Near-earth area underground crop detection system and positioning method based on continuous microwaves

Abstract

The invention belongs to the technical field of underground crop detection, and particularly relates to a near-earth area underground crop detection system and positioning method based on continuous microwaves, and the system comprises an antenna module and a signal system module. The antenna module comprises a detection structure and a flow blocking groove, the detection structure comprises two antenna structures, one antenna structure is used for transmitting signals, and the other antenna structure is used for receiving signals; the target information of the near-earth area is obtained through the detection structure in a scanning detection mode, classification identification is carried out by analyzing the features of the target information and combining the k-means algorithm, whether the target exists in the near-earth area or not is effectively identified, and the feasibility of the method is verified through experiments. And meanwhile, by collecting a distance parameter x, establishing a relationship between the distance parameter x and an actual distance and analyzing an output characteristic Uout of whether a detection target exists or not, an effective trough point is found, so that the positioning of the near-earth region target is realized.

Description

Technical Field

[0001] This invention belongs to the field of underground crop detection technology, specifically a near-ground underground crop detection system and positioning method based on continuous microwave. Background Technology

[0002] With the increasing demand from fields such as agriculture, ecological environmental protection, archaeology, and disaster search and rescue, the importance of underground target detection technology is becoming increasingly prominent. Especially in agriculture, accurately acquiring the status of underground biological targets is of great significance for improving crop yields and resource utilization efficiency. However, existing underground target detection technologies mostly rely on manual judgment and traditional equipment, resulting in limitations such as low detection efficiency, low accuracy, and complex data processing.

[0003] Most existing microwave positioning methods involve transmitting pulsed waves, which interact with the target object to obtain echo signals. By analyzing these echo signals, the target's location can be accurately determined, allowing researchers to visually analyze the target's distribution area. This method has strong interpretability and is widely used in fields such as underground target detection, medical positioning, and archaeology. However, to achieve high-precision positioning, microwave positioning systems rely on multi-band signal processing and complex reconstruction algorithms, which not only places high demands on hardware design but also requires powerful computing resources to support data processing.

[0004] CN1 12712132A discloses a method, apparatus, and system for detecting winter bamboo shoots based on microwave reflection. The method involves acquiring sample calibration data as a training dataset for a convolutional neural network model. The sample calibration data includes a detection dataset showing the presence of winter bamboo shoot samples and a detection dataset showing the absence of winter bamboo shoot samples. The training dataset is preprocessed to obtain a preprocessed training dataset. This preprocessed training dataset is then input into the convolutional neural network model for training, resulting in a reflection echo classification model. A region to be detected is determined. Real-time detection data within the region to be detected is used as input to the reflection echo classification model to obtain a classification result indicating whether winter bamboo shoots exist within the region to be detected.

[0005] CN1 18884567A discloses a high-precision method for detecting and reconstructing the 3D morphology of underground winter bamboo shoots. The method includes: measuring ground voltage data using a resistivity detection module to conduct large-scale preliminary screening of the underground growth location of winter bamboo shoots and initially delineating key areas; performing refined 3D electromagnetic detection on the key areas of underground winter bamboo shoots using a ground-penetrating radar module to obtain three-dimensional electromagnetic field data; scanning the exposed winter bamboo shoots in the area after refined 3D electromagnetic detection using a laser scanning module to measure the true three-dimensional morphology of the winter bamboo shoots and obtain three-dimensional image data; and based on the collected three-dimensional electromagnetic field data, voltage data, and three-dimensional image data, establishing spatial location prediction and 3D morphology prediction models using machine learning algorithms, and predicting the 3D morphology and spatial location of the underground winter bamboo shoots using these models.

[0006] CNl19717025A discloses a method for determining the optimal microwave transmission signal frequency band of a winter bamboo shoot detection device. This method involves measuring the moisture content and dielectric properties of winter bamboo shoots and the soil within the bamboo forest, analyzing the variation patterns of the dielectric properties of winter bamboo shoots and soil at different microwave signal frequencies, and the variation of the dielectric properties of the soil under different moisture contents. The difference in moisture content between winter bamboo shoots and soil is transformed into a difference in dielectric properties. Based on the measured dielectric properties, the frequency range with the greatest dielectric loss factor between winter bamboo shoots and soil is analyzed and used as the antenna bandwidth. A winter bamboo shoot-soil dielectric model under different moisture contents is constructed to analyze its influence on the S-parameters of the antenna transmission signal, thereby determining the center frequency of the optimal microwave transmission signal band for the winter bamboo shoot detection device.

