Single-time camera-mounted hyperspectral remote sensing damage range detection method

Abstract

The invention relates to a hyperspectral image processing and damage assessment method, in particular to a single-time camera-borne hyperspectral remote sensing damage range detection method, and solves the problems that an existing satellite-borne hyperspectral remote sensing image is low in spatial resolution, a single method is high in false alarm rate, and the detection accuracy is high. And damage range detection after the target or the target is damaged is difficult to realize. The method comprises the following steps: firstly, acquiring original data of an airborne hyperspectral remote sensing image of a damaged target area and reflectivity data of a ground object spectrometer; then carrying out pretreatment; calculating a mahalanobis distance difference value and a spectral angle from the target to the center of the background pixel by pixel, and calculating a specific value of the mahalanobis distance difference value and the spectral angle; and finally, carrying out binary segmentation to obtain a damage range. According to the single-time camera-mounted hyperspectral remote sensing damage range detection method, the scattering range or the damage range of the damage target material is extracted by using the airborne hyperspectral remote sensing image and the ground object spectrometer, and the damage range can be obtained without comparing the scattering range or the damage range with the remote sensing image before damage to perform change detection, so that the damage range detection effect is effectively improved.

Description

Technical Field

[0001] The invention relates to a hyperspectral image processing and damage assessment method, and in particular to a single-time camera-borne hyperspectral remote sensing damage range detection method. Background Art

[0002] As a non-contact long-distance detection and data acquisition method, remote sensing is widely used in the fields of land resources, agriculture and forestry, material detection and identification, etc. In terms of damage effect assessment, visual or remote sensing images are usually used; based on change detection technology, the damage information of the target or object is detected by analyzing the differences between remote sensing images of the same area before and after the damage. Hyperspectral remote sensing has the characteristics of unified map and spectrum, integrating image information and spectral information. Its image information can reflect the geometric shape and other information of the detection object, and the spectral information can fully reflect the physical structure and chemical differences of the detection object, and has a more detailed description of the material of the detection object. After the target or object is damaged, due to the influence of explosions, shock waves, etc., the target material will be scattered to a certain range around the target, or the target will be affected by the impact on the surface within a certain range and cause anomalies. Therefore, the damage range detection can be achieved by detecting the target material distribution range or the surface anomaly.

[0003] In practical applications, hyperspectral images have the following problems: different objects with the same spectrum, different spectra for the same object, mixed pixels, and low spectral signal-to-noise ratio. This has a negative impact on its application effect, which is reflected in algorithms such as material identification and anomaly detection, and usually has a high false alarm rate. In addition, since the hyperspectral imaging bands in the range of 400nm-2500nm are as high as hundreds of bands, the spatial resolution of satellite-borne hyperspectral remote sensing images is low, which is suitable for larger-scale target research and makes it difficult to detect the target or the damage range after the target is damaged. Summary of the invention

[0004] The purpose of the present invention is to solve the technical problems that the existing satellite-borne hyperspectral remote sensing images have low spatial resolution, high false alarm rate of a single method, and difficulty in realizing damage range detection after the target or object is damaged, and to provide a single-time camera satellite-borne hyperspectral remote sensing damage range detection method.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A single-time camera onboard hyperspectral remote sensing damage range detection method, which is special in that it includes the following steps:

[0007] 1] Obtain the original data of airborne hyperspectral remote sensing images of the target area after damage;

[0008] 2] Preprocess the original data of airborne hyperspectral remote sensing images to obtain airborne hyperspectral remote sensing images and reflectance data of airborne hyperspectral remote sensing images;

[0009] 3] Use the ground object spectrometer to collect reflectance data of the target area after damage, and obtain the ground object spectrometer reflectance data;

[0010] 4] Preprocess the preprocessed ground object spectrometer reflectance data to obtain smoothed ground object spectrometer reflectance data;

[0011] 5] Re-collect the smoothed ground object spectrometer reflectance data obtained in step 4] according to the number of bands, spectral range and spectral resolution of the airborne hyperspectral remote sensing image reflectance data to obtain the re-collected ground object spectrometer reflectance data;

