6 GHz; the method further includes: calculating the bow-tie arm lengths of the first frequency band antenna, the second frequency band antenna and the third frequency band antenna according to the low frequency band wavelength and characteristic impedance; wherein the calculation formula of the bow-tie arm length is: In the formula, li is the bow-tie arm length of the ith frequency band antenna, is the characteristic impedance of the antenna in the ith frequency band, is the opening angle of the antenna in the i-th frequency band.

[0104] In this embodiment, the multi-band ground penetrating radar antenna includes a first-band antenna, a second-band antenna, a third-band antenna, and a plurality of reflector structures, wherein the frequency band of the first-band antenna is 900 MHz, the frequency band of the second-band antenna is 1.6 GHz, and the frequency band of the third-band antenna is 2.6 GHz. The three-band antenna can meet the needs of effectively detecting plant roots at different depths.

[0105] Among them, the 1st band antenna, 2nd band antenna, and 3rd band antenna are all Bow-Tie antennas. Figure 7 The antenna adopts a symmetrical butterfly-shaped structure and optimizes the frequency response by adjusting the opening angle. The opening angle is designed to be between 60° and 90° to meet the requirements of wide-band characteristics. The arm length affects the low-band frequency of the antenna. The larger the value, the better the low-frequency coverage performance of the bow-tie antenna. The relationship between the bow-tie arm length and the low-frequency wavelength of the i-th frequency band antenna is: In the formula, l i is the bow-tie arm length of the ith frequency band antenna, is the characteristic impedance of the antenna in the ith frequency band, is the opening angle of the antenna in the i-th frequency band.

[0106] It should be noted that for the 2.6GHz antenna (0-0.3m), the antenna in this frequency band is suitable for shallow detection (0-0.3m), can provide higher resolution, and is suitable for detecting shallow plant roots. Higher frequency brings better resolution, but its penetration depth is shallower. The arm length is about 2.9cm, and the opening angle is about 60°-90°.

[0107] For 1.6GHz antenna (0.3-0.5m), this frequency band antenna is used to detect plant roots at a depth of 0.3-0.5 meters, taking into account both good resolution and deep penetration, and is suitable for medium-depth root detection. The arm length l is about 4.7cm, and the opening angle θ is about 70°-85°.

[0108] For 900MHz antenna (0.5-1m), this frequency band antenna is used for deeper (0.5-1m) plant root detection, which can provide strong penetration ability and is suitable for detecting deeper soil layers, but the resolution is relatively low. The arm length is about 8.3cm and the opening angle is about 80°-90°.

[0109] In a specific embodiment, the first frequency band antenna, the second frequency band antenna and the third frequency band antenna are respectively provided with a corresponding reflector structure; the method also includes: determining the wavelength of the i-th frequency band antenna according to the frequency band of the i-th frequency band antenna; determining the size of the reflector structure corresponding to the i-th frequency band antenna according to a first preset ratio and the wavelength of the i-th frequency band antenna; determining the distance between the reflector structure corresponding to the i-th frequency band antenna and the vibrator of the i-th frequency band antenna according to a second preset ratio and the vibrator length of the i-th frequency band antenna.

[0110] It should be noted that the planar Bow-Tie antenna has a bidirectional radiation characteristic, that is, there is a field distribution of the same intensity in front and behind. If the radiation in one direction is not effectively suppressed, the radiation in the rear will lead to the loss of the ground penetrating radar radiation efficiency and weaken the ability of electromagnetic energy to concentrate in the target detection direction. In addition, bidirectional radiation will also cause the clutter from the surrounding scatterers to be confused with the echo signal in the target direction, increase the complexity of the later waveform processing and the risk of misjudgment, thereby affecting the detection accuracy and effectiveness of the radar target. Therefore, in order to improve the radiation efficiency and signal quality of the ground penetrating radar, it is necessary to focus on the energy emission and echo reception in the target direction by designing a reflective surface or other radiation suppression measures.