[0007] Compared to the microwave positioning methods mentioned above, continuous wave detection systems are more attractive in some applications due to their lower cost, simpler structure, and easier operation. However, continuous wave detection can only provide limited energy information, and its detection accuracy and target resolution are relatively low, making it difficult to meet the needs of high-precision target positioning.

[0008] Therefore, the present invention provides a near-ground underground crop detection system and positioning method based on continuous microwave. Summary of the Invention

[0009] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0010] The technical solution adopted by this invention to solve its technical problem is as follows: The near-ground underground crop detection system based on continuous microwave of this invention includes an antenna module and a signal system module; the antenna module includes a detection structure and a flow-blocking groove, the detection structure includes two antenna structures, one of which is used to transmit signals and the other is used to receive signals; the antenna structure includes a metal shielding cavity, a coaxial feed, and an antenna dipole; the outer wall of the metal shielding cavity is made of a four-sided folded metal plate; the coaxial feed introduces the signal through a central coaxial line and is connected to the antenna dipole; the antenna dipole uses a butterfly-shaped dipole as the core radiating element; a support rod structure is provided between the antenna dipoles, the support rod structure includes an outer core and an inner core, the outer core is directly connected to one side of the antenna dipole through a metal connector, one end of the inner core is provided with a polytetrafluoroethylene connector, and the inner core is connected to the other side of the antenna dipole; a short-circuit ring structure is provided in the middle of the antenna dipole; the flow-blocking groove is provided between the outer walls of the two metal shielding cavities.

[0011] Preferably, the signal system module includes a signal synthesis unit and a signal processing unit. The signal synthesis unit includes a frequency control circuit, a frequency synthesis circuit, and a power amplification circuit. The signal processing unit includes a power amplification circuit, a radio frequency detection circuit, and a display output circuit. A support column is fixedly connected to the top of the choke groove. A mounting bracket is fixedly connected to the support column, and a power supply is installed on the mounting bracket via a snap-fit. A signal mounting plate is also fixedly connected to the support column, and the signal system module is mounted on the signal mounting plate.

[0012] Preferably, a near-ground crop detection and location method based on continuous microwaves includes the following steps:

[0013] S1: Bury the target in the soil near the ground, fix the detection structure on the crossbeam of the movable slide rail for horizontal scanning detection, collect the complete electromagnetic wave signal of the target, and perform signal preprocessing.

[0014] S2: Through k-means clustering analysis, determine the signal category of whether there is a detection target in the near-ground area, and establish a signal feature classifier to classify the different extracted feature vectors, thereby obtaining a classification model; where k-means clustering includes the training process and the actual identification process;

[0015] S3: The location of the target is determined using an LLM-driven localization algorithm.

[0016] Preferably, the signal preprocessing method includes the following steps:

[0017] S101: The detection and scanning are performed by moving the detection structure on the crossbeam of the slide rail, while simultaneously acquiring the detection output characteristic U.out The parameter x is related to distance;

[0018] S102: Process the collected data and convert it from the time domain to the spatial domain representation;

[0019] S103: The moving average smoothing method is used to reduce signal noise, which effectively reduces signal noise while preserving target features, providing a data basis for target detection.

[0020] S104: The relationship between the actual distance d and the distance parameter x can be derived from the actual measurement.

[0021] Preferably, the training process includes the following steps:

[0022] S2011: Extract features from the collected electromagnetic wave signals to obtain feature vectors that reflect the characteristics of the detected target;

[0023] S2012: When a target is present in the near-ground region, the signal is significantly weaker near the target than when there is no target, and is represented by the signal mean μ and the signal standard deviation σ.

[0024] The signal mean μ is used to reflect the overall signal strength, and its expression is:

[0025]

[0026] Where s(m) is the obtained electromagnetic wave signal, and M is the number of signal sampling points used for feature extraction, M = 3000;

[0027] The standard deviation σ of the signal is used to reflect information about the amplitude of signal fluctuations, and its expression is:

[0028]

[0029] Preferably, the actual identification process includes the following steps:

[0030] S2021: Utilize the detection structure to detect electromagnetic wave signals in the near-Earth region, extract the corresponding feature vector η from it, and input it into the signal feature classifier;

[0031] S2022: Calculate the eigenvector η and the corresponding cluster center c of the eigenvalues ​​containing the detected target using the formula. h Clustering centers c of eigenvalues ​​without detected targets n The distance;

[0032] d1=||η n -c h || 2 ;

[0033] d2=||η n *c n || 2 ;

[0034] S2023: Compare d1 and d2. If d1 > d2, it indicates that there is no target in the near-ground area; otherwise, it indicates that there is a target.