[0012] 6] Calculate the cosine angle between the reflectance data of the airborne hyperspectral remote sensing image obtained in step 2] and the reflectance data of the ground object spectrometer after re-collection obtained in step 5] pixel by pixel to obtain the spectral angle of all pixels;

[0013] 7] Calculate the Mahalanobis distance difference between the pixels of all airborne hyperspectral remote sensing images preprocessed in step 2] and the target center and the background center pixel by pixel;

[0014] 8] Ratio the Mahalanobis distance difference obtained in step 7] with the corresponding spectral angle obtained in step 6] pixel by pixel to obtain the similarity of all pixels;

[0015] 9] Calculate the mean and variance of the similarity of all pixels, perform binary segmentation, and obtain a binary image. The area in the binary image with pixel values ​​greater than the threshold is taken as the damage range to complete the single-time camera-mounted hyperspectral remote sensing damage range detection.

[0016] Further, step 1] is specifically as follows:

[0017] 1.1. Calibrate the airborne hyperspectral sensor;

[0018] 1.2. The calibrated airborne hyperspectral sensor is used to collect the original data of the airborne hyperspectral remote sensing images of the damaged target area according to the standard data collection process; the standard data collection process includes planning waypoints, collecting whiteboard data, collecting standard reflectance gray cloth data and executing the route.

[0019] Further, step 2] is specifically as follows:

[0020] The raw data of airborne hyperspectral remote sensing images are preprocessed to obtain reflectance data of airborne hyperspectral remote sensing images and floating-point airborne hyperspectral remote sensing images; the preprocessing includes lens correction, reflectance extraction, atmospheric correction, geometric correction and image stitching; each pixel value of the airborne hyperspectral remote sensing image represents the reflectance of the ground object; each pixel value is between 0 and 1.0.

[0021] Further, step 3] is specifically as follows:

[0022] 3.1. Calibration of ground object spectrometer;

[0023] 3.2. Perform dark current correction and optimization on the calibrated ground feature spectrometer, and use the ground feature spectrometer to collect whiteboard data;

[0024] 3.3. Use a ground object spectrometer to collect reflectance data of the target material in the target area after damage, and obtain ground object spectrometer reflectance data.

[0025] Furthermore, in step 4], preprocessing the reflectance data of the ground object spectrometer includes format conversion, export, water vapor absorption and noise band removal, and smoothing of the reflectance data of the ground object spectrometer.

[0026] Further, step 5] is specifically as follows:

[0027] 5.1. Extract pixel spectral curve from any pixel in the airborne hyperspectral remote sensing image reflectance data in step 2];

[0028] 5.2. The spectral range of the smoothed ground object spectrometer reflectance data in step 4] is intercepted to be consistent with the spectral range of the pixel spectral curve described in step 5.1;

[0029] 5.3. According to the number of bands and spectral resolution of the pixel spectral curve described in step 5.1, re-sample the smoothed ground object spectrometer reflectance data intercepted in step 5.2 to obtain the re-sampled ground object spectrometer reflectance data.

[0030] Furthermore, the reflectance data of the airborne hyperspectral remote sensing image obtained in step 2] and the reflectance data of the ground object spectrometer after re-sampling obtained in step 5.3 are vectors of the same dimension.

[0031] Further, step 7] is specifically as follows:

[0032] 7.1. Calculate the average value M of all pixels of the airborne hyperspectral remote sensing image in step 2;

[0033] 7.2. Calculate the whole image covariance matrix τ of the airborne hyperspectral remote sensing image in step 2] -1 ;

[0034] 7.3. The Mahalanobis distance difference D(x) between the pixel of the entire airborne hyperspectral remote sensing image and the target center and the background center is calculated pixel by pixel using the following formula:

[0035]

[0036] Where d is the reflectance data of the ground object spectrometer after resampling, and x is the pixel value.

