A method for crop hyperspectral change detection based on spatial reconstruction

By employing a spatial reconstruction-based hyperspectral change detection method, which utilizes boundary filling and spectral weight calculation, the problem of low accuracy in hyperspectral detection is solved, achieving high accuracy and effective differentiation in crop change detection.

CN118675056BActive Publication Date: 2024-11-05CHENGDU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411166037.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-05
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

Existing hyperspectral change detection methods have low detection accuracy in agricultural contexts, are greatly affected by environmental variables, lack specialized algorithms, and have high experimental costs.

Method used

A hyperspectral change detection method for crops based on spatial reconstruction is adopted. By acquiring a hyperspectral dataset, performing boundary filling and sliding window processing, and combining spectral and spatial features, the crop change results are determined by adjusting the weight of neighboring spectra and calculating spectral weights.

Benefits of technology

It improves the accuracy of crop change detection, eliminates reliance on prior knowledge, highlights the boundary details of hyperspectral datasets, reduces noise and boundary effects, and achieves effective differentiation between changed and unchanged pixels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118675056B_ABST
    Figure CN118675056B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of hyperspectral detection of crops, and specifically discloses a method for detecting hyperspectral changes of crops based on spatial reconstruction, comprising the following steps: obtaining a hyperspectral data set of crops, the hyperspectral data set comprising a processing set and a control set, the processing set and the control set being acquired at a first time and a second time, respectively, and the acquisition positions of the processing set and the control set are completely consistent; filling the boundaries of the processing set to obtain an initial processing set; processing the initial processing set based on each central spectrum in the initial processing set and the adjacent spectrum corresponding to each central spectrum to obtain a result set; determining the change results of the hyperspectral data set based on the result set and the control set. The present invention makes full use of spatial morphological properties, integrates Laplace smoothing constraints with spectral information, highlights the boundary details of the hyperspectral data set, and improves the accuracy of the crop change detection task. The problem of low accuracy in hyperspectral image detection of crops in the prior art is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of crop hyperspectral detection, relates to a detection method, and specifically to a crop hyperspectral change detection method based on spatial reconstruction. Background Art

[0002] Sustainable agricultural productivity and food security are two of the most important challenges facing the world in the 21st century. Crop change detection research is the foundation of agricultural sustainability and plays a key role in enhancing the stability of agricultural systems, reducing production risks, and improving the well-being of rural communities, thereby ensuring the sustainability of food supply. Change detection is one of the key applications of remote sensing technology. It has been widely used in crop change detection, environmental monitoring, land change analysis, urban expansion assessment, disaster detection and assessment, and military battlefield monitoring, and has achieved remarkable results.

[0003] In recent years, as the world's major economies have paid more and more attention to the "low-altitude economy", the demand for technologies related to the "low-altitude economy" has also increased. As an important technical branch, hyperspectral / multispectral change detection has gradually become a new research hotspot. Hyperspectral images (HSIs) are very different in structure from ordinary RGB images. HSIs combine spatial and spectral dimensions to present data in the form of a three-dimensional (3D) cube. This structure enables HSIs to provide complex spectral "diagnostic" information, making them a valuable tool for a variety of applications. The rich data information, coupled with its inherent advantages such as wide coverage and fast detection cycle, gives hyperspectral change detection important potential in accurately identifying crop changes and diversified agricultural transformation.

[0004] Although the above algorithms have shown remarkable effectiveness in different disciplines, their practical deployment still faces significant obstacles due to environmental variables such as seasonal fluctuations, topography, and meteorological conditions. Notably, these obstacles include interference caused by anomalous components and insufficient utilization of spatial data within the detection framework. In addition, there is a lack of algorithms specifically designed for crop change detection, as well as the high experimental time cost of these methods. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for detecting crop hyperspectral changes based on spatial reconstruction, which solves the problem of low detection accuracy of the existing detection methods.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for detecting crop hyperspectral changes based on spatial reconstruction, which specifically includes the following steps:

[0007] S1. Acquire a hyperspectral dataset of crops, the hyperspectral dataset comprising a processing set and a control set, the processing set and the control set are acquired at a first time and a second time, respectively, and the acquisition locations of the processing set and the control set are completely consistent;

[0008] S2, filling the boundaries of the processing set to obtain an initial processing set;

[0009] S3, processing the initial processing set based on each central spectrum in the initial processing set and the adjacent spectrum corresponding to each central spectrum to obtain a result set;

[0010] The central spectrum is the spectrum corresponding to the pixel point in the second row and the second column in the sliding window of a preset size, and the preset size of the sliding window is 3*3;

[0011] S4. Determine the change result of the hyperspectral data set according to the result set and the control set, and determine the change result as the change result of the crop.

[0012] The beneficial effects of the above scheme are:

[0013] The present invention can eliminate the reliance on prior knowledge and sample data in the context of agricultural change detection, integrate spectral information, emphasize the boundary lines of different crops, and make full use of spatial morphological characteristics, solving the problem of low detection efficiency caused by irregular spectral pixel interference, achieving effective distinction between changed and unchanged pixels, highlighting the boundary details of hyperspectral datasets, and improving the accuracy of crop change detection tasks.

[0014] On the basis of the above technical solution, the present invention can also be improved as follows:

[0015] Further, step S2 includes:

[0016] S21, determining the size of the sliding window;

[0017] S22. Use a sliding window to perform boundary processing on the processing set according to a preset boundary filling rule to obtain an initial processing set.

[0018] Furthermore, the preset boundary filling rule in step S22 includes:

[0019] The boundary spectrum closest to the spectrum to be filled is determined as the target spectrum, and the boundary spectrum is the boundary spectrum of each data in the hyperspectral dataset;

[0020] The value of the target spectrum is used as the value of the spectrum to be filled to expand the boundaries of the processing set.

[0021] The above further scheme has the beneficial effect that the loss of boundary information can be avoided. When performing sliding window operation on a hyperspectral dataset, the boundary is usually affected by the incomplete window, resulting in the loss of boundary information. The integrity of the boundary can be preserved by boundary filling, and the accuracy of the hyperspectral dataset processing can be improved. At the same time, the boundary effect can be avoided. In the absence of boundary filling, the sliding window operation may cause discontinuity or distortion at the boundary of the hyperspectral dataset. By adding boundary filling, this adverse effect can be reduced, making the hyperspectral dataset processing result smoother and more accurate.

[0022] Further, step S3 includes:

[0023] S31, determining the adjacent spectrum corresponding to each central spectrum;

[0024] S32, determining the adjustment weight of the adjacent spectrum corresponding to each central spectrum according to the value of the adjacent spectrum corresponding to each central spectrum;

[0025] S33, adjusting the value of each central spectrum according to a preset formula, the value of each central spectrum, the value of the adjacent spectrum corresponding to each central spectrum, and the adjustment weight of the adjacent spectrum corresponding to each central spectrum to obtain a result set.

[0026] Further, in step S33, the preset formula is:

[0027]

[0028] in, represents the value of each center spectrum after adjustment, represents the trade-off parameter of the central spectrum, represents the value of the central spectrum, represents the value of the first neighboring spectrum of the central spectrum, represents the adjustment weight of the first neighboring spectrum, represents the value of the second neighboring spectrum of the central spectrum, represents the adjustment weight of the second adjacent spectrum, represents the value of the third neighboring spectrum of the central spectrum, represents the adjustment weight of the third neighboring spectrum, represents the value of the fourth neighboring spectrum of the central spectrum, Represents the adjustment weight of the fourth adjacent spectrum.

[0029] The beneficial effects of the above further scheme are: the hyperspectral dataset is smoothed, and by setting weights so that the surrounding spectra have less influence on the central spectrum, the hyperspectral dataset can be smoothed, the noise and details in the hyperspectral dataset can be reduced, and the hyperspectral dataset can be sharpened to enhance the edges and details of the hyperspectral dataset.