[0035] Preferably, the process of using the signal feature classifier includes the following steps:

[0036] A1: For each eigenvector η n Calculate its relationship with the two cluster centers c. n and c h The Euclidean distance is used to assign the cluster to either cluster D1 or D2, where the expression is:

[0037]

[0038] A2: Based on the new clusters D1 and D2, recalculate the centroids c1 and c2 of the two clusters, where the expression is:

[0039]

[0040] A3: Use the results of the above steps to determine whether J has converged and the cluster center c. n and c h Does it no longer change; if J converges or c n and c h If the value stops changing, stop iterating; otherwise, continue iterating until J converges or the cluster center c is reached. n and c h It will no longer change.

[0041] Preferably, the positioning process of the LLM-driven positioning algorithm includes the following steps:

[0042] S301: After obtaining the data of the detection signal, find the valley point Min of the data obtained in the whole process. In order to ensure that each valley point is obtained accurately, the distance between the valley points is limited according to the sampling frequency, including at least 500 data points.

[0043] S302: Using a sensor probe to scan above the target, based on the scanning time and sampling frequency, with the trough point as the data center, data points within a 0.1s range to the left and right of the trough point are extracted. Calculate the maximum data point within this range, Max.

[0044] S303: Calculate ΔD by subtracting the maximum value Max from the trough value Min, and find the maximum difference ΔD. max Find the trough point of the corresponding point, calculate the distance parameter x corresponding to the trough point, and obtain the actual distance d to determine the position of the target.

[0045] The beneficial effects of this invention are as follows:

[0046] 1. The near-ground underground crop detection system and positioning method based on continuous microwave as described in this invention acquires target information in the near-ground area through scanning detection using a detection structure. By analyzing the characteristics of the target information and combining it with the k-means algorithm, classification and identification are performed, effectively identifying the presence or absence of targets in the near-ground area. The feasibility of the method has been verified through experiments. Simultaneously, by collecting the distance parameter x, its relationship with the actual distance is established, and the output characteristics U indicating the presence or absence of detected targets are analyzed. out They found effective trough points, thus enabling the positioning of near-ground targets.

[0047] 2. The near-ground crop detection system and positioning method based on continuous microwave in this invention utilizes k-means clustering analysis. During detection, the feature values ​​of signals with detected targets are closer to the cluster center located at the center of the image on the coordinate axis, and farther from another cluster center. Therefore, the cluster center at the center of the image corresponds to the cluster with the feature values ​​of the detected target. Conversely, the feature values ​​of signals without detected targets are farther from the cluster center at the center of the image on the coordinate axis, and closer to the cluster center located at the lower right corner of the coordinate axis. Therefore, the cluster center at the lower right corner corresponds to the cluster with the feature values ​​of the undetected target.

[0048] 3. The near-ground underground crop detection system and positioning method based on continuous microwave described in this invention, through the setting of a detection structure, utilizes an antenna structure with excellent electromagnetic performance and high design flexibility. The antenna as a whole uses pure metal material as the antenna arm to ensure its structural stability and high conductivity. The external structure is designed as a four-sided folded metal back cavity. Through the design of the folded metal cavity, not only is higher radiation efficiency provided, but the directivity of the antenna is also effectively enhanced, which is suitable for mid-frequency communication requirements. By setting coaxial feeding, stable signal transmission and good impedance matching are achieved. The coaxial feeding introduces the signal through the central coaxial line and achieves precise connection with the antenna dipole element, while reducing the reflection coefficient to optimize the working performance of the antenna. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a perspective view of the present invention;

[0051] Figure 2 This is a cross-sectional view of the metal shielding cavity in this invention;

[0052] Figure 3 This is a flowchart of the near-ground underground crop detection system based on continuous microwave in this invention;

[0053] Figure 4 This is a flowchart illustrating the composition of the signal synthesis unit in this invention;

[0054] Figure 5 This is a flowchart illustrating the composition of the signal processing unit in this invention;

[0055] Figure 6 This is a flowchart of a positioning method according to an embodiment of the present invention;

[0056] Figure 7 This is a flowchart of signal characteristic identification in this invention;

[0057] Figure 8 This is a classification flowchart of the actual identification process in this invention;

[0058] Figure 9 This is a flowchart of the positioning algorithm driven by LLM in this invention.

