A wheat plant height inversion method based on GNSS reflected signals
By employing a dual-channel parallel inversion architecture and multi-satellite data fusion, the problem of insufficient reliability of GNSS reflection signals in dynamic environments was solved, enabling robust and accurate monitoring of wheat plant height and adapting to different growth stages and environmental changes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING FORESTRY UNIV
- Filing Date
- 2025-09-29
- Publication Date
- 2026-06-23
AI Technical Summary
Existing GNSS reflection signal ranging methods lack an effective mechanism to integrate and judge the relative reliability of ranging results in dynamic observation environments, resulting in insufficient stability and accuracy of inversion results, especially when the carrier phase signal is blocked or lost lock, performance drops sharply.
A dual-channel parallel inversion architecture is adopted, combining a code-phase combination-based ranging method and a signal strength-based ranging method. Multipath reflection components are extracted through spectrum analysis and detrending processing. The initial reflection distance is fused using time-varying reliability weights, and refined correction is performed by combining phenological information and multi-satellite data. Finally, the wheat plant height is calculated.
It achieves robustness and accuracy of inversion results in dynamic environments, improves the continuity and precision of wheat plant height monitoring, adapts to the monitoring needs of the entire wheat growth cycle, and reduces noise and outlier interference.
Smart Images
Figure CN121276565B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of ground distance measurement using radio navigation signals, specifically a method for inverting wheat plant height based on GNSS reflection signals. Background Technology
[0002] In the field of precision agriculture, continuous measurement of the height of the wheat canopy top is a key ranging task for dividing growth stages, estimating biomass, and providing early warning of lodging disasters.
[0003] In Earth observation missions, accurately determining the vertical distance of a specific surface target, such as the top surface of a vegetation canopy, relative to a reference point is a fundamental geometric measurement task. Traditional surveying methods have inherent limitations in terms of efficiency and cost-effectiveness when dealing with large-scale and dynamically changing monitoring targets. While active ranging systems using dedicated emission sources, such as laser ranging systems, can achieve high-precision distance measurements, their deployment costs are high, and the propagation of their electromagnetic waves in the atmosphere is easily affected by weather conditions.
[0004] Surface reflection measurement using GNSS chance signals is a passive ranging technique. This technique utilizes the direct navigation signal simultaneously received by a standard antenna and the multipath signal reflected from the ground surface. These two signals interfere at the antenna, forming a measurable composite signal field. By analyzing specific parameters of this composite signal field, the technique deduces the additional propagation path length of the reflected signal compared to the direct signal, thereby determining the effective distance from the reflecting surface to the antenna.
[0005] In practical implementation, existing technologies mainly analyze the interferometric field by examining two types of observations derived from raw GNSS data: one is signal strength, i.e., the oscillation characteristics of the signal-to-noise ratio; the other is a combined sequence constructed using code pseudorange observation data and carrier phase observation data. Ranging methods based on signal strength are susceptible to interference from changes in the electromagnetic characteristics of the reflecting surface, while the reliability of ranging methods based on the combination of code pseudorange observation data and carrier phase observation data directly depends on the continuity of the carrier phase observation data, making them extremely sensitive to loss of lock caused by signal obstruction or fading.
[0006] However, existing technologies typically treat these two ranging methods as independent approaches. In dynamic observation environments, when the quality of one type of observation deteriorates due to specific problems in signal propagation or reception, the system lacks an effective mechanism to integrate and determine the relative reliability of the ranging results from both methods, thus limiting the stability and accuracy of the final distance measurement.
[0007] To address this, a method for inverting wheat plant height based on GNSS reflection signals is proposed. Summary of the Invention
[0008] The purpose of this invention is to provide a method for wheat plant height inversion based on GNSS reflection signals, thereby solving the problems mentioned in the background art.
[0009] To achieve the above objectives, the present invention provides the following technical solution:
[0010] A method for inverting wheat plant height based on GNSS reflection signals, comprising:
[0011] S1: Acquire GNSS pseudorange observation data, carrier phase observation data, and signal strength observation data; extract multipath reflection components based on the pseudorange observation data and the carrier phase observation data, and obtain the first reflection distance through spectrum analysis; perform detrending processing on the signal strength observation data, extract the reflection interference signal, and obtain the second reflection distance through spectrum analysis;
[0012] S2: Perform cycle slip detection on the carrier phase observation data to generate phase anomaly markers;
[0013] S3: Determine the time-varying reliability weight based on the phase anomaly marker; use the time-varying reliability weight to perform weighted fusion of the first reflection distance and the second reflection distance to obtain the preliminary reflection distance;
[0014] S4: Based on the normalized amplitude of the reflected interference signal and the preset wheat phenological information, the initial reflection distance is corrected according to the growth stage to obtain the corrected reflection distance;
[0015] S5: Perform multi-satellite data fusion on the corrected reflection distance to obtain the final effective reflection height, and calculate the wheat plant height based on the final effective reflection height and the initial antenna height.
