Thermal infrared remote sensing image adaptive denoising method based on background information

Abstract

The invention discloses a thermal infrared remote sensing image adaptive denoising method based on background information, which comprises the following steps: carrying out nonlinear fitting on background data output under a background radiation condition, and establishing a nonlinear fitting model of the background information of a remote sensor; performing longitudinal rough matching on the noise data of the original image and the nonlinear fitting model of the background information, and selecting the background information matched with the noise data; judging the initial position of the stripe period of the noise data frame by frame, performing transverse fine matching with the obtained matched background information, and obtaining a background fitting curve consistent with the noise data; establishing a one-to-one correspondence relationship between the fitted curve and the noise of the original image, and performing periodic noise removal on the original image; and repeating the steps until the periodic noise of the image is completely removed. Original detail information of the image can be reserved to the maximum extent, the image quality is effectively improved, the automation degree is high, and too much manual intervention is not needed.

Description

Technical Field

[0001] The invention relates to the technical field of infrared spectrum imaging image processing, and in particular to a thermal infrared remote sensing image adaptive denoising method based on background information. Background Art

[0002] As long as the temperature of a substance is above absolute zero, it will emit electromagnetic radiation. The temperature of the objects and atmosphere on the earth's surface is much lower than that of the solar photosphere, so the self-emitted radiation of the earth's surface is mainly concentrated in the thermal infrared spectrum with a longer wavelength than that of visible light. In the thermal infrared atmospheric window spectrum with high atmospheric transmittance, thermal infrared remote sensing technology can detect the temperature and emissivity (also known as relative emissivity) of land and ocean, and then identify the type of objects and environmental conditions; in the non-window spectrum with low atmospheric transmittance, thermal infrared remote sensing technology can detect the vertical temperature distribution of the atmosphere, the content of absorbing gases and their vertical distribution. Compared with visible, near-infrared and short-wave infrared, thermal infrared remote sensing technology has unique advantages. It mainly uses thermal induction to obtain information. It can obtain large-area ground temperature field information in an instant or in a relatively short time. It has very wide applications in night target detection, hidden target detection and forest fire monitoring.

[0003] Thermal infrared imaging payloads are usually composed of scanning mirrors (pointing mirrors), optical systems, detectors, information processors, system controllers, calibration black bodies, cooling systems and other components. Considering the impact of imaging on speed and sensitivity, it is necessary to use a high-performance, high-signal-to-noise ratio thermal infrared focal plane detector. In the prior art, a cooling infrared focal plane device is generally used, which consists of pixels arranged in a matrix and peripheral circuits. Its imaging method is as follows: Figure 1As shown, the spatial pixel is a plurality of pixels arranged in parallel, and a slow scan and a fast return are performed along the scanning direction. The first scanning strip 01 and the second scanning strip 02 are slow scanning imaging strips, and the fast return strip 03 is a fast return strip, which is not imaged. However, due to the inconsistency of the response characteristics of each detector in the focal plane array and the inconsistency of the readout circuit, the thermal infrared imaging detector finally causes a low signal-to-noise ratio of the image formed when detecting low-temperature and other uniform target objects. In satellite-borne thermal infrared images, the most common form of noise is strip noise, which is a special noise with certain periodicity, directionality and strip distribution in remote sensing images. However, through noise analysis based on the imaging mechanism and working characteristics of the thermal infrared camera, it is found that a scene image is spliced ​​by multiple continuous scanning strips, and the starting phases of the noise presented on multiple scanning strips of each scene image are inconsistent. The noise of each scanning strip presents a periodic phenomenon, but the periodicity is not completely fixed. The presence of strip noise causes problems in both qualitative and quantitative use of thermal infrared remote sensing images. In qualitative remote sensing interpretation and judgment, strip noise seriously affects the identification and extraction of ground object types by operators. In quantitative remote sensing product inversion, strip noise will distort the radiation information, and the impact on the inversion results cannot be evaluated, affecting the subsequent application of remote sensing images.