[0059] In the diagram: 1. Metal shielding cavity; 2. Coaxial feed; 3. Antenna dipole; 4. Baffle; 5. Support column; 6. Mounting bracket; 7. Power supply; 8. Signal mounting plate. Detailed Implementation

[0060] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0061] like Figures 1 to 2 As shown in the embodiment of the present invention, the near-ground underground crop detection system based on continuous microwave includes an antenna module and a signal system module. The antenna module includes a detection structure and a flow-blocking groove 4. The detection structure includes two antenna structures, one of which is used to transmit signals and the other is used to receive signals. The antenna structure includes a metal shielding cavity 1, a coaxial feed 2, and an antenna dipole 3. The outer wall of the metal shielding cavity 1 is made of a four-sided folded metal plate. The coaxial feed 2 introduces the signal through a central coaxial line, and... The antenna dipole element 3 is connected to the antenna dipole element 3. The antenna dipole element 3 uses a butterfly-shaped dipole as the core radiating element. A support rod structure is provided between the antenna dipole elements 3 and the antenna dipole element 3. The support rod structure includes an outer core and an inner core. The outer core is directly connected to one side of the antenna dipole element 3 through a metal connector. One end of the inner core is provided with a polytetrafluoroethylene connector. The inner core is connected to the other side of the antenna dipole element 3. A short-circuit ring structure is provided in the middle of the antenna dipole element 3. The flow-blocking groove 4 is provided between the outer walls of the two metal shielding cavities 1.

[0062] The antenna structure provided by this invention adopts a novel butterfly antenna with a center frequency of 1.07 GHz. It possesses excellent electromagnetic performance and high design flexibility. The antenna arm is made entirely of pure metal to ensure structural stability and high conductivity. The external structure is designed as a four-sided folded metal cavity. This folded metal cavity design not only provides higher radiation efficiency but also effectively enhances the antenna's directivity, making it suitable for mid-frequency communication needs. The inner layer of the antenna is designed with a 50-ohm coaxial feed 2 to achieve stable signal transmission and good impedance matching. The coaxial feed 2 introduces the signal through a central coaxial line and achieves precise connection with the antenna dipole element 3, while reducing the reflection coefficient to optimize the antenna's operating performance. The antenna employs a butterfly-shaped dipole as its core radiating element, constructed from a 1.5mm thick aluminum plate. This material choice not only makes the structure lightweight but also provides excellent electrical conductivity and durability, meeting the high conductivity requirements of mid-frequency antennas. To achieve precise connection between the feed structure and the antenna dipole 3, a support rod structure is used. The outer core of the support rod is directly connected to one side of the butterfly antenna dipole via a metal connector, while the inner core is isolated by a PTFE plastic connector, ensuring electrical insulation. Through the PTFE connector, the inner core is precisely connected to the other side of the dipole. The metal connector at this point acts as a signal conductor, transmitting the signal from the inner core to the antenna dipole 3, forming a complete feed path. To further optimize antenna performance and adjust its standing wave ratio (VSWR), a short-circuit ring structure is added to the middle of the antenna dipole 3. The addition of the short-circuit ring plays a crucial tuning role; by adjusting its size and position, the VSWR of the antenna can be effectively improved, thereby enhancing its overall radiation efficiency and signal transmission quality. Improve the accuracy of underground crop detection in near-ground areas through antenna structure design.

[0063] Linear Localization Method (LLM) is a class of algorithmic strategies used to determine the location of a target. For example... Figures 3 to 5 As shown, the signal system module includes a signal synthesis unit and a signal processing unit. The signal synthesis unit includes a frequency control circuit, a frequency synthesis circuit, and a power amplification circuit. The signal processing unit includes a power amplification circuit, a radio frequency detection circuit, and a display output circuit. A support column 5 is fixedly connected to the top of the flow choke 4. A mounting bracket 6 is fixedly connected to the support column 5, and a power supply 7 is installed on the mounting bracket 6 by means of a snap fastener. A signal mounting plate 8 is also fixedly connected to the support column 5, and the signal system module is mounted on the signal mounting plate 8.

