A blind pixel detection method for push-broom infrared hyperspectral imager

Abstract

The present invention discloses a blind pixel detection method applicable to a push-broom infrared hyperspectral imager, and relates to the field of infrared hyperspectral remote sensing. The problem to be solved is the blind pixel detection problem of a push-broom infrared hyperspectral imager. The method has the following steps: (1) black body data acquisition; (2) blind pixel detection based on black body data; (3) data repair based on blind pixel mask; (4) flash pixel detection and data repair based on infrared hyperspectral imaging data itself. The present invention makes full use of the spectral continuity characteristics of infrared hyperspectral data to realize blind pixel detection of infrared hyperspectral imagers, and is a highly efficient, accurate and stable blind pixel detection method for infrared hyperspectral imagers.

Description

Technical Field

[0001] The invention relates to the field of infrared hyperspectral remote sensing, and in particular to a blind pixel detection method suitable for a push-broom infrared hyperspectral imager. Background Art

[0002] Infrared hyperspectral remote sensing is an important branch of the field of hyperspectral remote sensing, and has great application potential in geological surveys, atmospheric environment monitoring and other fields. As the core component of the infrared imaging system, the infrared focal plane is inevitably affected by factors such as manufacturing process, materials, and ambient temperature, and blind pixels will inevitably exist. Blind pixels are generally divided into dead pixels, noise pixels, dark currents, and flicker pixels. From the characteristics of blind pixels, it can be seen that blind pixels often lose their normal energy response capabilities and must be detected and repaired from the data.

[0003] Blind pixel detection methods can be divided into calibration methods and scene-based detection methods. The calibration method is to obtain an image with uniform incident radiation by imaging a reference source (such as a black body), and use difference detection or sequential image statistics. Difference detection subtracts high and low temperature images, and the result is greater than the threshold value, which is considered a blind pixel. The sequential image statistics method calculates the pixel response rate and noise voltage at different temperatures, and compares the value of each pixel with the average response rate or pixel noise voltage. Those that are greater than 10 times the threshold or less than 1 / 10 of the threshold are considered blind pixels. The scene-based detection method is mainly based on the difference in the response characteristics of blind pixels and normal pixels, directly detecting the position of blind pixels from actual imaging data, and often using a two-dimensional space method with a window template.

[0004] The current blind pixel detection method is mainly based on the image space dimension, that is, it processes the single-band array infrared image, which is limited in the blind pixel detection of the infrared hyperspectral imager with push-broom imaging mode. The existing blind pixel detection method for the infrared hyperspectral imager with push-broom imaging mode requires black body data of multiple temperatures as a reference. The present invention is aimed at the blind pixel detection of the push-broom thermal infrared hyperspectral imager, fully utilizes the spectral continuity characteristics of the infrared hyperspectral data, and proposes a blind pixel detection method suitable for the push-broom infrared hyperspectral imager. Summary of the invention

[0005] In view of the existing technical gaps and shortcomings, the technical problem to be solved by the present invention is to provide a blind pixel detection method suitable for a push-broom infrared hyperspectral imager with better effect.

[0006] In order to solve the above technical problems, the present invention provides a blind pixel detection method applicable to a push-broom infrared hyperspectral imager. The specific technical solution adopted by the present invention is as follows:

[0007] 1) Acquisition of blackbody data. An infrared hyperspectral imager is used to image the blackbody and obtain a high-temperature blackbody data and a low-temperature blackbody data;

[0008] 2) Blind pixel detection based on blackbody data. It mainly includes spectral dimension filtering detection, spatial dimension noise detection and response slope detection. It can detect dead pixels, noise pixels and dark pixels and obtain blind pixel masks.

[0009] The spectral filtering detection method is:

[0010] (1) The time dimension average value of the obtained high and low temperature calibration blackbody data cubes is calculated as follows:

[0011]

[0012] Among them, DN bb (i, j, k) is the calibration blackbody data cube, the values ​​of bb are cbb and wbb, representing low-temperature blackbody and high-temperature blackbody respectively; i, j, k are the column, row, and band numbers of the data cube respectively; N L,bb is the number of rows in the data cube; DN bb (i, k) is the calibrated blackbody data after time dimension averaging.