[0016] Preferably, the process of obtaining the first reflection distance specifically includes:
[0017] The pseudorange-carrier phase differential processing is employed to multiply the carrier phase observation data by the signal wavelength at the corresponding frequency point to obtain the equivalent distance value. The equivalent distance value at the same frequency is then subtracted from the pseudorange observation data at the same frequency to extract the multipath reflection component. Power spectral density analysis is applied to the multipath reflection component to generate a spectral distribution containing the frequency and corresponding amplitude. The frequency with the highest amplitude is extracted from the spectral distribution as the dominant frequency of the low-frequency oscillation. Based on the correspondence between the dominant frequency and the multipath delay, the first reflection distance is obtained through numerical conversion.
[0018] Preferably, the process of obtaining the second reflection distance specifically includes:
[0019] The observed signal intensity is converted into a linear-scale signal amplitude ratio. A polynomial function is used to fit the signal amplitude ratio to generate a fitted curve. The difference between the signal amplitude ratio and the fitted curve is used to separate the reflected interference signal. Power spectral density analysis is applied to the reflected interference signal to generate a spectral distribution. The frequency point with the highest amplitude is selected from the spectral distribution and determined as the dominant oscillation frequency of the reflected interference signal. Combining the geometric relationship between the dominant oscillation frequency and the reflection path difference, the second reflection distance is calculated through numerical conversion.
[0020] Preferably, the cycle slip detection process specifically includes:
[0021] The cycle slip detection employs an inter-epoch difference method. It selects continuous epoch observations of the carrier phase observation data, performs difference operations on the observation data of adjacent epochs within the continuous epoch observations, and obtains a first-order difference value. Based on the first-order difference value, multiple rounds of adjacent difference operations are performed to generate higher-order difference values. These higher-order difference values are compared one by one with a preset variable difference threshold. For epoch positions where the higher-order difference value exceeds the difference threshold, a phase anomaly marker containing the epoch number is generated.
[0022] Preferably, the process of obtaining the initial reflection distance specifically includes:
[0023] The cycle slip density in the phase anomaly marker is divided into intervals to generate discretized cycle slip density levels; based on preset physical association rules, corresponding initial reliability weights are assigned to the cycle slip density levels, with higher weights assigned to lower cycle slip densities; time smoothing is performed on the initial reliability weights to generate time-varying reliability weights; using the time-varying reliability weights, a weighted combination calculation is performed on the first reflection distance and the second reflection distance to generate the preliminary reflection distance;
[0024] Preferably, the process of obtaining the corrected reflection distance includes:
[0025] The entire growth period of wheat is divided into three stages: the seedling stage, the jointing and heading rapid growth stage, and the maturity and harvest stage. By monitoring whether the normalized amplitude of the reflected interference signal is lower than a preset threshold and combining it with the annual accumulated day information, the growth stage to which the current observation time point belongs is automatically determined.
[0026] Different correction models are applied for different growth stages. In the seedling stage, the initial reflection distance is the corrected reflection distance.
[0027] During the rapid growth stage of jointing and heading, the corrected reflection distance is obtained by subtracting the systematic deviation correction amount from the initial reflection distance; the systematic deviation correction amount is calibrated in advance through experiments.
[0028] During the mature harvest stage, the initial reflection distance is corrected using a preset mapping model to obtain the corrected reflection distance. The mapping model is established by regression analysis on a synchronous observation dataset containing an attenuation index and a corresponding penetration deviation. The attenuation index is obtained by normalizing the peak amplitude of the reflected interference signal. The penetration deviation is obtained by adding the initial reflection distance to the synchronously measured actual plant height on the ground and subtracting the initial antenna height.
[0029] Preferably, the process of obtaining the wheat plant height specifically includes:
[0030] Satellite identifiers are parsed on the corrected reflection distance to separate reflection distance subsequences corresponding to different satellites; time synchronization processing is performed on the reflection distance subsequences of each satellite to generate a time-aligned multi-satellite reflection distance dataset; quality assessment is performed on the time-aligned multi-satellite reflection distance dataset to calculate the continuity score and outlier ratio of each satellite's data; based on the continuity score and outlier ratio, a corresponding fusion weight is assigned to each satellite to generate a satellite-weight mapping table; using the satellite-weight mapping table, a weighted aggregation operation is performed on the time-aligned multi-satellite reflection distance dataset to generate the final effective reflection height; coordinate systematization processing is performed on the initial antenna height to generate antenna height data in a standard coordinate system; the difference between the final effective reflection height and the antenna height data is calculated to generate the wheat plant height.
[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0032] 1. This invention employs a dual-channel parallel inversion architecture, utilizing a ranging method based on code phase combination and a ranging method based on signal strength to establish an adaptive fusion mechanism based on real-time carrier phase quality. When the carrier phase signal quality degrades due to obstruction or loss of lock, the system can automatically reduce its dependence on the first reflection distance and instead rely more on the second reflection distance inverted from signal strength; conversely, the same applies. This "optimal selection" dynamic weighted fusion strategy effectively avoids the risk of a sudden performance drop under specific conditions for a single technical approach, achieving complementary technical advantages. This ensures the continuity and robustness of the initial reflection distance estimate, solves the technical problem of insufficient reliability of inversion results due to reliance on a single signal source in existing technologies, and improves the stability and reliability of the inversion results.