[0030] Further, step S4 includes:

[0031] S41, determining the spectral weight of each spectrum in the result set according to the value of each spectrum in the result set and the value of the spectrum at the corresponding position in the reference set and according to the spectral weight calculation formula;

[0032] S42, determining a parameter size of the detection window, where the parameter size is an odd number greater than 1;

[0033] S43. According to the value of each spectrum within the parameter size in the result set, the value of each spectrum within the parameter size at the corresponding position in the control set, and the spectral weight of the central spectrum of the detection window, the change result of the hyperspectral data set is determined according to the detection formula, and the change result is determined as the change result of the crop.

[0034] Furthermore, the spectrum weight calculation formula in step S41 is:

[0035]

[0036] in, Indicates the result set The value of the spectrum, Indicates the first position of the corresponding position in the control set The value of the spectrum.

[0037] Furthermore, the detection formula in step S43 is:

[0038]

[0039] in, represents the total number of spectral bands, represents the spectral weight, Indicates the first detection window in the 3*3 size in the result set. The spectrum is The value of the band, Indicates the first detection window of size 3*3 in the corresponding position of the control set. The spectrum is The value of a band.

[0040] The beneficial effect of the above further scheme is that according to the spectral weight, the relationship between different spectra can be distinguished more accurately, which helps to extract more meaningful feature information. The parameter size of the detection window is determined in order to take into account the information of surrounding pixels in data processing, so as to analyze the data more comprehensively. Combining the spectral weight and the pixel value within the detection window can more accurately evaluate the changes in the data set, which helps to extract target information or features. The above steps and formulas can improve the accuracy and reliability of data processing, help extract more meaningful information and features from hyperspectral data, and provide support for subsequent analysis and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The figure is a flowchart of a crop hyperspectral change detection method based on spatial reconstruction. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, a crop hyperspectral change detection method based on spatial reconstruction includes the following steps:

[0044] S1. Acquire a hyperspectral dataset of crops. The hyperspectral dataset includes a processing set and a control set. The processing set and the control set are acquired at a first time and a second time, respectively, and the acquisition locations of the processing set and the control set are completely consistent.

[0045] Exemplarily, the hyperspectral dataset of crops can be the Hermiston dataset (i.e., the Hermiston dataset). The Hermiston dataset was collected in Hermiston, Oregon, USA, including farmland and river areas. The Hermiston dataset includes multiple hyperspectral images, which were taken in 2004 and 2007 using the Hyperion sensor (Hyperion sensor, also known as hyperspectral imaging spectrometer, is the first satellite-borne hyperspectral mapping instrument, and is currently the only satellite-borne hyperspectral imaging spectrometer in orbit and the only publicly available hyperspectral measuring instrument). The Hyperion sensor contains a total of 242 spectral bands, each of which is 390×200 pixels in size.

[0046] Optionally, the hyperspectral dataset of crops may include a result set, and the parameters of the result set may include: 242 spectral bands, each band having a size of 390×200 pixels. The result set may be a comparison result of the processing set and the control set standard in the hyperspectral dataset, and may be used for verification with the comparison result of the hyperspectral dataset to improve the accuracy of the experiment.

[0047] Optionally, among the 242 spectral bands in the above embodiment, bands B001-B007, B058-B076 and B225-B242 are not calibrated, so the data of these three bands need to be removed. Therefore, before step S2, the hyperspectral data set of crops can be preprocessed.

[0048] In this embodiment, five different types of changes related to crop conversion are identified. In addition, the Hermiston image data in the complete Hermiston dataset is finely labeled with 78,000 pixels, including 9986 positive pixels and 68,014 negative pixels.

[0049] S2. Fill the boundaries of the processing set to obtain an initial processing set.

[0050] Specifically, step S2 includes:

[0051] S21. Determine the size of the sliding window.

[0052] S22. Use a sliding window to perform boundary processing on the processing set according to a preset boundary filling rule to obtain an initial processing set.