[0111] In this embodiment, a metal reflector is installed on the back of each antenna parallel to the vibrator plane. The size of the reflector is larger than the maximum size of the antenna vibrator to ensure that the reflection effect of the electromagnetic wave covers the entire antenna working range. The reflector structure suppresses the lateral and backward radiation signals, further improving the directivity and signal-to-noise ratio. The space formed is filled with absorbing materials to reduce the self-reflection of the reflective surface. The schematic diagram is as follows Figure 8 shown.

[0112] Among them, the wavelength of the i-th frequency band antenna is determined according to the frequency band of the i-th frequency band antenna; the size of the reflector structure corresponding to the i-th frequency band antenna is determined according to a first preset ratio and the wavelength of the i-th frequency band antenna; the distance between the reflector structure corresponding to the i-th frequency band antenna and the vibrator of the i-th frequency band antenna is determined according to a second preset ratio and the vibrator length of the i-th frequency band antenna.

[0113] Exemplarily, the distance between the reflector and the antenna element is 1 / 4 wavelength of the length of the element.

[0114] At the same time, 2.6GHz is a higher frequency band, so a smaller reflective surface is required to effectively reflect electromagnetic waves. Due to the high frequency, the reflective surface needs to be designed to ensure that it can fully reflect signals with shorter wavelengths to increase the forward gain of the antenna. For 2.6GHz antennas, the wavelength is about 0.115m. In order to effectively reflect the signal, the size of the reflective surface is usually 3 to 5 times the wavelength, about 0.35m to 0.58m long, about 0.2m to 0.3m wide, and 0.05m to 0.1m high. For 1.6GHz antennas, the wavelength is about 0.1875m. In order to effectively reflect the signal, the size of the reflective surface is usually 4 to 6 times the wavelength, about 0.75m to 1.13m long, about 0.4m to 0.6m wide, and 0.08m to 0.12m high. For 900MHz antennas, the wavelength is about 0.333m. In order to effectively reflect the signal, the size of the reflecting surface is usually 5 to 7 times the wavelength, about 1.67m to 2.33m long, about 0.6m to 1m wide, and 0.1m to 0.15m high.

[0115] It should be noted that the overall structure of the multi-band ground penetrating radar antenna consists of a host, a transmitting path, a receiving path, a data processing module and a real-time inversion module. Fig. 9 shown.

[0116] The host module is the core control unit of the system, which is used to coordinate and manage the workflow of various parts of the system. It triggers the operation of the transmission path through control signals, receives signal data from the receiving path, and passes the data to the data processing module for analysis. The functions of the host module include: (1) Control signal generation: trigger the transmission signal shaping circuit to generate high-frequency electromagnetic waves. (2) Signal data aggregation: receive and integrate sampled signals from the receiving path to ensure signal integrity. (3) Data transmission: pass the processed data to the data processing module to support subsequent inversion and visualization.

[0117] The transmission path is responsible for converting the control signal generated by the host into high-frequency electromagnetic waves and transmitting them to the underground target area. Its components include: (1) Transmission signal shaping circuit: Shapes the control signal output by the host, optimizes the signal waveform, and ensures that the frequency and amplitude of the signal meet the requirements of the transmitting antenna. (2) Amplification module: Amplifies the power of the shaped signal to enhance the signal strength to ensure that the signal can penetrate deeper soil layers. (3) Limiting circuit: Limits the signal amplitude to prevent excessively strong signals from damaging subsequent modules or transmitting antennas. (4) Transmitting antenna: Uses a Bow-Tie antenna to convert the processed high-frequency electrical signal into electromagnetic waves and transmit it underground to complete the signal irradiation of underground targets.

[0118] The receiving path is used to capture the electromagnetic wave signal reflected from the underground target and process and amplify the signal. Its components include: (1) Receiving antenna: The electromagnetic wave signal reflected from the underground target is received by the receiving antenna and converted into an electrical signal. (2) Sampling and mixing module: Mixing the received high-frequency signal and converting it into a lower-frequency signal for subsequent processing. (3) Low-frequency amplification circuit: Amplifying the mixed low-frequency signal to enhance the signal strength and improve the signal-to-noise ratio. (4) Sampling signal amplification and shaping circuit: Amplifying the weak signal output by the sampling and mixing module to enhance the signal strength.