[0033] 2. This invention introduces a refined segmented correction model based on phenological information. This model can automatically identify the growth stage of wheat based on physical characteristics such as the normalized amplitude of the reflected interference signal, and apply targeted correction rules with clear physical meaning. This segmented, nonlinear correction method simulates the interaction mechanism between GNSS signals and the dynamically changing vegetation canopy, corrects systematic biases caused by model mismatch at key growth nodes, improves the accuracy of plant height inversion results throughout the entire wheat growth cycle, and enhances the monitoring accuracy and adaptability throughout the entire growth cycle.
[0034] 3. This invention fully utilizes the redundant information of the GNSS multi-satellite system by performing quality assessment and weighted fusion of observation data from multiple satellites. By evaluating the continuity and outlier ratio of each satellite's data to determine its fusion weight, the interference of poor-quality satellite data on the final result can be effectively suppressed, while smoothing random errors. This multi-satellite fusion mechanism is equivalent to conducting collaborative observation and verification of the wheat canopy from multiple different angles, improving the statistical reliability and measurement accuracy of the final effective reflectance height, making the final calculated wheat plant height closer to the true value. Attached Figure Description
[0035] Figure 1 A schematic diagram of a wheat plant height inversion method based on GNSS reflection signals provided in an embodiment of the present invention;
[0036] Figure 2 This is a flowchart illustrating the weighted fusion process for obtaining the initial reflection distance according to an embodiment of the present invention.
[0037] Figure 3 This is a flowchart illustrating the wheat plant height acquisition process based on growth stage correction and multi-satellite fusion, as described in an embodiment of the present invention. Detailed Implementation
[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0039] Reference Figure 1 The specific implementation steps of the method proposed in this invention include:
[0040] S1: Acquire GNSS pseudorange observation data, carrier phase observation data, and signal strength observation data; extract multipath reflection components based on the pseudorange observation data and the carrier phase observation data, and obtain the first reflection distance through spectrum analysis; perform detrending processing on the signal strength observation data, extract the reflection interference signal, and obtain the second reflection distance through spectrum analysis;
[0041] S2: Perform cycle slip detection on the carrier phase observation data to generate phase anomaly markers;
[0042] S3: Determine the time-varying reliability weight based on the phase anomaly marker; use the time-varying reliability weight to perform weighted fusion of the first reflection distance and the second reflection distance to obtain the preliminary reflection distance;
[0043] S4: Based on the normalized amplitude of the reflected interference signal and the preset wheat phenological information, the initial reflection distance is corrected according to the growth stage to obtain the corrected reflection distance;
[0044] S5: Perform multi-satellite data fusion on the corrected reflection distance to obtain the final effective reflection height, and calculate the wheat plant height based on the final effective reflection height and the initial antenna height.
[0045] Example 1
[0046] This embodiment provides a specific application of a wheat plant height inversion method based on GNSS reflection signals. A typical application scenario is a wheat field (A) to be measured, where a wheat plant height inversion method based on GNSS reflection signals is introduced for wheat plant height detection. This method is executed using a GNSS receiver at a fixed observation station deployed near the wheat field (A).
[0047] Reference Figure 1 The specific implementation process of the method proposed in this invention is as follows:
[0048] Further, GNSS pseudorange observation data, carrier phase observation data, and signal strength observation data are acquired; based on the pseudorange observation data and the carrier phase observation data, multipath reflection components are extracted, and the first reflection distance is obtained through spectrum analysis; the signal strength observation data is detrended, the reflection interference signal is extracted, and the second reflection distance is obtained through spectrum analysis. Corresponding to step S1 above, the specific process is as follows:
[0049] On a selected observation day, such as the year DOY124, a GNSS receiver is started and continuously acquires raw observation data from multiple global navigation satellite systems, including GPS, GLONASS, Galileo, and BDS, through its connected antenna. This includes pseudorange observation data, carrier phase observation data, and signal strength observations. This example uses the processing of the E1 frequency signal of a Galileo E01 satellite flying at a high elevation angle as an example.
[0050] This invention separates satellite pseudorange and carrier phase observation data at the same frequency from the data stream. Pseudorange has lower accuracy but no integer ambiguity, while carrier phase has extremely high accuracy but suffers from ambiguity and is sensitive to signal interruptions. This invention employs pseudorange-carrier phase differential processing to differentially combine the two data points, constructing a new combined observation measurement to obtain multipath reflection components.
[0051] The multipath reflection components were then fed into the Lomb-Scargle periodogram algorithm for power spectral density analysis, generating a spectral distribution containing frequencies and corresponding amplitudes. In the generated spectral distribution, a highly concentrated energy peak appeared at 0.162 Hz, which is the dominant frequency of the low-frequency oscillation. Based on the physical relationship between this dominant frequency and the multipath delay, the first reflection distance was calculated to be 1.830 meters through numerical conversion.
