A grain and oil crop scattering wave trough identification method based on harmonic amplitude and phase analysis

Abstract

The present application belongs to the technical field of spaceborne radar data processing and remote sensing mapping, and particularly relates to a grain and oil crop scattering wave trough identification method based on harmonic amplitude and phase analysis. The present application realizes scattering time series reconstruction and denoising through harmonic fitting, and realizes scattering waveform detection and rapid identification of scattering wave trough through amplitude and phase analysis. The present application flexibly and efficiently realizes scattering time series denoising without presetting a sliding window size and without assuming a local fitting function, and rapidly and accurately identifies grain and oil crop scattering wave troughs without relying on complex time series point-by-point operation and secondary discrimination, thereby providing support for long time series and large area agricultural application of radar data. The present application effectively overcomes the problems in the corresponding industry caused by the fact that the existing window statistical method and local fitting method are difficult to determine an adaptive sliding window and a fitting function, and the fact that the difference search method needs complex time series point-by-point operation and secondary discrimination.

Description

Technical Field

[0001] This invention belongs to the field of spaceborne radar data processing and remote sensing mapping technology, specifically relating to a method for identifying troughs in the scattering waves of grain and oil crops based on harmonic amplitude and phase analysis. Background Technology

[0002] Grain and oilseed crops include cereals, legumes, and oilseeds, which are rich in starch, plant protein, and plant oil, respectively, meeting people's basic food and energy needs. Long-term remote sensing Earth observation is an effective means of investigating the dynamics of large-area grain and oilseed crop planting. Optical satellite imaging is easily affected by frequent cloud and rain weather, while satellites equipped with synthetic aperture radar can overcome cloud and fog interference and provide continuous, stable, high-quality long-term radar scattering data, thus making it an important alternative to optical satellites. However, using radar scattering time series to monitor grain and oilseed crop planting requires solving two key problems: time series denoising and time series feature identification.

[0003] Commonly used time series denoising methods are mainly divided into window statistics and local fitting methods. The former calculates time series statistics (such as mean, median, maximum, and minimum values) within a specific sliding window to suppress outliers, while the latter performs function fitting (such as Gaussian, logistic, and polynomial fitting) on ​​time series values ​​within a specific time range to suppress abnormal fluctuations. Both methods can suppress time series noise. Window statistics rely on the reasonable setting of the sliding window size; local fitting relies on reasonable assumptions about local time series trends. However, scattering time series exhibit complex and diverse random noise and local trends, making it difficult to guarantee the adaptability of a specific sliding window and fitting function to different scattering time series.

[0004] Temporal characteristics mainly refer to the peaks (local maxima) and troughs (local minima) of time-series curves. The physical meaning of radar scattering peaks is not clear, as the heading of grain and oil crops, soil roughness, and increased moisture content can all lead to scattering peaks. Scattering troughs generally represent bare and flat ground, corresponding to the land leveling period before planting grain and oil crops. Therefore, scattering troughs are the most reliable temporal characteristics in grain and oil crop planting monitoring. The commonly used method for identifying temporal characteristics is the differential search method. By calculating the first-order difference of the time-series data, the zero point of the first-order difference is searched. Combined with the rising or falling trend of the curve near the zero point, it is determined whether the zero point corresponds to a peak or trough in the scattering time series. However, the differential search method requires point-by-point calculation of the time-series data, thus consuming a large amount of memory and time. In addition, from a longer time window perspective, the troughs identified by the differential search method are sometimes not low scattering values. Therefore, complex secondary discrimination is required for the troughs to determine whether they correspond to flat ground. Summary of the Invention

[0005] To address the aforementioned problems and shortcomings, and to solve the issues of poor adaptability to temporal denoising and low efficiency of temporal feature recognition in existing scattering valley identification methods, this invention provides a method for identifying scattering valleys in grain and oil crops based on harmonic amplitude and phase analysis. This method achieves scattering temporal reconstruction and denoising through harmonic fitting, and detects scattering waveforms and rapidly identifies scattering valleys through amplitude and phase analysis.

[0006] A method for identifying troughs in the scattering waves of grain and oil crops based on harmonic amplitude and phase analysis, the specific steps of which are as follows:

[0007] Step 1. Timing harmonic fitting;

[0008] For the target area, acquire radar VH (Vertically transmitting and Horizontally receiving) polarization scattering data to form a scattering time series S[t], where t represents time and its value range is normalized to 0 to 1, corresponding to the first day to the last day of the year;

[0009]

[0010] In the above formula, 'a' is a constant term, representing the annual average of scattering. 'i' takes values ​​of 1, 2, and 3, representing the order of the cosine term. A i and The magnitudes and phases of the i-th cosine term, respectively, are a and A. i , The value of is unknown and is obtained through least squares fitting.