[0037] Further, step 9] is specifically as follows:

[0038] 9.1. Calculate the mean value u of the similarity of all pixels obtained in step 8;

[0039] 9.2. Calculate the variance σ of the similarity of all pixels obtained in step 8;

[0040] 9.3. Use the following formula to perform binary segmentation on the similarity of all pixels obtained in step 8] to obtain a binary image;

[0041]

[0042] In the formula, binaryimg represents the pixel value after segmentation, and piexl represents the similarity;

[0043] 9.4. The area with pixel value 1 in the binary image is taken as the damage range to complete the single-time camera-borne hyperspectral remote sensing damage range detection.

[0044] Furthermore, the atmospheric correction in step 2] is completed using a standard reflectivity gray cloth;

[0045] The reflectivity data of the damaged target material in step 3.3] is the average value of no less than 5 reflectivity data.

[0046] Beneficial effects of the present invention:

[0047] 1. The present invention discloses a single-time camera onboard hyperspectral remote sensing damage range detection method, which utilizes an airborne hyperspectral remote sensing image and a ground object spectrometer to extract the distribution range or damage range of a damaged target material. It does not need to compare with the remote sensing image before the damage to perform change detection to obtain the damage range, thus reducing image registration, change detection and other processing processes, reducing systematic processing errors, and making up for the problem of low spatial resolution of existing satellite-borne hyperspectral remote sensing images.

[0048] 2. The present invention provides a single-time camera-borne hyperspectral remote sensing damage range detection method, which utilizes the ratio of the Mahalanobis distance difference and its corresponding spectral angle to effectively reduce the false alarm rate of damage range identification and effectively improve the damage range detection effect.

[0049] 3. The single-time camera-mounted hyperspectral remote sensing damage range detection method of the present invention can be applied to the fields of target material diffusion range or abnormal range detection after being affected in other scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of an embodiment of a single-time camera-mounted hyperspectral remote sensing damage range detection method of the present invention;

[0051] Figure 2It is a reflectance data diagram of red brick materials used in buildings collected by a ground object spectrometer in an embodiment of a single-time camera-mounted hyperspectral remote sensing damage range detection method of the present invention;

[0052] Figure 3 Yes Figure 2 Schematic diagram of the red brick material reflectance data after removing the noise band;

[0053] Figure 4 Yes Figure 3 Schematic diagram of the red brick material reflectivity data after smoothing;

[0054] Figure 5 It is a schematic diagram of re-collecting reflectivity data of a ground object spectrometer in an embodiment of a single-time camera-borne hyperspectral remote sensing damage range detection method of the present invention;

[0055] Figure 6 It is a schematic diagram of the damage range in an embodiment of a single-time camera-borne hyperspectral remote sensing damage range detection method of the present invention. DETAILED DESCRIPTION

[0056] like Figure 1 As shown, a single-time camera-mounted hyperspectral remote sensing damage range detection method includes the following steps:

[0057] Step 1: Obtain airborne hyperspectral remote sensing images:

[0058] 1.1. Calibrate the airborne hyperspectral sensor in the laboratory;

[0059] 1.2. The calibrated airborne hyperspectral sensor is used to collect the original data of the airborne hyperspectral remote sensing images of the damaged target area according to the standard data collection process; the standard data collection process includes planning waypoints, collecting whiteboard data, collecting standard reflectance gray cloth data and executing the route;

[0060] Step 2: preprocessing the original data of the airborne hyperspectral remote sensing image to obtain the airborne hyperspectral remote sensing image and the reflectance data of the airborne hyperspectral remote sensing image;

[0061] The raw data of airborne hyperspectral remote sensing images of the damaged target area are preprocessed, including lens correction, reflectivity extraction, atmospheric correction, geometric correction and image stitching, to obtain the reflectivity data of airborne hyperspectral remote sensing images and floating-point airborne hyperspectral remote sensing images; wherein each pixel value of the airborne hyperspectral remote sensing image represents the reflectivity of the ground object, and each pixel value is between 0 and 1.0.