[0052] Pseudorange-carrier phase differential processing effectively reduces common-mode errors such as ionospheric delay, tropospheric delay, satellite clock bias, and receiver clock bias. This allows the differential sequence to retain primarily multipath effects, carrier phase ambiguity, and noise, thus successfully separating and highlighting weak multipath reflection components from the complex raw observation data. This provides a clean, higher signal-to-noise ratio input signal for subsequent high-precision spectrum analysis.
[0053] Simultaneously, the acquired signal strength observations undergo detrending processing. The original signal strength observations exhibit a smooth macroscopic trend with changes in satellite elevation angle, which masks the subtle fluctuations caused by reflection interferometry. To separate these fluctuations, a low-order polynomial function is used to fit the signal strength sequence, accurately modeling and removing the aforementioned macroscopic trend.
[0054] For all linearized signal intensity observations within a single satellite observation arc, this method uses multiple polynomial functions of different orders to fit the data, generating a set of candidate fitting curves. Subsequently, the Bayesian information criterion is used to quantitatively score the performance of each candidate fitting curve. This criterion balances the goodness of fit and complexity of the model, automatically selecting the order with the optimal score by comparing the scores of all candidate orders. Finally, the selected optimal order polynomial function is used to generate the final fitting curve for detrending processing.
[0055] This processing selects the most suitable detrending model for each satellite observation arc segment, ensuring that the reflected interference signal can be separated most accurately from the raw signal intensity data of various forms. It avoids overfitting or underfitting problems that may be caused by the fixed model order, making the extracted reflected interference signal purer and directly improving the accuracy of the second reflection distance calculation.
[0056] Specifically, the signal strength observations, expressed in decibels (dB / Hertz), are converted to linear units using an antilogarithmic operation. Then, for all linearized signal strength observations within a single satellite observation arc, a second-order polynomial function is used for least-squares fitting to characterize and remove the macroscopic trend caused by changes in satellite elevation angle. Subtracting the corresponding fitted trend curve from the original signal strength observations yields the residual sequence, which is the reflected interferometric signal. Next, spectral analysis is performed; in this embodiment, Lomb-Scargle periodogram analysis is also executed on the reflected interferometric signal to extract its dominant oscillation frequency caused by multipath interference, which is 0.160 Hz. Finally, based on the interferometric geometric relationship between this dominant oscillation frequency and the reflection path difference, the second reflection distance is calculated to be 1.850 meters.
[0057] GNSS signal strength observations exhibit a significant trend term with changes in satellite elevation angle, which can mask interference oscillations caused by reflection. This method employs a polynomial function to fit and eliminate this trend, providing an efficient and accurate detrending approach. It separates the macroscopic variation trend of the direct signal from the microscopic interference oscillations of the reflected signal, ensuring that subsequent spectrum analysis deals with a pure reflected interference signal. By separating the pure reflected interference signal, subsequent spectrum analysis focuses only on periodic oscillations caused by multipath interference, avoiding interference from signal trends in the extraction of the dominant frequency and guaranteeing the accuracy of the second reflection distance calculation.
[0058] Further, cycle slip detection is performed on the carrier phase observation data to generate phase anomaly markers. Corresponding to step S2 above, the specific process is as follows:
[0059] Cycle slip detection is performed on the carrier phase observation data using an epoch-based differential method. This method calculates the higher-order differential values of the carrier phase observation data between consecutive epochs and compares these values with a pre-set differential threshold. If the differential value exceeds the threshold, a cycle slip event is determined to have occurred at that epoch, and a phase anomaly marker containing the satellite identifier and the epoch time is generated.
[0060] Cycle slip detection was performed on the carrier phase observation data at frequency E1 of the E01 satellite. During the observation period, no higher-order differential values were found to exceed the preset variable differential threshold. Therefore, no phase anomaly markers were generated during this period, indicating that the carrier phase data quality is good and its continuity is high.
[0061] The epoch difference method is a classic cycle slip detection method that is computationally simple, highly real-time, and easy to implement in engineering. It can sensitively capture minute discontinuities in carrier phase observation data. This simple and effective approach provides a timely and reliable input signal for determining reliability weights, ensuring the reliability of the final result.
[0062] Further, based on the phase anomaly marker, a time-varying reliability weight is determined; the first reflection distance and the second reflection distance are weighted and fused using the time-varying reliability weight to obtain a preliminary reflection distance. Corresponding to step S3 above, the specific process is as follows:
[0063] Subsequently, based on the generated phase anomaly markers, the relative reliability of the first and second reflection distances is quantitatively evaluated. Within a sliding time window, the frequency of occurrence of phase anomaly markers, i.e., cycle slip density, is statistically analyzed and divided into several levels. According to preset physical rules, corresponding initial reliability weights are assigned to different cycle slip density levels: during clean periods without cycle slips, the first reflection distance is given a higher weight; while during periods with frequent cycle slips, the weight of the second reflection distance is conversely increased. To ensure a smooth weight transition and avoid abrupt jumps in the fusion result, a time-smoothing filter is applied to the initial weight sequence to generate a set of time-varying reliability weights that change continuously over time.