[0004] Image denoising is to use various technical means to filter out image noise and improve image quality to meet the subsequent application of the image. At present, the methods for removing stripe noise in infrared images mainly include Fourier transform method, moment matching method, histogram matching method, wavelet transform denoising method, etc. Although these traditional methods have certain effects on removing stripe noise in thermal infrared images, these methods are based on the conditions of the image itself for denoising, requiring the type of ground objects in the image to be single, and to a certain extent will cause the noise equivalent temperature difference to become larger and lose the effective information of the image. In particular, for the working characteristics of thermal infrared cameras, its main imaging mode is slow scanning and fast return swing scanning imaging, but the starting phase of the noise of each scanning line is inconsistent, showing the phenomenon of inconsistency between the first scanning line and the second scanning line on the image. The traditional stripe denoising method based on the image itself is difficult to eliminate this type of noise. Therefore, in order to ensure the effective removal of image periodic noise, maintain the original sensitivity characteristics of the image, and improve image quality and clarity, it is necessary to study the noise suppression and removal method from the mechanism of this type of noise generation. Summary of the invention

[0005] The present invention proposes an adaptive denoising method for thermal infrared remote sensing images based on background information, which can maximally retain the original detail information of the image while eliminating the stripe noise of the thermal infrared image, effectively improve the image quality and ensure the image clarity. The method has a simple algorithm, fast processing speed and high automation, and can realize batch processing of massive thermal infrared image data without much manual intervention.

[0006] To this end, the present invention adopts the following technical solutions:

[0007] A thermal infrared remote sensing image adaptive denoising method based on background information, the method comprising the following steps:

[0008] Step 1. Perform nonlinear fitting on the background data output by the thermal infrared remote sensor detector under background radiation conditions according to the scanning strip sequence to establish a nonlinear fitting model of the remote sensor background information;

[0009] Step 2. Perform longitudinal rough matching between the noise data of the original image of the thermal infrared remote sensor and the nonlinear fitting model of the background information, and select the background information data matching the noise data from the remote sensor background information;

[0010] Step 3. Determine the starting position of the noise data stripe period frame by frame, and perform horizontal precise matching with the background information data obtained in step 2 to match the background data position consistent with the starting phase of the noise data, and obtain a background data fitting curve consistent with the noise data;

[0011] Step 4. Fit the background data obtained in step 3 to a curve, establish a one-to-one correspondence with the original image noise, remove periodic noise from the original image, and obtain a denoised image;

[0012] Step 5. Use the denoised image as the original image and repeat the above steps until the periodic noise of the image is completely removed.

[0013] The method for establishing a nonlinear fitting model of remote sensor background information in step 1 includes the following steps:

[0014] 1.1) Arrange the background data according to the scan sequence 1, 2, ..., n-1, n, and the corresponding background data is darkDN n (i, j), where i is the scan row number and j is the scan column number, where 1≤i≤400; 1≤j≤7500;

[0015] 1.2) Assume that the data length is x, 1≤x≤j, where j is the scan column number; use the Fourier transform method to transform the data length x and the background data darkDN of each scan strip. n (i, j) is subjected to nonlinear fitting to establish the nonlinear fitting curve f(x) of each scanning line of the scanning strip:

[0016]

[0017] Among them, a0 is the initial amplitude, a k is the amplitude of the real frequency cosine component, b kis the amplitude of the real frequency sinusoidal component, and K is the number of collected signals.

[0018] Among them, according to the cold space data, calculate a i , b i , according to the orthogonality of the trigonometric function system, we can get their calculation formulas as follows:

[0019]

[0020] 1.3) According to step 1.2), a nonlinear fitting curve is established for each scan line in each background data, and a nonlinear fitting model f is formed for each background scan strip. n (x), nonlinear fitting model f for all background data n (x) Combined to form the nonlinear fitting model F of the entire background strip n {i,x}.