[0062] Preferably, in the preprocessing, the atmospheric correction needs to be completed using a gray cloth with a standard reflectivity. If the atmospheric correction is performed during the period of strong sunlight at noon in summer, a gray cloth with a reflectivity of no more than 50% is used for atmospheric correction, and the reflectivity data of the gray cloth used for atmospheric correction should be the average value of more than 5 pieces;

[0063] Step 3: Obtain reflectivity data from the ground object spectrometer:

[0064] 3.1. Calibrate the ground object spectrometer in the laboratory;

[0065] 3.2. Perform dark current correction and optimization on the calibrated ground feature spectrometer, and use the ground feature spectrometer to collect whiteboard data;

[0066] 3.3. Use the ground object spectrometer to collect the reflectivity data of the target material in the target area after damage, collect 5 reflectivity data and average them to obtain the reflectivity data of the ground object spectrometer;

[0067] like Figure 2 As shown, the reflectance data of red brick materials used in buildings was collected using a ground object spectrometer.

[0068] Step 4, preprocessing the ground object spectrometer reflectance data to obtain smoothed ground object spectrometer reflectance data;

[0069] Since water vapor in the atmosphere absorbs light near the 1400nm and 1800nm ​​bands, there is almost no signal at these wavelengths. Therefore, it is necessary to remove the data near these bands.

[0070] Preprocessing the reflectance data of the ground object spectrometer to obtain smoothed reflectance data of the ground object spectrometer; preprocessing the reflectance data of the ground object spectrometer includes format conversion, export, water vapor absorption and noise band removal, and smoothing of the reflectance data of the ground object spectrometer; Figure 3 For the general Figure 2 Schematic diagram of the red brick material reflectivity data after removing the noise band. Figure 4 For the general Figure 3 Schematic diagram of the data after smoothing the red brick material reflectivity data after removing the noise band;

[0071] Step 5: Resample the reflectance data of the ground object spectrometer to obtain the reflectance data of the ground object spectrometer after resample:

[0072] Generally, the number of bands, band ranges and spectral resolutions of airborne hyperspectral sensors and ground object spectrometers are different. The number of bands of ground object spectrometers is much larger than that of airborne hyperspectral sensors. The reflectance data of ground object spectrometers are re-collected to the level of spectral range, number of bands and spectral resolution of airborne hyperspectral remote sensing images, specifically:

[0073] 5.1. Extracting a pixel spectral curve from any pixel in the airborne hyperspectral remote sensing image reflectance data in step 2;

[0074] 5.2. The spectral range of the smoothed ground object spectrometer reflectance data in step 4 is intercepted to be consistent with the spectral range of the pixel spectral curve in step 5.1;

[0075] 5.3. Re-collect the smoothed ground object spectrometer reflectance data intercepted in step 5.2 according to the number of bands and spectral resolution of the pixel spectral curve described in step 5.1 to obtain the re-collected ground object spectrometer reflectance data;

[0076] like Figure 5 As shown, the data is obtained by re-collecting the reflectance data of the ground object spectrometer obtained in step 4 after smoothing;

[0077] Step 6. Calculate the spectral angle:

[0078] Calculate the cosine angle between the reflectance data of the airborne hyperspectral remote sensing image obtained in step 2 and the reflectance data of the ground object spectrometer after recollection obtained in step 5 pixel by pixel to obtain the spectral angle of all pixels; the reflectance data of the airborne hyperspectral remote sensing image and the reflectance data of the ground object spectrometer after recollection are vectors of the same dimension, so the calculated spectral angle is the cosine angle of the two vectors;

[0079] Step 7: Calculate the Mahalanobis distance:

[0080] In real airborne hyperspectral remote sensing images, it is usually impossible to obtain the true statistical values ​​of the target and background. Therefore, the statistical value of the entire airborne hyperspectral remote sensing image obtained in step 2 is used as the background statistical value, and the statistical value of the reflectance of the ground object spectrometer obtained in step 5.3 is used as the target statistical value;

[0081] 7.1. The pixel mean M of the computer-borne hyperspectral remote sensing image;

[0082] 7.2. The whole image covariance matrix τ of computer-borne hyperspectral remote sensing images -1 ;