[0064] Since the cycle slip density is zero during this period, the first reflection distance is assigned a high initial reliability weight of 0.9, while the second reflection distance is weighted at 0.1. After time smoothing, the two distance values are weighted and combined using this time-varying reliability weight. During this observation period, the fused initial reflection distance stabilizes at approximately 1.832 meters.
[0065] By using the time-varying reliability weight, the first reflection distance and the second reflection distance at the same moment are weighted and summed to obtain the initial reflection distance that combines the advantages of both and has stronger robustness.
[0066] This invention transforms discrete cycle slip events into quantified cycle slip density levels and further generates time-varying reliability weights, which is more refined than simple binary judgments. It considers not only the presence or absence of cycle slips but also their frequency, and avoids drastic weight jumps through time smoothing. This refined and dynamic weighting strategy can intelligently allocate resources based on the real-time quality of the two data sources, resulting in a smooth transition of the initial reflection distance after fusion, leading to more stable results that better reflect physical reality.
[0067] Furthermore, based on the normalized amplitude of the reflected interference signal and the preset wheat phenological stage information, the initial reflection distance is corrected for the growth stage to obtain the corrected reflection distance. Corresponding to step S4 above, the specific process is as follows:
[0068] Considering that the physical morphology and electromagnetic properties of wheat plants are not constant throughout the growth cycle, this method further refines the initial reflection distance. By comprehensively analyzing the annual day information of the observation date and the normalized amplitude of the reflected interference signal, the current growth stage of wheat is automatically determined.
[0069] In the early growth stage, when the annual accumulated days are small and the reflected signal amplitude is low, it is determined to be the seedling stage; when the signal amplitude significantly increases and stabilizes at a high level, it is determined to be the rapid growth stage of jointing and heading; when the signal amplitude shows specific attenuation characteristics in the later stage, it is determined to be the maturity and harvest stage.
[0070] A transition period of several days is set between the three main growth stages. When the observation date falls within this transition period, the model is corrected and a weighted combination is used to achieve a smooth transition. Within the transition period from the jointing and heading rapid growth stage to the maturity and harvest stage, the weights of the systematic bias correction applied to the previous stage and the mapping relationship model applied to the subsequent stage change linearly with the date: the former's weight gradually decreases from 1 to 0, while the latter's weight gradually increases from 0 to 1. The final correction is calculated by weighting these two models according to the dynamic weights of the day.
[0071] This treatment eliminates the non-physical step that may occur in the final plant height time series due to abrupt changes in the correction model when switching growth stages, making the inverted corrected reflection distance sequence more smooth and continuous, and more accurately simulating the physical morphology of wheat during growth.
[0072] Based on the year-day (DOY124) of the observation date and the normalized amplitude of the real-time monitored reflected interference signal, if the value is stable below the threshold (0.7 in this embodiment), the growth status of wheat is automatically determined to be in the rapid growth stage of jointing and heading.
[0073] For this stage, the modified model is activated. During the rapid growth stage of jointing and heading, the modified reflection distance is obtained by subtracting the systematic deviation correction from the initial reflection distance. The systematic deviation correction was pre-calibrated experimentally and its value is 0.03 meters. After correction, the modified reflection distance is 1.802 meters.
[0074] This phase of correction generates a more accurate revised reflection distance.
[0075] The wheat growth period is clearly divided into three stages, and a dual assessment is made using normalized amplitude and accumulated days per year, making the identification of growth stages more accurate and robust, avoiding misjudgments that may arise from a single indicator. Applying different correction models with clear physical meanings to different stages provides a precise simulation of the changes in the GNSS signal reflection mechanism during crop growth. This refined growth stage correction strategy overcomes the problem that traditional single models cannot adapt to the entire growth cycle, corrects systematic inversion biases existing at specific growth stages, and ensures the physical authenticity and accuracy of the final plant height time series.
[0076] Further, multi-satellite data fusion is performed on the corrected reflection distance to obtain the final effective reflection height. The wheat plant height is then calculated based on the final effective reflection height and the initial antenna height. Corresponding to step S5 above, the specific process is as follows:
[0077] To obtain the global optimal solution, the observation results from multiple satellites are finally aggregated. This method will perform satellite identifier resolution on the corrected reflection distance sequences calculated by multiple systems and frequency bands such as GPS G15 (L1), GLONASS R08 (G1), Galileo E01 (E1), and BDS C30 (B1-2), and separate the reflection distance subsequences corresponding to different satellites from the data stream that has been corrected during the growth stage.