[0053] The preset boundary filling rule in step S22 includes:

[0054] The boundary spectrum closest to the spectrum to be filled is determined as the target spectrum, and the boundary spectrum is the boundary spectrum of each data in the hyperspectral dataset; the value of the target spectrum is used as the value of the spectrum to be filled to expand the boundary of the processing set.

[0055] Exemplarily, the size of the sliding window may be 3x3. The center of the sliding window is placed on the spectrum to be filled, that is, the spectrum to be filled may be regarded as the central spectrum. Therefore, among the 8 nearest spectra of the central spectrum, the boundary spectrum nearest to the central spectrum to be filled is determined as the target spectrum, and the value of the target spectrum is used as the value of the spectrum to be filled, so as to expand the boundary of the processing set.

[0056] S3. Based on each central spectrum in the initial processing set and the adjacent spectrum corresponding to each central spectrum, the initial processing set is processed to obtain a result set; the central spectrum is a spectrum corresponding to the pixel point in the second row and second column in a sliding window of a preset size, and the preset size of the sliding window is 3*3.

[0057] Specifically, step S3 includes:

[0058] S31, determining the adjacent spectrum corresponding to each central spectrum.

[0059] S32 . Determine, according to the value of the adjacent spectrum corresponding to each central spectrum, the adjustment weight of the adjacent spectrum corresponding to each central spectrum.

[0060] S33, adjusting the value of each central spectrum according to a preset formula, the value of each central spectrum, the value of the adjacent spectrum corresponding to each central spectrum, and the adjustment weight of the adjacent spectrum corresponding to each central spectrum.

[0061] In step S33, the preset formula is:

[0062]

[0063] in, represents the value of each center spectrum after adjustment, represents the trade-off parameter of the central spectrum, represents the value of the central spectrum, represents the value of the first neighboring spectrum of the central spectrum, represents the adjustment weight of the first neighboring spectrum, represents the value of the second neighboring spectrum of the central spectrum, represents the adjustment weight of the second adjacent spectrum, represents the value of the third neighboring spectrum of the central spectrum, represents the adjustment weight of the third neighboring spectrum, represents the value of the fourth neighboring spectrum of the central spectrum, Represents the adjustment weight of the fourth adjacent spectrum.

[0064] Optionally, when processing the initial set, a sliding window with a preset size of 3*3 may be selected for processing, and the central spectrum is the spectrum corresponding to the pixel point in the second row and second column of the 3*3 sliding window.

[0065] S4. Determine the change result of the hyperspectral data set according to the result set and the control set, and determine the change result as the change result of the crop.

[0066] Specifically, step S4 includes:

[0067] S41, determining the spectral weight of each spectrum in the result set according to the value of each spectrum in the result set and the value of the spectrum at the corresponding position in the reference set and according to the spectral weight calculation formula.

[0068] The spectral weight calculation formula in step S41 is:

[0069]

[0070] in, Indicates the result set The value of the spectrum, Indicates the first position of the corresponding position in the control set The value of the spectrum.

[0071] S42: Determine a parameter size of the detection window, where the parameter size is an odd number greater than 1.

[0072] S43. According to the value of each spectrum within the parameter size in the result set, the value of each spectrum within the parameter size at the corresponding position in the control set, and the spectral weight of the central spectrum of the detection window, the change result of the hyperspectral data set is determined according to the detection formula, and the change result is determined as the change result of the crop.

[0073] For example, the parameters of the change result of the hyperspectral data set may be: including 242 spectral bands, and the size of each band is 390×200 pixels.

[0074] In this embodiment, the detection formula in step S43 is:

[0075]

[0076] in, represents the total number of spectral bands, represents the spectral weight, Indicates the first detection window in the 3*3 size in the result set. The spectrum is The value of the band, Indicates the first detection window of size 3*3 in the corresponding position of the control set. The spectrum is The value of a band.

[0077] Optionally, after obtaining the change result of the hyperspectral dataset, the change result of the hyperspectral dataset may be compared with a result set included in the hyperspectral dataset of crops to verify the accuracy of the obtained change result.