[0011] Step 2. Harmonic index calculation;

[0012] The i-th cosine term has i troughs, and the trough dates are determined by... The calculation shows that the first-order cosine term has one trough, let the trough date be V1. The second-order cosine term has two troughs, let the trough dates be V1 and V2 respectively. 2a and V 2b The third-order cosine term has three troughs, let the dates of the troughs be V. 3a V 3b and V 3c .

[0013]

[0014] Calculate the proportion of the magnitude of the i-th cosine term to the sum of the magnitudes of all cosine terms:

[0015] P i =A i / (A1+A2+A3)

[0016] Calculate the phase shift of the i-th cosine term relative to the j-th cosine term:

[0017]

[0018] Step 3. Scattering waveform detection;

[0019] The scattering waveform is classified into three types: single-valley, double-valley, and triple-valley.

[0020] When P i and The scattering waveform is determined to be a single-valley S-type when the following four conditions are met:

[0021] S0 type: P1>2 / 3

[0022] Type S1:

[0023] S2 type:

[0024] S3 type:

[0025] When P i and The scattering waveform is determined to be a double-valley type D waveform when the following four conditions are met:

[0026] Type D0: P1 < 1 / 3, P2 ≥ 1 / 3 or P3 ≤ 1 / 2

[0027] Type D1:

[0028] Type D2:

[0029] Type D3: or

[0030] When P i and The scattering waveform is determined to be a three-valley T-type when the following conditions are met:

[0031] Type T0: P1 < 1 / 3, P2 < 1 / 3, P3 > 1 / 2;

[0032] Step 4. Identification of scattering valleys in grain and oil crops: This involves four steps: determining the baseline component, finding the nearest valley, calculating the weighted average, and limiting the valley magnitude.

[0033] Reference component determination: For a single-trough type, the reference component is a first-order cosine term; for a double-trough type, the reference component is a second-order cosine term; for a triple-trough type, the reference component is a third-order cosine term. The number of troughs in the reference component is the same as the number of troughs in the scattered waveform, and the date V of any trough in the scattered waveform is... P All of these can be determined by the date V of the trough of the nearest reference component. x To approximate it.

[0034] Recent trough search: Based on the trough date calculation method in step 2, the dates of all troughs in the three cosine terms are obtained. For V P The corresponding V x Search for the cosine terms that are the same as V in the other two types of cosine terms respectively. x The corresponding reference component trough is the date V of the nearest trough. y and V z Thus, the date V corresponding to the target trough in the scattering waveform has been successfully located. P The trough corresponding date V of the three nearest cosine terms x V y and V z .

[0035] Weighted average calculation: Let V x V y and V z The order i corresponding to the cosine term is i x i y and i z Amplitude A i A respectively x A y and A z Then it can be determined by V x V y and V z V is obtained by weighted average calculation P 1, V P 1 is V P The approximation value is given by the following formula:

[0036]

[0037] Valley magnitude limit: The date of the scattering valley V P Substitute 1 as the t-value into the harmonic fitting formula in step 1 to obtain the magnitude of the scattering valleys. Set 0.02 as a loose threshold for the valley magnitude, and only retain scattering valleys with magnitudes below 0.02. These valleys are the scattering valleys of grain and oil crops. This completes the identification of scattering valleys of grain and oil crops.

[0038] The principles involved in steps 1 to 4:

[0039] The principle of time-series harmonic fitting: The annual scattering time sequence of grain and oil crops can be regarded as the superposition of multiple periodic cosine processes. Grain and oil crops in farmland can be planted up to three times a year. Therefore, the three orders of cosine terms can fully simulate the diversified scattering time sequence of grain and oil crops while suppressing higher-order random noise. Their amplitudes represent the strength of the periodic changes in cultivated land, and their phases represent the time of periodic changes in cultivated land.

[0040] Harmonic index calculation principle: In the initial state, the three orders of cosine terms have the same amplitude and phase. After least squares fitting, the amplitude and phase of the three orders of cosine terms change. By calculating the amplitude ratio of each cosine term, the energy contribution of each cosine term to the scattering time sequence can be determined. By calculating the phase shift between each pair of cosine terms, the positional relationship between the troughs of the cosine terms can be determined.