[0078] Subsequently, time synchronization processing was performed on the reflection distance subsequences of each satellite, aligning these subsequences from different sampling times onto a unified time grid to generate a time-aligned multi-satellite reflection distance dataset. Next, the quality of the time-aligned multi-satellite reflection distance dataset was evaluated, calculating the continuity score and outlier ratio for each satellite's data. For each subsequence in the dataset, the continuity score and outlier ratio for each satellite's data were calculated. The evaluation results showed that the Galileo E01 data sequence had a continuity score of 0.99 and an outlier ratio of only 0.5%; while a BDS satellite with a lower elevation angle might have a continuity score of 0.91 and an outlier ratio reaching 3%.
[0079] Based on the continuity of the data sequence and the proportion of outliers, a corresponding fusion weight is assigned to each satellite, generating a satellite-weight mapping table. E01 is given a higher weight due to its high-quality data and high spatial overlap.
[0080] At a certain moment, the corrected reflection distances from the four satellites mentioned above are 1.810 meters, 1.798 meters, 1.802 meters, and 1.805 meters, respectively. After weighted aggregation, a single, highly stable final effective reflection height of 1.803 meters is generated.
[0081] When aggregating multi-satellite data using a satellite-weighted mapping table, a robust estimation method with stronger resistance to outliers is employed: the weighted median method. At each observation epoch, the corrected reflectance distances of each satellite are sorted in ascending order of value; the fusion weights corresponding to each reflectance distance value remain unchanged, and the weights are accumulated along the sorting sequence until they reach 50% of the total weight.
[0082] The reflection distance value corresponding to the first time the cumulative weight reaches or exceeds 50% is taken as the final effective reflection height.
[0083] This processing significantly enhances the robustness of the multi-satellite data fusion step, ensures a more stable final effective reflectance height sequence, reduces the possibility of isolated spike pulses in the final wheat plant height results, and improves the overall reliability of the inversion results.
[0084] Finally, the initial antenna height was processed using a coordinate system to generate antenna height data of 2.505 meters in a standard coordinate system. Subtracting the calculated final effective reflection height of 1.803 meters, the accurate inversion of the wheat plant height was completed, and the final wheat plant height result was 0.702 meters.
[0085] By independently assessing the quality (continuity, outlier ratio) of reflectance distance subsequences from different satellites and assigning fusion weights accordingly, high-quality satellite data is given greater weight, while low-quality or outlier data is suppressed from contaminating the final result. This multi-perspective weighted aggregation operation is equivalent to performing multiple independent measurements of the target and optimizing their combination. It can effectively smooth random noise and remove outliers, thereby giving the generated final effective reflectance height higher statistical confidence and accuracy.
[0086] This invention significantly improves the overall performance of plant height monitoring through multi-source data fusion and multi-dimensional technology optimization. By employing complementary analysis and dynamic weighting fusion of multiple types of observation data, it effectively reduces noise and outlier interference, improving inversion accuracy and reliability. Combined with adaptive correction based on growth stages, it can adapt to the morphological characteristics of wheat at different growth stages, ensuring monitoring stability. Furthermore, by integrating multi-satellite data and using automated processing, it expands the spatiotemporal coverage and improves monitoring efficiency. This invention possesses good compatibility and scalability, adapting to the monitoring needs of different regions and planting patterns, providing timely and reliable technical support for precision agricultural production management.
[0087] Example 2
[0088] In this embodiment, a wheat plant height inversion method based on GNSS reflection signals is applied to wheat experimental field B, focusing on the weighted fusion acquisition describing the initial reflection distance. (See reference...) Figure 2 The specific implementation process of the method proposed in this invention is as follows:
[0089] To obtain the first reflection distance, this method first performs differential combination of pseudorange and carrier phase observations of the satellite. After this operation, the multipath reflection components are successfully separated, with values fluctuating within a range of plus or minus tens of centimeters, clearly revealing the periodic oscillations caused by wheat canopy reflection. Subsequently, power spectral density analysis is performed on the multipath reflection components to generate a power spectral density map. In the map, a sharp and significant energy peak appears at a frequency of 0.150 Hz, much higher than the surrounding noise floor. This frequency is determined to be the dominant frequency of the low-frequency oscillation. This frequency represents the rate of change of the interferometric field with the satellite elevation angle. Through a predetermined physical conversion relationship, this frequency value is calculated into a specific distance, yielding a first reflection distance of 1.65 meters.
[0090] Meanwhile, to obtain the second reflection distance, the signal strength observations of the same satellite were processed. The original signal strength observations traced a smooth arc during the observation period, rising from a peak of 42 dBH to 49 dBH before slowly decreasing. A third-order polynomial function was used to accurately fit this macroscopic trend. Subtracting this fitted curve from the signal strength observations, which had been converted to a linear scale, yielded a weak reflection interferometric signal with an amplitude of approximately 0.4 linear units. Spectral analysis of this interferometric signal also identified a distinct energy peak in its power spectrum, with a corresponding dominant oscillation frequency of 0.147 Hz, highly consistent with the results of the first method. Combining this with the satellite's geometric position information, this dominant frequency was converted into the second reflection distance, which was 1.68 meters.