[0078] The present invention can eliminate the reliance on prior knowledge and sample data in the context of agricultural change detection, integrate spectral information, emphasize the boundary lines of different crops, and make full use of spatial morphological characteristics, solving the problem of low detection efficiency caused by irregular spectral pixel interference, achieving effective distinction between changed and unchanged pixels, highlighting the boundary details of hyperspectral datasets, and improving the accuracy of crop change detection tasks.

[0079] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.

Claims

1. A method for detecting crop hyperspectral changes based on spatial reconstruction, characterized in that: The method comprises: S1. Acquire a hyperspectral dataset of crops, wherein the hyperspectral dataset includes a processing set and a control set, wherein the processing set and the control set are acquired at a first time and a second time, respectively, and acquisition locations of the processing set and the control set are completely consistent; S2, filling the boundaries of the processing set to obtain an initial processing set; S3, processing the initial processing set based on each central spectrum in the initial processing set and the adjacent spectrum corresponding to each central spectrum to obtain a result set; The central spectrum is a spectrum corresponding to the pixel point in the second row and the second column in a sliding window of a preset size, and the preset size of the sliding window is 3*3; S4, determining the change result of the hyperspectral data set according to the result set and the control set, and determining the change result as the change result of the crop; The step S4 specifically includes: S41, determining the spectral weight of each spectrum in the result set according to the value of each spectrum in the result set and the value of the spectrum at the corresponding position in the reference set and according to a spectral weight calculation formula; The spectrum weight calculation formula in step S41 is: in, Indicates the result set The value of the spectrum, Indicates the first position of the corresponding position in the control set The value of the spectrum; S42, determining a parameter size of the detection window, where the parameter size is an odd number greater than 1; S43, determining the change result of the hyperspectral data set according to the value of each spectrum within the parameter size in the result set, the value of each spectrum within the parameter size at the corresponding position in the reference set, and the spectral weight of the central spectrum of the detection window according to the detection formula, and determining the change result as the change result of the crop; The detection formula in step S43 is: in, represents the total number of spectral bands, represents the spectral weight, Indicates the first detection window in the 3*3 size in the result set. The spectrum is The value of the band, Indicates the first detection window of size 3*3 in the corresponding position of the control set. The spectrum is The value of a band.

2. The method according to claim 1, characterized in that The step S2 comprises: S21, determining the size of the sliding window; S22: Use the sliding window to perform boundary processing on the processing set according to a preset boundary filling rule to obtain the initial processing set.

3. The method according to claim 2, characterized in that The preset boundary filling rule in step S22 includes: Determine a boundary spectrum closest to the spectrum to be filled as a target spectrum, wherein the boundary spectrum is a boundary spectrum of each data in the hyperspectral dataset; The value of the target spectrum is used as the value of the spectrum to be filled, so as to expand the boundary of the processing set.

4. The method according to claim 1, characterized in that The step S3 comprises: S31, determining an adjacent spectrum corresponding to each central spectrum; S32, determining the adjustment weight of the adjacent spectrum corresponding to each central spectrum according to the value of the adjacent spectrum corresponding to each central spectrum; S33, adjusting the value of each central spectrum according to a preset formula, the value of each central spectrum, the value of the adjacent spectrum corresponding to each central spectrum, and the adjustment weight of the adjacent spectrum corresponding to each central spectrum to obtain the result set.

5. The method according to claim 4, characterized in that In step S33, the preset formula is: in, represents the value of each center spectrum after adjustment, represents the trade-off parameter of the central spectrum, represents the value of the central spectrum, represents the value of the first neighboring spectrum of the central spectrum, represents the adjustment weight of the first neighboring spectrum, represents the value of the second neighboring spectrum of the central spectrum, represents the adjustment weight of the second adjacent spectrum, represents the value of the third neighboring spectrum of the central spectrum, represents the adjustment weight of the third neighboring spectrum, represents the value of the fourth neighboring spectrum of the central spectrum, Represents the adjustment weight of the fourth adjacent spectrum.