[0119] The data processing module is the core computing and analysis unit of the ground penetrating radar system, responsible for digitizing, processing and analyzing the signal data provided by the receiving path. The main functions include: (1) Analog-to-digital conversion unit (ADC): digitizes the analog signal output by the receiving path to generate a processable digital signal. (2) Preprocessing module: performs preliminary processing on the digital signal to improve the signal quality. (3) Storage and transmission module: transmits the processed data to the subsequent inversion part.

[0120] In a specific embodiment, step S103 includes: obtaining measured label data based on the environmental measured data and the underground target parameters in the preset indoor experimental environment; obtaining simulation label data based on the environmental simulation data; and constructing the training data set based on the environmental measured data, the measured label data, the environmental simulation data and the simulation label data.

[0121] In this embodiment, the measured environmental data and the simulated environmental data are B-Scan images, which are visualization results of signal data, showing the information of radar waves reflected from underground targets, and can reflect the depth, shape, and relative position of underground objects with respect to other underground objects or media. The label data is used to mark the real properties of the underground objects corresponding to each signal point in the B-Scan image. The label data is usually the actual physical properties of the underground target, such as the depth, shape, and position of the target. These labels can be used to train the model so that the model can learn how to extract and predict the relevant features of the underground target from the B-Scan image.

[0122] Specifically, when using gprMax to generate an underground soil environment model, the generated environment simulation data will contain known information about the underground structure. For each signal point, the true properties of the corresponding underground objects (such as roots, soil layers) are known, so this information can be directly used as simulation label data.

[0123] During the measurement process, the signals in the measured data can be annotated by comparing them with the parameters of known underground targets in a preset indoor experimental environment (for example, using drilling technology or directly observing underground targets), thereby obtaining measured label data. For example, measuring the depth of the root system, or using sensors to detect soil moisture.

[0124] In a specific embodiment, the training data set is constructed based on the environmental measured data, the measured label data, the environmental simulation data and the simulated label data, including: obtaining a first correspondence between the environmental measured data and the measured label data; obtaining a second correspondence between the environmental simulation data and the simulated label data; integrating the environmental measured data with the measured label data according to the first correspondence to obtain a first data set; integrating the environmental simulation data with the simulation label data according to the second correspondence to obtain a second data set; and obtaining the training data set based on the first data set and the second data set.

[0125] In this embodiment, the environment measured data is matched with the measured label data to obtain a first corresponding relationship; the environment simulation data is matched with the simulation label data to obtain a second corresponding relationship.

[0126] According to the first corresponding relationship, the measured environmental data and the measured label data are integrated to obtain a first data set; according to the second corresponding relationship, the simulated environmental data and the simulated label data are integrated to obtain a second data set; and the first data set and the second data set are combined to obtain a training data set.

[0127] See also Fig.10 In a specific embodiment, step S104 includes steps S1041 to S1042, and each step is described in detail below.

[0128] Step S1041, preprocessing the training data set to obtain a preprocessed training set.

[0129] In this embodiment, a plurality of preprocessing operations are performed on the training data set to obtain a preprocessed training set, so as to improve the efficiency and accuracy of model training.

[0130] In a specific embodiment, step S1041 includes: denoising the training data set to obtain a denoised training set; normalizing the denoised training set to obtain a normalized training set; and enhancing the normalized training set to obtain the preprocessed training set.

[0131] In this embodiment, the training data set is cleaned and denoised to remove irrelevant noise or abnormal data to obtain a denoised training set to ensure the accuracy of the signal; the denoised training set is standardized and normalized to adapt it to the input requirements of the model and improve the training efficiency; the normalized training set is data enhanced (such as rotation, translation, scaling, etc.) to increase the diversity of the training data and prevent overfitting.

[0132] Step S1042: Based on the dynamically adjustable joint loss function, the deep learning inversion model is trained according to the preprocessed training set to obtain the ground penetrating radar inversion model.

[0133] In this embodiment, the preprocessed training set is used as input, which includes radar transmission signals and target reflection signals.