[0021] The method for longitudinal rough matching in step 2 includes the following steps:

[0022] 2.1) According to the nonlinear fitting model F of the entire background strip n {i,x} can fit the DN value curve of each specific background information:

[0023] PdarkDN n (i,j)=F n {i,j}

[0024] Among them, i is the scan row number, j is the scan column number, and n is the scan stripe number;

[0025] 2.2) Select the corresponding image data DN of a thermal infrared remote sensor original image n (i,j), calculate its difference with the background information PdarkDN n The difference between (i,j) is:

[0026] ΔDN n (i,j)=DN n (i,j)-PdarkDN n (i,j)

[0027] 2.3) Then ΔDN n (i, j) is transformed by two-dimensional Fourier transform and inverse transform, and its absolute value is taken to obtain the transform function FFT2 n (i,j), take the transformation function FFT2 n (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) n (i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j columnn (i);

[0028] Among them, the two-dimensional Fourier transform is to convert an image into a series of periodic functions for processing. From a physical effect point of view, the Fourier transform converts from the spatial domain to the frequency domain. In other words, the Fourier transform converts the grayscale distribution function of the image into the frequency distribution function of the image. In fact, the spectrum obtained by performing a two-dimensional Fourier transform on the image is the distribution map of the image gradient. The bright spots of varying brightness seen on the Fourier spectrum are the strength of the difference between a certain point and the domain, that is, the size of the gradient (the frequency size of the point).

[0029]

[0030] FFT2 n (i,j)=|F -1 (u,v)|

[0031]

[0032] D n (i,j): grayscale image;

[0033] F(u,v):D n The spectrum of (i,j);

[0034] M:D n The number of rows of (i,j) is 400 for this thermal infrared detector;

[0035] N:D n The number of columns (i, j) corresponding to this thermal infrared detector is 7500;

[0036] FFT2 n (i,j): is the frequency spectrum of F(u,v) after translation.

[0037] 2.4) According to FMAX1 n (i) and FMAX2 n (i) Calculate the matching function stdB(n):

[0038]

[0039] 2.5) The background strip h corresponding to the minimum value of the matching function stdB(n) is the image data DN of the corresponding scanning strip n (i,j) matching background information data:

[0040]

[0041] 2.6) According to the above steps, the background data h(m) matching each strip noise data in each original image can be calculated in sequence, where m=1, 2, ..., n-1, n, n is the scanning strip number.

[0042] The method of transverse precise matching in step 3 specifically includes the following steps:

[0043] 3.1) Select the noise data DN of the original image of a scan strip n (i, j), assuming that the number of columns occupied by periodic noise is T, fit the image DN value curve of its background scanning strip h(m):

[0044] PdarkDN h (i,L+j-1)=F h {i,j+3T}

[0045] Among them, i is the scan row number, j is the scan column number, h is the background scan strip; L is a certain sliding window, L=L1+Lstep, L1 is the initial value of the sliding window, L1=1; Lstep is the sliding step, L is the integer part of 3T / Lstep and L≤3T, T is the fluctuation period of the signal value.

[0046] 3.2) Set the background scan strip h(m) to a certain sliding window L, move the sliding window L according to the step size Lstep, and obtain the noise data DN of the original image n (i, j) and the background data HdarkDN under each step Lstep sliding window L The difference HD of (i,j) h (i,j), that is:

[0047] HD h (i,j)=DN n (i,j)-HdarkDN L (i,j)

[0048] Among them, the background data under each step Lstep sliding window is:

[0049] HdarkDN L (i,j)=PdarkDN h (i,L+j-1)

[0050] 3.3) HD h (i,j) performs a two-dimensional Fourier transform to obtain the transform function FFT2 h (i,j), take the transformation function FFT2 h (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) h(i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j column h (i);

[0051] 3.4) According to FMAX1 h (i) and FMAX2 h (i) Calculate the matching function stdH(h):

[0052]

[0053] 3.5) Take the background scan strip window L corresponding to the minimum value of the matching function stdH(h), which is the noise curve data darkDN that is most consistent with the background data fitting the noise data n :

[0054]

[0055] 3.6) Obtained noise curve data darkDN n The noise data DN of the original image n (i, j) establish a one-to-one correspondence, perform periodic noise removal on the original image, and obtain the denoised image:

[0056]

[0057] in, The noise curve data darkDN is fitted to the background data that is most consistent with the noise data obtained in step 3.5) n Uniform value of .