[0041] Scattered waveform detection principle: The scattered waveform is the result of the combined effect of three orders of cosine terms. The larger the amplitude proportion of the cosine term, the greater its influence on the scattered waveform. Furthermore, when the phase shifts between the cosine terms are in different intervals, the timing of the enhancement and weakening effects between the troughs of the cosine terms will change. By numerically segmenting the amplitude proportion and phase shift of the cosine terms, the type of scattered waveform can be detected.

[0042] Identification of scattering troughs in grain and oil crops: After detecting the scattering waveform, the number of troughs in the scattering time sequence can be determined. Each scattering trough is contributed by the troughs of the three adjacent cosine terms. The trough dates of the three cosine terms can be obtained through phase calculation, and the scattering trough dates can be obtained by weighted averaging of the trough dates of the three cosine terms. After determining the trough dates, the magnitude of the scattering trough can be calculated using the harmonic fitting formula. Leveling the land before planting grain and oil crops results in very low scattering intensity (around 0.01), while other trough magnitudes are usually above 0.02. Therefore, only when the scattering trough magnitude is less than 0.02 is the trough related to the leveling of the land before planting grain and oil crops.

[0043] In summary, this invention achieves scattering time sequence reconstruction and denoising through harmonic fitting, and realizes the detection of scattering waveforms and rapid identification of scattering valleys through amplitude and phase analysis. Existing window statistics methods and local fitting methods are difficult to determine adaptive sliding windows and fitting functions, while the difference search method requires complex time sequence point-by-point calculations and secondary discrimination. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the process of the present invention;

[0045] Figure 2 The timing harmonic fitting diagram is shown in the example.

[0046] Figure 3The scattering waveform detection diagram is shown in the example.

[0047] Figure 4 This is an example of identifying the scattering valleys of grain and oil crops. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0049] A method for identifying troughs in the scattering waves of grain and oil crops based on harmonic amplitude and phase analysis (e.g.) Figure 1 As shown in the figure, the development environment in this embodiment is GEE (Google Earth Engine), and the programming language is JavaScript.

[0050] Step 1: Retrieve Sentinel-1 ascending VH polarization data for Guigang City, Guangxi Zhuang Autonomous Region in 2020. Construct scattering time series using the imageCollection function and perform harmonic fitting using the image.linearRegression function to obtain the undetermined coefficients a and A. i and

[0051] Step 2: Using the mathematical operators add, subtract, multiply, and divide, perform pixel-by-pixel calculations on the following parameters: the trough dates of the three-order cosine components (V1, V...). 2a V 2b V 3a V 3b and V 3c ), the amplitude proportion of each cosine term P i Phase shift between each pair of cosine terms

[0052] Step 3: Use the logical operators and, or, not, gt, and lt to apply the logic to P. i and Pixel-by-pixel numerical segmentation is performed, and the scattering waveforms that conform to different numerical ranges are divided into single-valley (S0, S1, S2 and S3) type, double-valley (D0, D1, D2 and D3) type and triple-valley (T0) type.

[0053] Step 4: Determine the reference component pixel by pixel based on the valley type of the scattering waveform. Use the logical operators gt and lt to search for the valleys among the other two cosine terms that are closest to the valley of the reference component. Use the mathematical operators reduce('sum') and reduce('mean') to calculate the weighted average V of the valley dates. P 1. Use the logical operators gt and mask to remove troughs with a scattering magnitude higher than 0.02.

[0054] This embodiment processes Sentinel-1 data from Guigang City, Guangxi Zhuang Autonomous Region in 2020. Figure 2 The time-series harmonic fitting diagram is shown in the example. Figure 3 The scattering waveform detection diagram is shown in the example. Figure 4 This is an example of identifying the scattering valleys of grain and oil crops.

[0055] As can be seen from the above embodiments, the present invention has achieved the identification of scattering valleys of grain and oil crops in Guigang City, Guangxi Zhuang Autonomous Region in 2020, with an accuracy of 91.3%. The method provided by the present invention flexibly and efficiently realizes scattering time-series denoising without presetting the sliding window size or assuming a local fitting function. Without relying on complex time-series point-by-point calculations and secondary discrimination, it quickly and accurately identifies the scattering valleys of grain and oil crops, providing support for the long-term, large-area agricultural application of radar data.