[0134] The deep learning inversion model uses multi-layer convolution to extract local features based on the preprocessed training set, and then uses the pyramid convolution structure to extract features of different scales. The Transformer module is then introduced to capture the global correlation of different positions in the B-Scan image. The self-attention mechanism can enhance the understanding of the relationship between long-distance targets in complex scenes and solve the problem of insufficient capture of long-distance dependencies by traditional CNN. In the decoding part, the multi-scale features of the feature extraction module are integrated, and the geometric shape and electrical parameters of the target are gradually reconstructed through upsampling.

[0135] At the same time, a dynamically adjustable joint loss function is used to combine the morphological inversion loss and the electrical parameter inversion loss, and the balance between the two is adjusted by weights to improve the overall performance of the ground penetrating radar inversion model. The Adam optimizer is used in combination with a dynamic learning rate adjustment strategy to quickly converge to the optimal parameters, and regularization techniques (such as L2 regularization and Dropout) are used to prevent model overfitting.

[0136] Among them, the expression of the dynamically adjustable joint loss function is: L = λ 1 L shape +λ 2 L electrical ; Where L is the value of the joint loss function, L shape is the morphological inversion loss value, L electrical is the electrical parameter inversion loss value, λ 1 and λ 2 is a preset adjustable weight factor.

[0137] In a specific embodiment, step S1042 includes: based on the dynamically adjustable joint loss function, preliminarily training the deep learning inversion model according to the environmental simulation data in the preprocessed training set to obtain an initial inversion model; determining the model initial layer and the model fine-tuning layer according to the initial inversion model; based on the preset learning rate and the preset weight factor, fine-tuning the model fine-tuning layer according to the actual measured environmental data in the preprocessed training set to obtain the model target layer; obtaining the ground penetrating radar inversion model according to the model initial layer and the model target layer.

[0138] In this embodiment, based on the dynamically adjustable joint loss function, the deep learning inversion model is preliminarily trained according to the environmental simulation data in the preprocessed training set to obtain an initial inversion model.

[0139] Further, the model initial layer and the model fine-tuning layer are determined according to the initial inversion model. The model initial layer may include a low-level feature extraction layer, and the remaining layers are model fine-tuning layers. Only the latter layers are fine-tuned, which can reduce the amount of calculation and speed up the training.

[0140] Furthermore, the learning rate is adjusted, and a smaller learning rate is used to fine-tune the network to avoid excessive parameter updates that lead to loss of previously learned knowledge; and the weight of the loss function is adjusted according to the task requirements, that is, the preset adjustable weight factor is adjusted. At the same time, more environmental measured data in the preprocessed training set is introduced for model fine-tuning, especially data in actual agricultural environments, which helps the model better adapt to complex real-world scenarios. At this time, samples of different depth levels, soil types, and crop types can also be added to enhance the generalization ability of the model.

[0141] It should be noted that when adjusting the weight of the loss function, in order to ensure the balance between morphological inversion and electrical parameter inversion during training, a larger λ can be set at the beginning of training. 1 Prioritize optimization of morphological inversion (e.g., root morphology, distribution, etc.) to ensure that the model can accurately predict the geometric morphology of underground targets. As training progresses, λ can be gradually increased. 2 This makes the optimization of electrical parameter inversion more important. This helps to optimize the physical properties of underground targets in the later stage of the model and avoid the model being too biased in one task. During the training process, the adaptive dynamic adjustment of λ 1 and λ 2 The value of is used to adjust the weight ratio in real time according to the change of the loss function to ensure the balance during training.

[0142] In a specific embodiment, step S105 includes: based on a segmented acquisition mechanism, acquiring scanning data of a preset section through a multi-band ground penetrating radar antenna at the bottom center of the plant protection robot; and determining the scanning data of the preset section as the data to be analyzed.