[0064] The signal system module provided by this invention comprises a signal generation unit consisting of a frequency control circuit, a frequency synthesis circuit, and a power amplifier circuit. The frequency control circuit interacts with an STM32 microcontroller via a touchscreen display and uses an ADF4350 chip to generate output signals ranging from 137.5 to 4400 MHz. The power amplifier circuit, based on the SBB5089 chip, supports signal amplification from 50 to 6000 MHz, with a maximum power gain of 20 dB, and features wide bandwidth, stable gain, low noise, and a large dynamic range. The signal processing unit consists of a power amplifier circuit, an RF detection circuit, and a display output circuit. The power amplifier circuit, also based on the SBB5089 chip, amplifies the original signal from the receiving antenna. The RF detection circuit, based on the AD8361 chip, performs high-precision power detection on microwave signals and converts the signal power into a linearly corresponding DC voltage output, meeting the needs of signal characteristic research at different frequencies.

[0065] Changes in voltage can directly reflect changes in the amplitude of the echo signal, thereby indirectly determining the presence or absence of the target.

[0066] like Figure 6 As shown, a method for detecting and locating underground crops in near-ground areas based on continuous microwaves includes the following steps:

[0067] S1: Bury the target in the soil near the ground, fix the detection structure on the crossbeam of the movable slide rail for horizontal scanning detection, collect the complete electromagnetic wave signal of the target, and perform signal preprocessing.

[0068] S2: Through k-means clustering analysis, determine the signal category of whether there is a detection target in the near-ground area, and establish a signal feature classifier to classify the different extracted feature vectors, thereby obtaining a classification model; where k-means clustering includes the training process and the actual identification process;

[0069] S3: The location of the target is determined using an LLM-driven localization algorithm.

[0070] The positioning method provided by this invention, when in use, measures the output characteristics U of the detection structure in the near-ground region using a flat scan method. out The system records the distance the detection structure moves, denoted by parameter x, and uses pattern recognition to determine whether a detection target exists in the near-ground region. Simultaneously, it achieves target localization by acquiring the target's movement parameter x and analyzing its relationship with the actual distance, ultimately yielding the detection and localization results.

[0071] like Figure 5 As shown, the signal preprocessing method includes the following steps:

[0072] S101: The detection and scanning are performed by moving the detection structure on the crossbeam of the slide rail, while simultaneously acquiring the detection output characteristic U. out The parameter x is related to distance;

[0073] S102: Process the collected data and convert it from the time domain to the spatial domain representation;

[0074] S103: The moving average smoothing method is used to reduce signal noise, which effectively reduces signal noise while preserving target features, providing a data basis for target detection.

[0075] S104: The relationship between the actual distance d and the distance parameter x can be derived from the actual measurement.

[0076] The signal preprocessing method provided by this invention involves first burying the target in the soil in the near-field region, then fixing the detection structure on the crossbeam of a movable slide rail. The actual length of the slide rail crossbeam is 170cm, and the experimental sliding length is 160cm. A horizontal scan is performed by moving the crossbeam on the slide rail, while simultaneously acquiring the detection output characteristics U. out The experiment recorded the distance-related parameter x. The dataset was acquired using a flat-scan method with a sampling frequency of 500Hz, obtaining a large number of amplitude data points. To more intuitively display the spatial characteristics of the data, the acquired data needed to be processed, converting it from the time domain to a spatial domain representation, thereby revealing the characteristic patterns of the data's spatial distribution. The flat-scan experiment collected data from the target area by moving the probe structure along a fixed slide rail. The raw signals collected were often affected by environmental noise and equipment errors, resulting in a large number of random fluctuations in the signals, making it difficult to directly reveal the characteristics of the target. Based on the flat-scan experiment data, the signal was denoised using a moving average smoothing method, effectively reducing signal noise while preserving target characteristics, providing a reliable data foundation for target detection. Finally, based on actual measurements, the relationship between the actual distance d and the distance parameter x can be derived.

[0077] like Figure 7 As shown, the training process includes the following steps:

[0078] S2011: Extract features from the collected electromagnetic wave signals to obtain feature vectors that reflect the characteristics of the detected target;

[0079] S2012: When a target is present in the near-ground region, the signal is significantly weaker near the target than when there is no target, and is represented by the signal mean μ and the signal standard deviation σ.