[0013] (2) Calibrated blackbody data DN after time dimension averaging bb (i, k) is spectrally filtered. Spectral filtering uses a combination of median filtering and SG filtering, first median filtering and then SG filtering. This operation can avoid the influence of large outliers on SG smoothing filtering and obtain a more accurate trend line. The calculation method is

[0014] DN bb,filtered (i,·)=SG{median[DN bb (i,·)]}

[0015] Among them, DN bb,filtered (i,·) are the data after spectral filtering; median(·) and SG(·) are median filtering and SG filtering functions respectively.

[0016] (3) Calculate the time-averaged calibrated blackbody data DN before and after filtering bb (i,k) and DN bb,filtered The difference between (i, k) is calculated as

[0017] DN bb，diff (i,k)=DN bb (i,k)-DN bb,filtered (i,k)

[0018] Among them, DN bb，diff (i, k) is the difference between the calibrated blackbody data after time dimension averaging before and after filtering.

[0019] (4) According to the difference image, the blind pixel mask 1 is extracted column by column using the 3σ criterion. 1 (i,k), the extraction method is

[0020]

[0021] Among them, σ bb (i) is the root mean square of the i-th column in the high and low temperature difference image of the calibrated blackbody. The calculation method of σ(i) is

[0022]

[0023] Among them, N B is the number of bands of the infrared hyperspectral imager.

[0024] The spatial noise detection method is:

[0025] (1) Calculate the noise of the obtained high and low temperature calibration blackbody data cube in the time dimension. The calculation method is:

[0026]

[0027] (2) Calculate the absolute deviation between the pixel noise in each band and the average pixel noise in that band. The calculation method is:

[0028]

[0029] Among them, Noise bb,AD (i,k) absolute deviation of pixel noise; is the mean pixel noise in a certain band.

[0030] (3) According to the absolute deviation image of pixel noise, the blind pixel mask 2 is extracted band by band using the 3σ criterion. 2 (i,k), the extraction method is

[0031]

[0032] Among them, σ bb,Noise (k) is the root mean square of the absolute deviation of the pixel noise in the kth band of the image. bb,Noise The calculation method of (k) is

[0033]

[0034] Among them, N B is the number of columns in the ATHIS data cube.

[0035] The response slope detection method is,

[0036] (1) Calculate the response slope of each detection unit based on the high and low temperature calibration blackbody data. The calculation method is:

[0037] Slope(i,k)=regress[X(i,k),Y(i,k)]

[0038] For the airborne calibration blackbody,

[0039] X(i,k)=[DN cbb (i,k),DN wbb (i,k)],Y(i,k)=[L cbb (k),L wbb (k)]

[0040] Where Slope(i,k) is the slope of the i-th column and the k-th band; regress(·) is the regression function; X(i,k) is the sequence of pixel values ​​of the black bodies calibrated at different temperatures after dark background correction; Y(i,k) is the sequence of radiation brightness corresponding to the black bodies calibrated at different temperatures.

[0041] (2) Extract blind pixel mask 3 based on the slope image, mask 3 (i,k), the extraction method is

[0042]

[0043] 3) Data restoration based on blind pixel mask. According to the blind pixel mask, the mean value of the non-blind pixel in the 3×3 neighborhood of each abnormal pixel in a frame of data is selected as the value of the abnormal pixel. If there is no non-abnormal pixel in the 3×3 neighborhood, the search window is expanded until the mean value of the non-blind pixel can be obtained;

[0044] 4) Flash detection based on infrared hyperspectral imaging data. For infrared hyperspectral imaging data, spectral filtering is used to detect abnormal bands in the spectrum of each pixel caused by flashes frame by frame. The values ​​of abnormal bands are repaired by interpolating adjacent bands, and flash detection and repair are completed to obtain the final infrared hyperspectral imaging data. The spectral filtering method is

[0045] (1) Infrared hyperspectral imaging data DN img (i, j, k) is used for spectral filtering. Spectral filtering uses a combination of median filtering and SG filtering, first median filtering and then SG filtering. The calculation method is:

[0046] DN img,filtered (i,j,·)=SG{median[DN img (i,j,·)]}

[0047] Among them, DN img,filtered(i,j,·) is the data after spectral filtering; median(·) and SG(·) are median filtering and SG filtering functions respectively.