[0083] 7.3. The Mahalanobis distance difference D(x) between the pixel of the entire airborne hyperspectral remote sensing image and the target center and the background center is calculated pixel by pixel using the following formula:

[0084]

[0085] In the formula, d is the reflectance data of the ground object spectrometer after re-sampling, and x is the pixel value;

[0086] Step 8: Calculate similarity:

[0087] In order to suppress the false alarm rate and enhance the true value of the detection, the ratio is used as the similarity, and each pixel corresponds to a spectral angle and Mahalanobis distance;

[0088] The Mahalanobis distance difference D(x) obtained in step 7 is compared with the corresponding spectral angle obtained in step 6 for each pixel to obtain the similarity of all pixels; each pixel represents a similarity value, and the higher the similarity value, the higher the probability of being in the damage range;

[0089] Step 9: Extract the damage range:

[0090] 9.1. Calculate the mean value u of the similarity of all pixels obtained in step 8;

[0091] 9.2. Calculate the variance σ of the similarity of all pixels obtained in step 8;

[0092] 9.3. Use the following formula to perform binary segmentation on the similarity of all pixels obtained in step 8 to obtain a binary image;

[0093]

[0094] In the formula, binaryimg represents the pixel value after segmentation, piexl represents the similarity, and the similarity is used as the pixel value before segmentation;

[0095] 9.4. The area with pixel value 1 in the binary image is taken as the damage range, such as Figure 6 As shown, Figure 6 The white area in the middle is the area where the pixel value is 1 after segmentation, indicating the scattered area of ​​red brick material after damage, that is, the target damage range of red brick material.

[0096] The research results show that the single-time camera onboard hyperspectral remote sensing damage range detection method of the present invention has a better damage range detection effect and a faster processing speed under the premise of only using the airborne hyperspectral remote sensing image after the damage. It can be used to detect the range of building materials scattered after the building is damaged, the range of surface damage after the damage, etc.

Claims

1. A single-time camera-borne hyperspectral remote sensing damage range detection method, characterized in that: The following steps are involved: 1] Obtain the original data of airborne hyperspectral remote sensing images of the target area after damage; 2] Preprocess the original data of airborne hyperspectral remote sensing images to obtain airborne hyperspectral remote sensing images and reflectance data of airborne hyperspectral remote sensing images; 3] Use the ground object spectrometer to collect reflectance data of the target area after damage, and obtain the ground object spectrometer reflectance data; 4] Preprocess the preprocessed ground object spectrometer reflectance data to obtain smoothed ground object spectrometer reflectance data; 5] Re-collect the smoothed ground object spectrometer reflectance data obtained in step 4] according to the number of bands, spectral range and spectral resolution of the airborne hyperspectral remote sensing image reflectance data to obtain the re-collected ground object spectrometer reflectance data; 6] Calculate the cosine angle between the reflectance data of the airborne hyperspectral remote sensing image obtained in step 2] and the reflectance data of the ground object spectrometer after re-collection obtained in step 5] pixel by pixel to obtain the spectral angle of all pixels; 7] Calculate the Mahalanobis distance difference between the pixels of all airborne hyperspectral remote sensing images preprocessed in step 2] and the target center and the background center pixel by pixel; 8] Ratio the Mahalanobis distance difference obtained in step 7] with the corresponding spectral angle obtained in step 6] pixel by pixel to obtain the similarity of all pixels; 9] Calculate the mean and variance of the similarity of all pixels, perform binary segmentation, and obtain a binary image. The area in the binary image with pixel values ​​greater than the threshold is taken as the damage range to complete the single-time camera-mounted hyperspectral remote sensing damage range detection.

2. According to claim 1, a single-time camera-mounted hyperspectral remote sensing damage range detection method is characterized in that: Step 1] Specifically: 1.

1. Calibrate the airborne hyperspectral sensor; 1.

2. The calibrated airborne hyperspectral sensor is used to collect the original data of the airborne hyperspectral remote sensing images of the damaged target area according to the standard data collection process; the standard data collection process includes planning waypoints, collecting whiteboard data, collecting standard reflectance gray cloth data and executing the route.