Claims

1. A method for identifying troughs in the scattering waves of grain and oil crops based on harmonic amplitude and phase analysis, characterized in that, The specific steps are as follows: Step 1. Timing harmonic fitting; For the target area, acquire radar VH polarization scattering data to form a scattering time series S[t], where t represents time and the value range is normalized to 0 to 1, corresponding to the first day to the last day of the year; In the above formula, 'a' is a constant term, representing the annual average of scattering; 'i' takes values ​​of 1, 2, and 3, representing the order of the cosine term; A i and The magnitudes and phases of the i-th cosine term, respectively, are a and A. i , The value of is unknown and is obtained through least squares fitting; Step 2. Harmonic index calculation; The i-th cosine term has i troughs, and the trough dates are determined by... The calculations show that the first-order cosine term has one trough, with the trough date denoted as V1; the second-order cosine term has two troughs, with the trough dates denoted as V1 and V2 respectively. 2a and V 2b The third-order cosine term has three troughs, let the dates of the troughs be V. 3a V 3b and V 3c ; Calculate the proportion of the magnitude of the i-th cosine term to the sum of the magnitudes of all cosine terms: P i =A i (A1+A2+A3) Calculate the phase shift of the i-th cosine term relative to the j-th cosine term: Step 3. Scattering waveform detection; The scattering waveform is classified into three types: single-valley, double-valley, and triple-valley. When P i and The scattering waveform is determined to be a single-valley S-type when the following four conditions are met: S0 type: P1>2 / 3 Type S1: 1 / 3 ≤ P1 ≤ 2 / 3, P2 > 2P3 S2 type: 1 / 3≤P1≤2 / 3, P3>2P2 S3 type: 1 / 3≤P1≤2 / 3, P2≤2P3, P3≤2P2 When P i and The scattering waveform is determined to be a double-valley type D waveform when the following four conditions are met: Type D0: P1 < 1 / 3, P2 ≥ 1 / 3 or P3 ≤ 1 / 2 Type D1: 1 / 3 ≤ P1 ≤ 2 / 3, P2 > 2P3 Type D2: 1 / 3 ≤ P1 ≤ 2 / 3, P3 > 2P2 D3 type: 1 / 3≤P1≤2 / 3, P2≤2P3, P3≤2P2, or When P i and The scattering waveform is determined to be a three-valley T-type when the following conditions are met: Type T0: P1 < 1 / 3, P2 < 1 / 3, P3 > 1 / 2; Step 4. Identification of scattering valleys in grain and oil crops: This involves four steps: determining the reference component, finding the nearest valley, calculating the weighted average, and limiting the valley magnitude. Reference component determination: For a single-trough type, the reference component is a first-order cosine term; for a double-trough type, the reference component is a second-order cosine term; for a triple-trough type, the reference component is a third-order cosine term; the number of troughs in the reference component is the same as the number of troughs in the scattered waveform, and the date V of any trough in the scattered waveform... P All of these can be determined by the date V of the trough of the nearest reference component. x To approximate; Recent trough search: Based on the method for calculating trough dates in step 2, obtain the dates of all troughs in the three cosine terms; for V P The corresponding V x In the other two cosine terms, find the terms that are related to V. x The dates of the corresponding reference component troughs from the most recent trough are V y and V z Thus, the date V corresponding to the target trough in the scattering waveform has been successfully located. P The trough corresponding date V of the three nearest cosine terms x V y and V z ; Weighted average calculation: Let V x V y and V z The order i corresponding to the cosine term is i x i y and i z Amplitude A i A respectively x A y and A z Then it can be determined by V x V y and V z V is calculated using a weighted average. P 1, V P 1 is V P The approximation value is given by the following formula: Valley magnitude limit: The date of the scattering valley V P Substitute 1 as the t value into the harmonic fitting formula in step 1 to obtain the magnitude of the scattering valley. Set 0.02 as a loose threshold for the magnitude of the valley, and only retain the scattering valleys with a magnitude lower than 0.

02. These valleys are the scattering valleys of grain and oil crops. Thus, the identification of scattering valleys of grain and oil crops is completed.

2. The method for identifying troughs in scattering waves of grain and oil crops based on harmonic amplitude and phase analysis as described in claim 1, characterized in that: In step 1, the imageCollection function is used to construct the scattering time series, and the image.linearRegression function is used to implement harmonic fitting.

Images