[0058] Wherein, the step 5. specifically includes the following:

[0059] 5.1) The denoised image newDN n Perform a two-dimensional Fourier transform to get the mean:

[0060]

[0061] And the maximum value:

[0062] W max =max(FFT2 new (i,j)

[0063] 5.2) Determine when FMAX1 h (i) and FMAX2 h When the value of (i) is greater than 1.1×W, or FMAX1 h (i) and FMAX2 h The value of (i) is equal to W max When , it indicates that there is still residual periodic noise in the data;

[0064] 5.3) Take the denoised image as the original image and repeat steps 1.-4. until FMAX1 is satisfied. h (i) and FMAX2 h When the values ​​of (i) are all less than 1.1×W, the periodic noise of the image is completely removed.

[0065] The present invention adopts the above technical scheme, and aims at the inconsistent initial phase of the noise of each scanning line of the thermal infrared camera, which presents the phenomenon of inconsistency between the first scanning line and the second scanning line on the image, collects the background data output by the thermal infrared remote sensor detector under background radiation conditions, and establishes a noise curve model by analyzing the characteristics of the noise. By means of longitudinal coarse matching and lateral fine matching, the initial phase is adaptively matched to obtain the background scanning strip fitting curve, and the strip noise is adaptively removed. While eliminating the strip noise of the thermal infrared image, this method can retain the original detail information of the image to the maximum extent, and will not introduce new noise, thereby effectively improving the image quality. In addition, the algorithm contained in the method is simple, the processing speed is fast, the degree of automation is high, and manual intervention is less, so it can realize batch processing of massive thermal infrared image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a schematic diagram of the swing scanning imaging method of the thermal infrared remote sensor in the prior art.

[0067] Figure 2 The figure is a flow chart of a method for adaptively denoising thermal infrared remote sensing images based on background information according to the present invention.

[0068] Figure 3 This is the imaging data of water bodies and land areas by the multi-spectral infrared camera imager carried by the XX satellite selected in the specific implementation manner of the present invention.

[0069] Figure 4 This is the noise impact of the multispectral infrared camera imager carried by the XX satellite selected in the specific implementation manner of the present invention on the imaging data of water bodies and land areas.

[0070] Figure 5 It is a schematic diagram of the denoising effect of the thermal infrared remote sensing image adaptive denoising method based on background information of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, features and advantages of the present invention clearer, a specific implementation of the present invention is described in more detail. In the following description, many specific details are elaborated to facilitate a full understanding of the present invention, but the present invention can be implemented in many other ways different from the description. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0072] like Figure 2 As shown, the present invention proposes a thermal infrared remote sensing image adaptive denoising method based on background information, the method comprising the following steps:

[0073] Step 1. Perform nonlinear fitting on the background data output by the thermal infrared remote sensor detector under background radiation conditions according to the scanning strip sequence to establish a nonlinear fitting model of the remote sensor background information, which specifically includes the following:

[0074] 1.1) Arrange the background data according to the scan strip number 1, 2, ..., n-1, n, and the corresponding background data is darkDN n (i, j), where i is the scan row number and j is the scan column number, where 1≤i≤400; 1≤j≤7500;

[0075] 1.2) Assume that the data length is x, 1≤x≤j, where j is the scan column number; use the Fourier transform method to transform the data length x and the background data darkDN n (i, j) performs nonlinear fitting and establishes the nonlinear fitting curve f(x) for each scan line:

[0076]

[0077] Where a0 is the initial amplitude, a k is the amplitude of the real frequency cosine component, b k is the amplitude of the real frequency sinusoidal component, and K is the number of collected signals.