[0143] In this embodiment, a segmented acquisition mechanism is used for data acquisition. When the plant protection robot is moving along the path, the multi-band ground penetrating radar antenna collects the reflected signals of the underground plant roots in the environment to be analyzed in real time. After each section of the path is completed (for example, 2-3 meters), the system stores the scan data collected from the preset section as independent B-Scan data, and transmits it to the computing unit inside the robot in real time through a cable as the data to be analyzed. The computing unit can then use the ground penetrating radar inversion model for inversion processing. This mechanism realizes the structured storage and real-time upload of data, providing an efficient and reliable foundation for subsequent real-time inversion processing.

[0144] In the inversion processing stage, the plant protection robot relies on the high-performance platform embedded in it, which is the NVIDIA Jetson Orin NX platform. It can run the self-trained ground-penetrating radar inversion model and quickly analyze the B-Scan data collected in real time. The ground-penetrating radar inversion model is used to extract and analyze the features of underground targets, and invert the depth, density and spatial distribution of the root system, the dielectric constant distribution of the soil, the layered structure and other underground distribution data. At the same time, the NVIDIA Jetson Orin NX is connected to an external high-resolution display to present the inversion results in real time for the operator to observe and analyze. The process is as follows: Fig.11 shown.

[0145] The close collaboration between data collection and inversion processing and the real-time display function form a closed-loop process, enabling the plant protection robot to dynamically generate underground target information during movement and provide real-time feedback to the operation interface. This design significantly improves operation efficiency and accuracy, and provides strong technical support for precision agricultural operations.

[0146] The ground penetrating radar antenna and real-time imaging method integrated in the plant protection robot proposed in this embodiment can adapt to the detection needs of plant roots at different depths by using a multi-band ground penetrating radar antenna; the ground penetrating radar is automatically collected indoors by controlling the mechanical arm, and multi-angle and multi-directional data can be accurately collected according to a preset path, reducing human errors, providing high-quality and standardized underground data, and laying a foundation for subsequent inversion model training and testing; in addition, the multi-band ground penetrating radar antenna is installed at the bottom center of the plant protection robot, and an integrated quick-release design is used to support the rapid replacement of antennas of different frequency bands, and the segmented acquisition mechanism can be used to transmit data in real time for inversion processing; at the same time, the ground penetrating radar inversion model based on deep learning is combined with the underground target morphological loss and electrical parameter loss for inversion processing, thereby improving the model inversion accuracy.

[0147] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0148] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0149] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A ground penetrating radar antenna integrated in a plant protection robot and a real-time imaging method, characterized in that: The method comprises: Using two UR5e robotic arms to control the ground penetrating radar to obtain environmental measured data of underground targets in a preset indoor experimental environment, where the preset indoor experimental environment is determined according to preset materials and preset space dimensions; Simulating an underground soil scene through a simulation model and obtaining environmental simulation data of the underground soil scene; Constructing a training data set according to the environmental measured data and the environmental simulation data; Based on a dynamically adjustable joint loss function, a deep learning inversion model is trained according to the training data set to obtain a ground penetrating radar inversion model, wherein the dynamically adjustable joint loss function is constructed according to a preset morphological inversion loss, a preset electrical parameter inversion loss and a preset adjustable weight factor; The data to be analyzed of the environment to be analyzed is obtained through the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot, and the data to be analyzed is subjected to ground penetrating radar inversion according to the ground penetrating radar inversion model to obtain real-time imaging of the underground soil and root system in the environment to be analyzed.

2. The ground penetrating radar antenna and real-time imaging method integrated in a plant protection robot according to claim 1, characterized in that: The UR5e robotic arm includes a data measurement robotic arm and a scatterer operation robotic arm. The two UR5e robotic arms are used to control the ground penetrating radar to obtain the measured environmental data of underground targets in a preset indoor experimental environment, including: Constructing the preset indoor experimental environment by operating the scatterer as a robotic arm; Determine the scanning path of the robot arm according to the preset indoor experimental environment structure and preset task requirements; The data measurement robot arm controls the ground penetrating radar to scan the underground target in the preset indoor experimental environment according to the robot arm scanning path to obtain the environmental measured data.

3. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 1, characterized in that: The step of constructing a training data set according to the environmental measured data and the environmental simulation data includes: Comparing the measured data of the environment with the underground target parameters in the preset indoor experimental environment to obtain measured tag data; Acquire simulation tag data according to the environmental simulation data; The training data set is constructed according to the environmental measured data, the measured label data, the environmental simulation data and the simulation label data.

4. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 3, characterized in that: The step of constructing the training data set according to the environmental measured data, the measured label data, the environmental simulation data and the simulation label data includes: Acquire a first corresponding relationship between the environmental measured data and the measured tag data; Acquire a second corresponding relationship between the environment simulation data and the simulation label data; Integrate the environmental measured data with the measured label data according to the first corresponding relationship to obtain a first data set; Integrating the environment simulation data with the simulation label data according to the second corresponding relationship to obtain a second data set; The training data set is obtained according to the first data set and the second data set.

5. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 1, characterized in that: The method of training the deep learning inversion model based on the dynamically adjustable joint loss function according to the training data set to obtain the ground penetrating radar inversion model includes: Preprocessing the training data set to obtain a preprocessed training set; Based on the dynamically adjustable joint loss function, the deep learning inversion model is trained according to the preprocessed training set to obtain the ground penetrating radar inversion model; The expression of the dynamically adjustable joint loss function is: L = λ1L shape +λ2L electrical ; Where L is the value of the joint loss function, L shape is the morphological inversion loss value, L electrical is the electrical parameter inversion loss value, λ1 and λ2 are the preset adjustable weight factors.

6. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 5, characterized in that: The method of performing model training on the deep learning inversion model based on the joint loss function and the pre-processed training set to obtain the ground penetrating radar inversion model includes: Based on the dynamically adjustable joint loss function, the deep learning inversion model is preliminarily trained according to the environmental simulation data in the preprocessed training set to obtain an initial inversion model; Determining a model initial layer and a model fine-tuning layer according to the initial inversion model; Based on a preset learning rate and the preset adjustable weight factor, the model fine-tuning layer is fine-tuned according to the measured environmental data in the preprocessed training set to obtain a model target layer; The ground penetrating radar inversion model is obtained according to the model initial layer and the model target layer.

7. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 5, characterized in that: The preprocessing of the training data set to obtain a preprocessed training set includes: Denoising the training data set to obtain a denoised training set; Normalizing the denoised training set to obtain a normalized training set; The normalized training set is enhanced to obtain the preprocessed training set.

8. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 1, characterized in that: The multi-band ground penetrating radar antenna comprises a first-band antenna, a second-band antenna, a third-band antenna and a plurality of reflector structures; the frequency band of the first-band antenna is 900 MHz, the frequency band of the second-band antenna is 1.6 GHz, and the frequency band of the third-band antenna is 2.6 GHz; The method further comprises: calculating the bow-tie arm lengths of the first frequency band antenna, the second frequency band antenna, and the third frequency band antenna according to the low frequency band wavelength and characteristic impedance; The calculation formula of the bow-tie arm length is: In the formula, l i is the bow-tie arm length of the ith frequency band antenna, is the characteristic impedance of the antenna in the ith frequency band, is the opening angle of the antenna in the i-th frequency band; The first frequency band antenna, the second frequency band antenna and the third frequency band antenna are each provided with a corresponding reflector structure; The method also includes: determining the wavelength of the i-th frequency band antenna according to the frequency band of the i-th frequency band antenna; determining the size of the reflector structure corresponding to the i-th frequency band antenna according to a first preset ratio and the wavelength of the i-th frequency band antenna; and determining the distance between the reflector structure corresponding to the i-th frequency band antenna and the vibrator of the i-th frequency band antenna according to a second preset ratio and the vibrator length of the i-th frequency band antenna.

9. The ground penetrating radar antenna and real-time imaging method integrated into a plant protection robot according to claim 1, characterized in that: The method of obtaining the data to be analyzed of the environment to be analyzed by using the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot includes: Based on the segmented acquisition mechanism, the scanning data of the preset road section is obtained through the multi-band ground penetrating radar antenna at the bottom center of the plant protection robot; The scanned data of the preset road section is determined as the data to be analyzed.

Images