[0080] The signal mean μ is used to reflect the overall signal strength, and its expression is:

[0081]

[0082] Where s(m) is the obtained electromagnetic wave signal, and M is the number of signal sampling points used for feature extraction, M = 3000;

[0083] The standard deviation σ of the signal is used to reflect information about the amplitude of signal fluctuations, and its expression is:

[0084]

[0085] In the training process provided by this invention, the detection structure is mainly used to acquire the signal strength of electromagnetic waves and represent the energy strength in the form of output voltage values. Extraction of feature signals is crucial for establishing a feature signal classifier and is also the core step in determining the presence or absence of a target, directly affecting the effectiveness of the determination. When a target is present in the near-ground region, the detection signal of the detection structure scanning above the target is significantly weaker near the target than when there is no target. Based on this conclusion, this paper introduces two time-domain statistical features: the signal mean μ and the signal standard deviation σ. Cluster analysis is performed using the mean and standard deviation as features, resulting in two clusters. One cluster is relatively concentrated, while the other is relatively dispersed. Although neither cluster has a central region, both clusters have relatively obvious cluster centers, and the distance between these two cluster centers is relatively large, with very clear boundaries. Therefore, clustering methods can be used to determine whether a target is present in the near-ground region.

[0086] like Figure 8 As shown, the actual identification process includes the following steps:

[0087] S2021: Utilize the detection structure to detect electromagnetic wave signals in the near-Earth region, extract the corresponding feature vector η from it, and input it into the signal feature classifier;

[0088] S2022: Calculate the eigenvector η and the corresponding cluster center c of the eigenvalues ​​containing the detected target using the formula. h Clustering centers c of eigenvalues ​​without detected targets n The distance;

[0089] d1=||η n -c h || 2 ;

[0090] d2=||η n -c n || 2 ;

[0091] S2023: Compare d1 and d2. If d1 > d2, it indicates that there is no target in the near-ground area; otherwise, it indicates that there is a target.

[0092] In the actual identification process provided by this invention, to verify the effectiveness of the clustering method and determine whether cluster D1 or D2 corresponds to the case with or without a detected target, during detection, the feature values ​​of the signal with a detected target are closer to the cluster center located at the center of the image on the coordinate axis, and farther away from the other cluster center. Therefore, the cluster center at the center of the image corresponds to the feature value cluster with a detected target. Conversely, the feature values ​​of the signal without a detected target are farther away from the cluster center at the center of the image on the coordinate axis, and closer to the cluster center located at the lower right corner of the coordinate axis. Therefore, the cluster center at the lower right corner corresponds to the feature value cluster without a detected target.

[0093] like Figure 8 As shown, the process of using the signal feature classifier includes the following steps:

[0094] A1: For each eigenvector η n Calculate its relationship with the two cluster centers c. n and c h The Euclidean distance is used to assign the cluster to either cluster D1 or D2, where the expression is:

[0095]

[0096] A2: Based on the new clusters D1 and D2, recalculate the centroids c1 and c2 of the two clusters, where the expression is:

[0097]

[0098] A3: Use the results of the above steps to determine whether J has converged and the cluster center c. n and c h Does it no longer change; if J converges or c n and c h If the value stops changing, stop iterating; otherwise, continue iterating until J converges or the cluster center c is reached. n and c h It will no longer change.

[0099] The signal feature classifier provided by this invention is crucial to the identification process. Its core principle is to categorize and group data through cluster training, facilitating subsequent data identification. This is a commonly used unsupervised machine learning algorithm, primarily used for cluster analysis. The algorithm divides the dataset into K categories, increasing the similarity between samples within the same cluster and maximizing the differences between samples from different clusters, thus enabling the identification of the presence or absence of targets in near-ground regions.

[0100] like Figure 9 As shown, the localization process of the LLM-driven localization algorithm includes the following steps:

[0101] S301: After obtaining the data of the detection signal, find the valley point Min of the data obtained in the whole process. In order to ensure that each valley point is obtained accurately, the distance between the valley points is limited according to the sampling frequency, including at least 500 data points.

[0102] S302: Using a sensor probe to scan above the target, based on the scanning time and sampling frequency, with the trough point as the data center, extract data points within a 0.1s range to the left and right of the trough point, and calculate the maximum data point Max within this range;

[0103] S303: Calculate ΔD by subtracting the maximum value Max from the trough value Min, and find the maximum difference ΔD. max Find the trough point of the corresponding point, calculate the distance parameter x corresponding to the trough point, and obtain the actual distance d to determine the position of the target.