[0078] Among them, according to the cold space data, calculate a i , b i , according to the orthogonality of the trigonometric function system, we can get their calculation formulas as follows:

[0079]

[0080] 1.3) According to step 1.2), a nonlinear fitting curve is established for each scan line in each background data, and a nonlinear fitting model f is formed for each background scan strip. n (x), nonlinear fitting model f for all background data n (x) is combined to form the nonlinear fitting model F of the entire background data n {i,x}.

[0081] Step 2. Perform a longitudinal rough match between the noise data of the original image of the thermal infrared remote sensor and the nonlinear fitting model of the background information, and select the background information scanning strip that matches the noise data from the remote sensor background information. The specific method is as follows:

[0082] 2.1) According to the nonlinear fitting model F of the entire background data n {i,x} can fit the DN value curve of each specific scanning strip background information:

[0083] PdarkDN n (i,j)=F n {i,j}

[0084] Among them, i is the scan row number, j is the scan column number, and n is the scan stripe number;

[0085] 2.2) Select the image data DN of the corresponding scanning strip of the original image of the thermal infrared remote sensor n (i, j), calculate its background information PdarkDN with the scanned strip n The difference between (i,j) is:

[0086] ΔDN n (i,j)=DN n (i,j)-PdarkDN n (i,j)

[0087] 2.3) Then ΔDN n (i, j) is transformed by two-dimensional Fourier transform and inverse transform, and its absolute value is taken to obtain the transform function FFT2 n (i,j), take the transformation function FFT2 n (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) n (i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j column n (i);

[0088] Among them, the two-dimensional Fourier transform is to convert an image into a series of periodic functions for processing. From a physical effect point of view, the Fourier transform converts from the spatial domain to the frequency domain. In other words, the Fourier transform converts the grayscale distribution function of the image into the frequency distribution function of the image. In fact, the spectrum obtained by performing a two-dimensional Fourier transform on the image is the distribution map of the image gradient. The bright spots of varying brightness seen on the Fourier spectrum are the strength of the difference between a certain point and the domain, that is, the size of the gradient (the frequency size of the point).

[0089]

[0090] FFT2 n (i,j)=|F -1 (u,v)|

[0091]

[0092] D n(i,j): grayscale image;

[0093] F(u,v):D n The spectrum of (i,j);

[0094] M:D n The number of rows of (i,j) is 400 for this thermal infrared detector;

[0095] N:D n The number of columns (i, j) corresponding to this thermal infrared detector is 7500;

[0096] FFT2 n (i,j): is the frequency spectrum of F(u,v) after translation.

[0097] 2.4) According to FMAX1 n (i) and FMAX2 n (i) Calculate the matching function stdB(n):

[0098]

[0099] 2.5) The background scanning strip h corresponding to the minimum value of the matching function stdB(n) is the image data DN of the corresponding scanning strip n (i,j) matching background information data:

[0100]

[0101] 2.6) According to the above steps, the background data h(m) matching the noise data of each scanning strip in each original image can be calculated in sequence, where m=1, 2, ..., n-1, n, n is the scanning strip number.

[0102] Step 3. Determine the starting position of the noise data stripe period frame by frame, and perform horizontal precise matching with the matching background information scanning strip obtained in step 2, match the background data position consistent with the starting phase of the noise data, and obtain the background data fitting curve consistent with the noise data, which specifically includes the following:

[0103] Step 4. Fit the background data obtained in step 3 to a curve, establish a one-to-one correspondence with the original image noise, remove periodic noise from the original image, and obtain a denoised image, which specifically includes the following:

[0104] 3.1) Select a noise data DN to scan the original image n (i, j), assuming that the number of columns occupied by periodic noise is T, fit the image DN value curve of its background scanning strip h(m):

[0105] PdarkDNh (i,L+j-1)=F h {i,j+3T}

[0106] Among them, i is the scan row number, j is the scan column number, h is the background scan strip; L is a certain sliding window, L=L1+Lstep, L1 is the initial value of the sliding window, L1=1; Lstep is the sliding step, L is the integer part of 3T / Lstep and L≤3T, T is the fluctuation period of the signal value.