[0091] During the aforementioned process, a quality monitoring procedure continuously performs cycle slip detection on the integrity of the carrier phase data. Through high-order inter-epoch differential analysis of the phase data, at 10:22:31 AM, a differential value suddenly increased to 0.35 cycles, significantly exceeding the preset stability threshold of 0.08 cycles. This indicates that a signal lock-off occurred at that moment. Therefore, a phase anomaly marker containing the satellite identifier and the epoch number at that time was generated.
[0092] Finally, the fusion stage begins. The generation of this phase anomaly marker directly affects the allocation of fusion weights. For the time window of 10:22:31 and its vicinity, the reliability level is determined to be low due to the instantaneous increase in cycle slip density. Accordingly, the time-varying reliability weight assigned to the first reflection distance is automatically reduced to 0.2, while the weight for the second reflection distance is correspondingly increased to 0.8. In other time periods without anomaly markers, the weights are typically set to 0.9 and 0.1 to prioritize the adoption of the more accurate phase method results. Using this time-varying weight, the two distance values are weighted and combined to obtain a fusion result of 1.674 meters. By performing this process over the entire observation period, a smooth and robust preliminary reflection distance time series is finally generated.
[0093] Example 3
[0094] This embodiment applies a wheat plant height inversion method based on GNSS reflection signals to a C wheat experimental field at the jointing stage, focusing on describing the wheat plant height acquisition process based on growth stage correction and multi-satellite fusion. (See also...) Figure 3 The specific implementation process of the method proposed in this invention is as follows:
[0095] First, growth stage correction is performed on the initial reflection distance. This method automatically determines the wheat's growth status by analyzing the normalized amplitude of the reflected interference signal and the year-day information. For example, on the observation day with a year-day of 115, the normalized amplitude of the reflected signal was consistently below the threshold (0.7 in this example), and field investigation confirmed that this day was the wheat heading stage. Therefore, this time point was automatically determined to be in the rapid growth stage of jointing and heading.
[0096] After determining the growth stage, a matching physical correction model was invoked. During this rapid growth stage of jointing and heading, the dense canopy causes a certain degree of GNSS signal penetration, resulting in a systematic downward shift of the effective reflecting surface relative to the top of the physical canopy. This leads to an overestimation of the initial inversion result; on DOY115, the uncorrected initial reflection distance was 1.840 meters. To correct this bias, a variable systematic bias correction amount, determined beforehand through experimental calibration, of 0.03 meters was subtracted from the initial reflection distance. After this correction, the corrected reflection distance for that day became 1.810 meters.
[0097] To obtain the global optimal solution, observation results from multiple satellites need to be aggregated. This method first performs satellite identifier resolution on the corrected reflection distance, separating independent reflection distance subsequences corresponding to multiple systems and frequency bands, such as GPS L1, GLONASSG1, Galileo E1, and BDS B1-2, from the data stream after the growth stage correction.
[0098] Subsequently, time synchronization processing is performed on the reflection distance subsequences of each satellite. By interpolation or selection of nearest neighbor epochs, these subsequences from different sampling times are aligned to a unified time grid, generating a time-aligned multi-satellite reflection distance dataset.
[0099] Next, the quality of the time-aligned multi-satellite reflectance distance dataset was evaluated. For each subsequence in the dataset, the continuity score and outlier ratio for each satellite were calculated. In this experiment, the continuity score for the observed GPS L1 band data sequence was 0.98, with an outlier ratio of 1.5%; while the continuity score for GLONASS G1 was 0.95, with an outlier ratio of 2.2%; and for BDS B1-2, the scores were 0.96 and 1.8%, respectively.
[0100] Based on the continuity score and the outlier ratio, a corresponding fusion weight is assigned to each satellite, generating a satellite-weight mapping table. Using this table, a weighted aggregation operation is performed on the time-aligned multi-satellite reflectance distance dataset. At the current time, the corrected reflectance distances corresponding to each satellite are weighted and summed according to the weights in the table. For example, at a certain time in DOY115, the corrected reflectance distances from GPS L1, GLONASS G1, and BDS B1-2 are 1.810 meters, 1.800 meters, and 1.820 meters, respectively. After weighted aggregation, a single, highly stable final effective reflectance altitude is generated, with a value of 1.808 meters.
[0101] Finally, by systematizing the initial antenna height using coordinate systems, antenna height data of 2.520 meters in the standard coordinate system was generated. Subtracting the calculated final effective reflection height of 1.808 meters, the accurate inversion of wheat plant height was completed. The final result obtained was 0.708 meters.