[0048] (2) Calculate the time-averaged calibration blackbody data DN before and after filtering img (i,j,k) and DN bb,filtered The difference between (i, k) is calculated as

[0049] DN img，diff (i,j,k)=DN img (i,j,k)-DN img,filtered (i,j,k)

[0050] Among them, DN img，diff (i, j, k) is the difference of the calibrated blackbody data after time dimension averaging before and after filtering.

[0051] (3) According to the difference image, the 3σ criterion is used to extract flash elements column by column. The extraction method is:

[0052]

[0053] Among them, σ bb (i, j) is the root mean square of pixel (i, j) in infrared hyperspectral imaging data. The calculation method of σ(i, j) is

[0054]

[0055] Among them, N B is the number of bands of the infrared hyperspectral imager.

[0056] The beneficial effects of the invention are as follows: The present invention makes full use of the continuous characteristics of infrared spectral information and performs blind pixel detection from multiple dimensions such as spectral dimension filtering, spatial dimension noise, response slope, and data's own information. It is a high-efficiency, low-computational complexity, high-accuracy, and stable blind pixel detection method for push-broom infrared hyperspectral imagers. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Flow of blind pixel detection method for push-broom infrared hyperspectral imager. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein take the airborne thermal-infrared hyperspectral imaging system (ATHIS) as an example, which is only used to explain the present invention and is not limited to the present invention. All similar methods and similar variations of the present invention should be included in the protection scope of the present invention.

[0059] like Figure 1 As shown, a blind pixel detection method applicable to a push-broom infrared hyperspectral imager comprises the following steps:

[0060] (1) Blackbody data acquisition. Use an airborne thermal infrared hyperspectral imaging system to image the blackbody and obtain a high-temperature and a low-temperature blackbody data cube. The data cube should have no less than 1000 frames.

[0061] (2) Blind pixel detection based on blackbody data. It mainly includes spectral dimension filtering detection, spatial dimension noise detection and response slope detection. It can detect dead pixels, noise pixels and dark pixels and obtain blind pixel masks;

[0062] (3) Data restoration based on blind pixel mask. According to the blind pixel mask, the mean value of the non-blind pixel in the 3×3 neighborhood of each abnormal pixel in a frame of data is selected as the value of the abnormal pixel. If there is no non-abnormal pixel in the 3×3 neighborhood, the search window is expanded until the mean value of the non-blind pixel can be obtained;

[0063] (4) Flash detection based on infrared hyperspectral imaging data. For infrared hyperspectral imaging data, spectral filtering is used to detect abnormal bands in the spectrum of each pixel caused by flashes frame by frame. The values ​​of abnormal bands are repaired by interpolating adjacent bands, completing flash detection and repair, and obtaining the final infrared hyperspectral imaging data.