[0107] 3.2) Set the background data h(m) to a certain sliding window L, move the sliding window L according to the step size Lstep, and obtain the noise data DN of the original image n (i, j) and the background data HdarkDN under each step Lstep sliding window L The difference HD of (i,j) h (i,j), that is:

[0108] HD h (i,j)=DN n (i,j)-HdarkDN L (i,j)

[0109] Among them, the background data under each step Lstep sliding window is:

[0110] HdarkDN L (i,j)=PdarkDN h (i,L+j-1)

[0111] 3.3) HD h (i,j) performs a two-dimensional Fourier transform to obtain the transform function FFT2 h (i,j), take the transformation function FFT2 h (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) h (i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j column h (i);

[0112] 3.4) According to FMAX1 h (i) and FMAX2 h (i) Calculate the matching function stdH(h):

[0113]

[0114] 3.5) The background data window L corresponding to the minimum value of the matching function stdH(h) is the noise curve data darkDN that is fitted with the background data that is most consistent with the noise data. n:

[0115]

[0116] 3.6) Obtained noise curve data darkDN n The noise data DN of the original image n (i, j) establish a one-to-one correspondence, perform periodic noise removal on the original image, and obtain the denoised image:

[0117]

[0118] in, The noise curve data darkDN is fitted to the background data that is most consistent with the noise data obtained in step 3.5) n Uniform value of .

[0119] Step 5. Use the denoised image as the original image and repeat the above steps until the periodic noise of the image is completely removed, which includes the following:

[0120] 5.1) The denoised image newDN n Perform a two-dimensional Fourier transform to get the mean:

[0121]

[0122] And the maximum value:

[0123] W max =max(FFT2 new (i,j)

[0124] 5.2) Determine when FMAX1 h (i) and FMAX2 h When the value of (i) is greater than 1.1×W, or FMAX1 h (i) and FMAX2 h The value of (i) is equal to W max When , it indicates that there is still residual periodic noise in the data;

[0125] 5.3) Take the denoised image as the original image and repeat steps 1.-4. until FMAX1 is satisfied. h (i) and FMAX2 h When the values ​​of (i) are all less than 1.1×W, the periodic noise of the image is completely removed.

[0126] The multi-spectral infrared camera carried by the XX satellite has a width of 720km and a spatial resolution of 48 / 96m, and can simultaneously obtain spatial, radiation and spectral information of surface objects in 9 spectral bands within the range of 630-125000nm. Figure 1 As shown in the figure, its long-wave linear array is 400 yuan. When scanning the ground, the angle when the scanning mirror points to the subsatellite point is defined as 45°. The ground is scanned slowly and quickly within the range of 45°±15°. Periodic noise appears in the uniform scanning direction. The imaging data of water and land areas on October 20, 2020 are selected as follows Figure 3 As shown, the noise affects the data as follows Figure 4 As shown in the figure, it can be seen that the starting phases of the noises presented on multiple scanning strips of the image are inconsistent, and the noises of each scanning strip are periodic, but the periodicity is not completely fixed. The above method of the present invention is verified. The noise removal effect is shown in the figure. Figure 5 As shown, it can be seen that this method effectively maintains the original sensitivity characteristics of the image. While eliminating the stripe noise of the thermal infrared image, it retains the original detail information of the image to the maximum extent without introducing new noise. It can effectively improve the image quality and lay a good foundation for subsequent image analysis and application.