[0104] The LLM-driven localization algorithm provided in this invention, in order to facilitate the acquisition of the position of targets in the near-ground area, uses sensors to determine the position of the detection structure and records the sliding distance. The detection structure performs a linear scan from the starting point, and simultaneously measures and records the output characteristics U of the received target signal. out The distance-related parameter x is recorded. When a target appears in the near-ground area and a clear trough appears, the signal strength rises and falls significantly. By finding the distance parameter x corresponding to this trough, the location of the target can be determined.

[0105] Working Principle: By designing the antenna structure, it possesses excellent electromagnetic performance and high design flexibility. The antenna arm is made of pure metal to ensure structural stability and high conductivity. The external structure is designed as a four-sided folded metal cavity. This folded metal cavity design not only provides higher radiation efficiency but also effectively enhances the antenna's directivity, making it suitable for mid-frequency communication needs. Coaxial feeding is used to achieve stable signal transmission and good impedance matching. The coaxial feed introduces the signal through a central coaxial line and achieves precise connection with the antenna dipole element, while reducing the reflection coefficient to optimize antenna performance. The positioning method provided by this invention, when in use, measures the output characteristics U of the detection structure in the near-ground region using a flat-scan method. out The system records the distance the detection structure moves, denoted by parameter x, and uses pattern recognition to determine whether a detection target exists in the near-ground region. Simultaneously, it achieves target localization by acquiring the target's movement parameter x and analyzing its relationship with the actual distance, ultimately yielding the detection and localization results.

[0106] During training, the detection structure is primarily used to acquire the signal strength of electromagnetic waves, representing the energy strength as an output voltage value. Feature signal extraction is crucial for establishing a feature signal classifier and is also the core step in determining the presence or absence of a target, directly impacting the effectiveness of the determination. When a target is present in the near-ground region, the detection signal of the detection structure scanning above the target is significantly weaker near the target than when no target is present. Based on this conclusion, this paper introduces two time-domain statistical features: the signal mean μ and the signal standard deviation σ. Cluster analysis is performed using the mean and standard deviation as features, resulting in two clusters: one relatively concentrated and the other relatively dispersed. Although neither cluster exhibits a region centered on a circle, both clusters have relatively clear cluster centers, and the distance between these centers is relatively large, with very distinct boundaries. Therefore, clustering can be used to determine the presence of a target in the near-ground region.

[0107] In the actual identification process, to verify the effectiveness of the clustering method and determine whether cluster D1 or D2 corresponds to the case with or without a detected target, during detection, the feature values ​​of the signal with a detected target are closer to the cluster center located at the center of the image on the coordinate axis, and farther away from the other cluster center. Therefore, the cluster center at the center of the image corresponds to the feature value cluster with a detected target. Conversely, the feature values ​​of the signal without a detected target are farther away from the cluster center at the center of the image on the coordinate axis, and closer to the cluster center located at the lower right corner of the coordinate axis. Therefore, the cluster center at the lower right corner corresponds to the feature value cluster without a detected target.

[0108] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A near-ground underground crop detection system based on continuous microwave, comprising an antenna module and a signal system module; characterized in that: The antenna module includes a detection structure and a choke slot (4). The detection structure includes two antenna structures, one of which is used to transmit signals and the other is used to receive signals. The antenna structure includes a metal shielding cavity (1), a coaxial feed (2), and an antenna dipole (3). The outer wall of the metal shielding cavity (1) is made of a four-sided folded metal plate. The coaxial feed (2) introduces the signal through a central coaxial line and is connected to the antenna dipole (3). A butterfly-shaped radiating element is used as the core radiating element; a support rod structure is provided between the antenna dipole (3) and the antenna dipole (3), the support rod structure includes an outer core and an inner core, the outer core is directly connected to one side of the antenna dipole (3) through a metal connector, one end of the inner core is provided with a polytetrafluoroethylene connector, and the inner core is connected to the other side of the antenna dipole (3); a short-circuit ring structure is provided in the middle of the antenna dipole (3); the flow-blocking groove (4) is provided between the outer walls of the two metal shielding cavities (1).

2. The near-ground underground crop detection system based on continuous microwave as described in claim 1, characterized in that: The signal system module includes a signal synthesis unit and a signal processing unit. The signal synthesis unit includes a frequency control circuit, a frequency synthesis circuit, and a power amplification circuit. The signal processing unit includes a power amplification circuit, a radio frequency detection circuit, and a display output circuit. A support column (5) is fixedly connected to the top of the choke groove (4). A mounting bracket (6) is fixedly connected to the support column (5). A power supply (7) is installed on the mounting bracket (6) by a snap fastener. A signal mounting plate (8) is also fixedly connected to the support column (5). The signal system module is mounted on the signal mounting plate (8).