Claims

1. A blind pixel detection method suitable for a push-broom infrared hyperspectral imager, characterized in that The following steps are involved: 1) Acquisition of blackbody data: using an infrared hyperspectral imager to image the blackbody and obtain a high-temperature and a low-temperature blackbody data; 2) Blind pixel detection based on blackbody data, including spectral dimension filtering detection, spatial dimension noise detection and response slope detection, to achieve the detection of dead pixels, noise pixels and dark pixels, and obtain blind pixel masks; The spectral filtering detection method is: (1) The average value of the obtained high and low temperature calibration blackbody data cubes in the time dimension is calculated as follows: Among them, DN bb (i, j, k) is the calibration blackbody data cube, the values ​​of bb are cbb and wbb, representing low-temperature blackbody and high-temperature blackbody respectively; i, j, k are the column, row, and band numbers of the data cube respectively; N L,bb is the number of rows in the data cube; DN bb (i, k) is the calibrated blackbody data after time dimension averaging; (2) Calibrated blackbody data DN after time dimension averaging bb (i, k) is subjected to spectral filtering. The spectral filtering adopts the combination of median filtering and SG filtering. Median filtering is performed first and then SG filtering. This operation can avoid the influence of large outliers on SG smoothing filtering and obtain a more accurate trend line. The calculation method is: DN bb,filtered (i,·)=SG{median[DN bb (i,·)]} Among them, DN bb,filtered (i,·) Data after spectral filtering; median(·) and SG(·) are median filtering and SG filtering functions, respectively; (3) Calculate the time-averaged calibrated blackbody data DN before and after filtering bb (i,k) and DN bb,filtered The difference between (i, k) is calculated as: DN bb，diff (i,k)=DN bb (i,k)-DN bb,filtered (i,k) Among them, DN bb，diff (i, k) is the difference of the calibrated blackbody data after time dimension averaging before and after filtering; (4) According to the difference image, the blind pixel mask 1, mask1(i,k), is extracted column by column using the 3σ criterion. The extraction method is: Among them, σ bb (i) is the root mean square of the i-th column in the high and low temperature difference image of the calibrated blackbody, and the calculation method of σ(i) is: Among them, N B is the number of bands of the infrared hyperspectral imager; The spatial noise detection method is: (1) Calculate the noise of the obtained high and low temperature calibration blackbody data cube in the time dimension. The calculation method is: (2) Calculate the absolute deviation between the pixel noise in each band and the average pixel noise in that band. The calculation method is: Among them, Noise bb,AD (i,k) absolute deviation of pixel noise; is the mean pixel noise value of a certain band, (3) According to the absolute deviation image of pixel noise, the blind pixel mask 2, mask2(i,k), is extracted band by band using the 3σ criterion. The extraction method is: Among them, σ bb,Noise (k) is the root mean square of the absolute deviation of the pixel noise in the kth band of the image, σ bb,Noise The calculation method of (k) is: Among them, N B is the number of columns in the ATHIS data cube, The response slope detection method is: (1) Calculate the response slope of each detection unit based on the high and low temperature calibration blackbody data. The calculation method is: Slope(i,k)=regress[X(i,k),Y(i,k)] For the airborne calibration blackbody, X(i,k)=[DN cbb (i,k),DN wbb (i,k)],Y(i,k)=[L cbb (k),L wbb (k)] Where Slope(i,k) is the slope of the kth band in the i-th column; regress(·) is the regression function; X(i,k) is the sequence of pixel values ​​of the black bodies calibrated at different temperatures after dark background correction; Y(i,k) is the sequence of radiation brightness corresponding to the black bodies calibrated at different temperatures; (2) Extract blind pixel mask 3, mask3(i,k) according to the slope image. The extraction method is: 3) Data restoration based on blind pixel mask: According to the blind pixel mask, the mean value of non-blind pixel in the 3×3 neighborhood of each abnormal pixel in one frame of data is selected as the value of the abnormal pixel. If there is no non-abnormal pixel in the 3×3 neighborhood, the search window is expanded until the mean value of the non-blind pixel can be obtained; 4) Based on the flash detection of infrared hyperspectral imaging data itself, the abnormal bands in the spectrum of each pixel caused by the flash are detected frame by frame by spectral filtering for infrared hyperspectral imaging data, and the values ​​of the abnormal bands are repaired by interpolation of adjacent bands. The flash detection and repair are completed to obtain the final infrared hyperspectral imaging data; the spectral filtering method is: (1) Infrared hyperspectral imaging data DN img (i, j, k) is subjected to spectral filtering. The spectral filtering is a combination of median filtering and SG filtering. Median filtering is performed first and then SG filtering. The calculation method is: DN img,filtered (i,j,·)=SG{median[DN img (i,j,·)]} Among them, DN img,filtered (i, j, ·) is the data after spectral filtering; median(·) and SG(·) are median filtering and SG filtering functions respectively; (2) Calculate the time-averaged calibration blackbody data DN before and after filtering img (i,j,k) and DN bb,filtered The difference between (i, k) is calculated as: DN img，diff (i,j,k)=DN img (i,j,k)-DN img,filtered (i,j,k) Among them, DN img，diff (i, j, k) is the difference of the calibrated blackbody data after the time dimension average before and after filtering; (3) According to the difference image, the flash element is extracted column by column using the 3σ criterion. The extraction method is as follows: Among them, σ bb (i, j) is the root mean square of pixel (i, j) in infrared hyperspectral imaging data, and the calculation method of σ(i, j) is: Among them, N B is the number of bands of the infrared hyperspectral imager.

Images