3. A single-time camera-mounted hyperspectral remote sensing damage range detection method according to claim 1 or 2, characterized in that: Step 2] Specifically: The raw data of airborne hyperspectral remote sensing images are preprocessed to obtain reflectance data of airborne hyperspectral remote sensing images and floating-point airborne hyperspectral remote sensing images; the preprocessing includes lens correction, reflectance extraction, atmospheric correction, geometric correction and image stitching; each pixel value of the airborne hyperspectral remote sensing image represents the reflectance of the ground object; each pixel value is between 0 and 1.

0.

4. According to claim 3, a single-time camera-mounted hyperspectral remote sensing damage range detection method is characterized in that: Step 3] Specifically: 3.

1. Calibration of ground object spectrometer; 3.

2. Perform dark current correction and optimization on the calibrated ground feature spectrometer, and use the ground feature spectrometer to collect whiteboard data; 3.

3. Use a ground object spectrometer to collect reflectance data of the target material in the target area after damage, and obtain ground object spectrometer reflectance data.

5. According to claim 4, a single-time camera-mounted hyperspectral remote sensing damage range detection method is characterized by: In step 4], the reflectance data of the ground object spectrometer is preprocessed, including format conversion, export, water vapor absorption and noise band removal, and smoothing of the reflectance data of the ground object spectrometer.

6. According to claim 5, a single-time camera-mounted hyperspectral remote sensing damage range detection method is characterized in that: Step 5] Specifically: 5.

1. Extract pixel spectral curve from any pixel in the airborne hyperspectral remote sensing image reflectance data in step 2]; 5.

2. The spectral range of the smoothed ground object spectrometer reflectance data in step 4] is intercepted to be consistent with the spectral range of the pixel spectral curve described in step 5.1; 5.

3. According to the number of bands and spectral resolution of the pixel spectral curve described in step 5.1, re-sample the smoothed ground object spectrometer reflectance data intercepted in step 5.2 to obtain the re-sampled ground object spectrometer reflectance data.

7. The method for detecting damage range using a single-time camera-mounted hyperspectral remote sensing according to claim 6, characterized in that: The reflectance data of the airborne hyperspectral remote sensing image obtained in step 2] and the reflectance data of the ground object spectrometer after re-sampling obtained in step 5.3 are vectors of the same dimension.

8. According to claim 7, a single-time camera-mounted hyperspectral remote sensing damage range detection method is characterized in that: Step 7] Specifically: 7.

1. Calculate the average value M of all pixels of the airborne hyperspectral remote sensing image in step 2]; 7.

2. Calculate the whole image covariance matrix τ of the airborne hyperspectral remote sensing image in step 2] -1 ; 7.

3. The Mahalanobis distance difference D(x) between the pixel of the entire airborne hyperspectral remote sensing image and the target center and the background center is calculated pixel by pixel using the following formula: Where d is the reflectance data of the ground object spectrometer after resampling, and x is the pixel value.

9. The method for detecting damage range using a single-time camera-mounted hyperspectral remote sensing according to claim 8, characterized in that: Step 9] Specifically: 9.

1. Calculate the mean value u of the similarity of all pixels obtained in step 8; 9.

2. Calculate the variance σ of the similarity of all pixels obtained in step 8; 9.

3. Use the following formula to perform binary segmentation on the similarity of all pixels obtained in step 8] to obtain a binary image; In the formula, binaryimg represents the pixel value after segmentation, and piexl represents the similarity; 9.

4. The area with pixel value 1 in the binary image is taken as the damage range to complete the single-time camera-borne hyperspectral remote sensing damage range detection.

10. The method for detecting damage range using a single-time camera-mounted hyperspectral remote sensing according to claim 9, characterized in that: In step 2], atmospheric correction is done using standard reflectivity grey cloth; The reflectivity data of the damaged target material in step 3.3] is the average value of no less than 5 reflectivity data.

Images