3. A method for detecting and locating underground crops in near-ground areas based on continuous microwaves, applicable to the near-ground underground crop detection system based on continuous microwaves as described in any one of claims 1 or 2, characterized in that: The positioning method includes the following steps: S1: The target is buried in the soil near the ground, the detection structure is fixed on the crossbeam of the movable slide rail for horizontal scanning detection, the complete electromagnetic wave signal of the target is collected, and the signal is preprocessed. S2: Through k-means clustering analysis, determine the signal category of whether there is a detection target in the near-ground area, and establish a signal feature classifier to classify the different extracted feature vectors, thereby obtaining a classification model; where k-means clustering includes the training process and the actual identification process; S3: The location of the target is determined using an LLM-driven localization algorithm.

4. The method for detecting and locating underground crops in near-ground areas based on continuous microwaves according to claim 3, characterized in that: The signal preprocessing method includes the following steps: S101: The detection and scanning are performed by moving the detection structure on the crossbeam of the slide rail, while simultaneously acquiring the detection output characteristic U. out The parameter x is related to distance; S102: Process the collected data and convert it from the time domain to the spatial domain representation; S103: The moving average smoothing method is used to reduce signal noise, which effectively reduces signal noise while preserving target features, providing a data basis for target detection. S104: The relationship between the actual distance d and the distance parameter x can be derived from the actual measurement.

5. The method for detecting and locating underground crops in near-ground areas based on continuous microwaves according to claim 4, characterized in that: The training process includes the following steps: S2011: Extract features from the collected electromagnetic wave signals to obtain feature vectors that reflect the characteristics of the detected target; S2012: When a target is present in the near-ground region, the signal is significantly weaker near the target than when there is no target, and is represented by the signal mean μ and the signal standard deviation σ. The signal mean μ is used to reflect the overall signal strength, and its expression is: Where s(m) is the obtained electromagnetic wave signal, and M is the number of signal sampling points used for feature extraction, M = 3000; The standard deviation σ of the signal is used to reflect information about the amplitude of signal fluctuations, and its expression is:

6. The method for detecting and locating underground crops in near-ground areas based on continuous microwaves according to claim 5, characterized in that: The actual identification process includes the following steps: S2021: Utilize the detection structure to detect electromagnetic wave signals in the near-Earth region, extract the corresponding feature vector η from it, and input it into the signal feature classifier; S2022: Calculate the eigenvector η and the corresponding cluster center c of the eigenvalues ​​containing the detected target using the formula. h Clustering centers c of eigenvalues ​​without detected targets n The distance; d1=||η n -c h || 2 ; d2=||η n -c n || 2 ; S2023: Compare d1 and d2. If d1 > d2, it indicates that there is no target in the near-ground area; otherwise, it indicates that there is a target.

7. The method for detecting and locating underground crops in near-ground areas based on continuous microwaves according to claim 6, characterized in that: The process of using the signal feature classifier includes the following steps: A1: For each eigenvector η n Calculate its relationship with the two cluster centers c. n and c h The Euclidean distance is used to assign the cluster to either cluster D1 or D2, where the expression is: A2: Based on the new clusters D1 and D2, recalculate the centroids c1 and c2 of the two clusters, where the expression is: A3: Use the results of the above steps to determine whether J has converged and the cluster center c. n and c h Does it no longer change; if J converges or c n and c h If the value stops changing, stop iterating; otherwise, continue iterating until J converges or the cluster center c is reached. n and c h It will no longer change.

8. The method for detecting and locating underground crops in near-ground areas based on continuous microwaves according to claim 7, characterized in that: The localization process of the LLM-driven localization algorithm includes the following steps: S301: After obtaining the data of the detection signal, find the valley point Min of the data obtained in the whole process. In order to ensure that each valley point is accurately obtained, the distance between the valley points is limited according to the sampling frequency, including at least 500 data points. S302: Using a sensor probe to scan above the target, based on the scanning time and sampling frequency, with the trough point as the data center, data points within a 0.1s range to the left and right of the trough point are extracted. Calculate the maximum data point within this range, Max. S303: Calculate ΔD by subtracting the maximum value Max from the trough value Min, and find the maximum difference ΔD. max Find the trough point of the corresponding point, calculate the distance parameter x corresponding to the trough point, and obtain the actual distance d to determine the position of the target.

Images