[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for inverting wheat plant height based on GNSS reflection signals, characterized in that, include: Acquire GNSS pseudorange observation data, carrier phase observation data, and signal strength observation data; Based on the pseudorange observation data and the carrier phase observation data, multipath reflection components are extracted, and the first reflection distance is obtained through spectrum analysis; the signal strength observations are detrended, the reflection interference signal is extracted, and the second reflection distance is obtained through spectrum analysis. Cycle slip detection is performed on the carrier phase observation data to generate phase anomaly markers; Based on the phase anomaly markers, determine the time-varying reliability weights; The first reflection distance and the second reflection distance are weighted and fused using time-varying reliability weights to obtain the preliminary reflection distance; Based on the normalized amplitude of the reflected interference signal and the preset wheat phenological information, the initial reflection distance is corrected for the growth stage to obtain the corrected reflection distance; The corrected reflection distance is fused with multi-satellite data to obtain the final effective reflection height. The wheat plant height is then calculated based on the final effective reflection height and the initial antenna height.
2. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The process of obtaining the first reflection distance specifically includes: The pseudorange-carrier phase differential processing is employed to multiply the carrier phase observation data by the signal wavelength at the corresponding frequency point to obtain the equivalent distance value. The equivalent distance value at the same frequency is then subtracted from the pseudorange observation data at the same frequency to extract the multipath reflection component. Power spectral density analysis is applied to the multipath reflection component to generate a spectral distribution containing the frequency and corresponding amplitude. The frequency with the highest amplitude is extracted from the spectral distribution as the dominant frequency of the low-frequency oscillation. Based on the correspondence between the dominant frequency and the multipath delay, the first reflection distance is obtained through numerical conversion.
3. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The process of obtaining the second reflection distance specifically includes: The observed signal intensity is converted into a linear-scale signal amplitude ratio. A polynomial function is used to fit the signal amplitude ratio to generate a fitted curve. The difference between the signal amplitude ratio and the fitted curve is used to separate the reflected interference signal. Power spectral density analysis is applied to the reflected interference signal to generate a spectral distribution. The frequency point with the highest amplitude is selected from the spectral distribution and determined as the dominant oscillation frequency of the reflected interference signal. Combining the geometric relationship between the dominant oscillation frequency and the reflection path difference, the second reflection distance is calculated through numerical conversion.
4. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The cycle slip detection process specifically includes: The cycle slip detection employs an inter-epoch difference method. It selects continuous epoch observations of the carrier phase observation data, performs difference operations on the observation data of adjacent epochs within the continuous epoch observations, and obtains a first-order difference value. Based on the first-order difference value, multiple rounds of adjacent difference operations are performed to generate higher-order difference values. These higher-order difference values are compared one by one with a preset variable difference threshold. For epoch positions where the higher-order difference value exceeds the difference threshold, a phase anomaly marker containing the epoch number is generated.
5. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The process of obtaining the initial reflection distance specifically includes: The cycle slip density in the phase anomaly marker is divided into intervals to generate discrete cycle slip density levels; based on preset physical association rules, corresponding initial reliability weights are assigned to the cycle slip density levels, with higher weights assigned to lower cycle slip densities; time smoothing is performed on the initial reliability weights to generate time-varying reliability weights; using the time-varying reliability weights, a weighted combination calculation is performed on the first reflection distance and the second reflection distance to generate the preliminary reflection distance.
6. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The process of obtaining the corrected reflection distance includes: The entire growth period of wheat is divided into three stages: the seedling stage, the jointing and heading rapid growth stage, and the maturity and harvest stage. By monitoring whether the normalized amplitude of the reflected interference signal is lower than a preset threshold and combining it with the annual accumulated day information, the growth stage to which the current observation time point belongs is automatically determined. Different correction models are applied for different growth stages. In the seedling stage, the initial reflection distance is the corrected reflection distance. During the rapid growth stage of jointing and heading, the corrected reflection distance is obtained by subtracting the systematic deviation correction amount from the initial reflection distance; the systematic deviation correction amount is calibrated in advance through experiments. During the mature harvest stage, the initial reflection distance is corrected using a preset mapping model to obtain the corrected reflection distance. The mapping model is established by regression analysis on a synchronous observation dataset containing an attenuation index and a corresponding penetration deviation. The attenuation index is obtained by normalizing the peak amplitude of the reflected interference signal. The penetration deviation is obtained by adding the initial reflection distance to the synchronously measured actual plant height on the ground and subtracting the initial antenna height.
7. The method for wheat plant height inversion based on GNSS reflection signals according to claim 1, characterized in that, The process of obtaining the wheat plant height specifically includes: parsing the satellite identifiers of the corrected reflection distance to separate the reflection distance subsequences corresponding to different satellites; performing time synchronization processing on the reflection distance subsequences of each satellite to generate a time-aligned multi-satellite reflection distance dataset; performing quality assessment on the time-aligned multi-satellite reflection distance dataset to calculate the continuity score and outlier ratio of each satellite's data; assigning a corresponding fusion weight to each satellite based on the continuity score and the outlier ratio to generate a satellite-weight mapping table; using the satellite-weight mapping table to perform a weighted aggregation operation on the time-aligned multi-satellite reflection distance dataset to generate the final effective reflection height; performing coordinate systematization processing on the initial antenna height to generate antenna height data in a standard coordinate system; and calculating the difference between the final effective reflection height and the antenna height data to generate the wheat plant height.