[0127] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for adaptive denoising of thermal infrared remote sensing images based on background information, characterized in that: The method comprises the following steps: Step 1. Perform nonlinear fitting on the background data output by the thermal infrared remote sensor detector under background radiation conditions according to the scanning order to establish a nonlinear fitting model of the remote sensor background information; Step 2. Perform longitudinal rough matching between the noise data of the original image of the thermal infrared remote sensor and the nonlinear fitting model of the background information, and select the background information data matching the noise data from the remote sensor background information; Step 3. Determine the starting position of the noise data stripe period frame by frame, and perform horizontal precise matching with the matching background information data obtained in step 2. Match the background data position consistent with the starting phase of the noise data, and obtain a background fitting curve consistent with the noise data; Step 4. Fit the background data obtained in step 3 to a curve, establish a one-to-one correspondence with the original image noise, remove periodic noise from the original image, and obtain a denoised image; Step 5. Use the denoised image as the original image and repeat the above steps until the periodic noise of the image is completely removed.

2. The method for adaptive denoising of thermal infrared remote sensing images based on background information according to claim 1, characterized in that: The method for establishing a nonlinear fitting model of remote sensor background information in step 1. comprises the following steps: 1).1 Arrange the background data according to the scan sequence 1, 2, ..., n-1, n, and the corresponding background data is darkDN n (i, j), where i is the scan row number and j is the scan column number, where 1≤i≤400; 1≤j≤7500; 1).2 Assume that the data length is x, 1≤x≤j, where j is the scan column number; use the Fourier transform method to transform the data length x and the background data darkDN of each scan strip n (i, j) performs nonlinear fitting and establishes the nonlinear fitting curve f(x) for each scan line: Where a0 is the initial amplitude, a k is the amplitude of the real frequency cosine component, b k is the amplitude of the real frequency sinusoidal component, K is the number of collected signals; 1).3 According to step 1.2), a nonlinear fitting curve is established for each scan line in each background data, and a nonlinear fitting model f is formed for each background data. n (x), the nonlinear fitting model fn(x) of all background data is combined to form the nonlinear fitting model F of the entire background scan strip n {i,x}.

3. The method for adaptive denoising of thermal infrared remote sensing images based on background information according to claim 1, characterized in that: The method for longitudinal rough matching in step 2. comprises the following steps: 2).1 According to the nonlinear fitting model F of the entire background data n {i,x} can fit the DN value curve of each specific background information: PdarkDN n (i,j)=F n {i,j} Among them, i is the scan row number, j is the scan column number, and n is the scan stripe number; 2) Select the corresponding image data DN of the original image of the thermal infrared remote sensor n (i,j), calculate its difference with the background information PdarkDN n The difference between (i,j) is: ΔDN n (i,j)=DN n (i,j)-PdarkDN n (i,j) 2).3 Then ΔDN n (i, j) is transformed by two-dimensional Fourier transform and inverse transform, and its absolute value is taken to obtain the transform function FFT2 n (i,j), take the transformation function FFT2 n (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) n (i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j column n (i); 2).4According to FMAX1 n (i) and FMAX2 n (i) Calculate the matching function stdB(n): 2).5 Take the background scanning strip h corresponding to the minimum value of the matching function stdB(n), which is the image data DN of the corresponding scanning strip n (i,j) matches the background information scan strip: 2).6 According to the above steps, the background scanning strip h(m) that matches the noise data of each scanning strip in each original image can be calculated in sequence, where m=1, 2,..., n-1, n, n is the scanning strip sequence number.

4. The method for adaptive denoising of thermal infrared remote sensing images based on background information according to claim 1, characterized in that: The method for lateral precise matching in step 3 specifically comprises the following steps: 3).1 Select the noise data DN of the original image of a scan strip n (i, j), assuming that the number of columns occupied by periodic noise is T, fit the image DN value curve of its background scanning strip h(m): PdarkDN h (i,L+j-1)=F h {i,j+3T} Where i is the scan row number in the scan strip, j is the scan column number, and h is the background scan strip; L is a certain sliding window, L = L1 + Lstep, L1 is the initial value of the sliding window, L1 = 1; Lstep is the sliding step, L is the integer part of 3T / Lstep and L≤3T, and T is the fluctuation period of the signal value; 3).2 Set the background scan strip h(m) to a certain sliding window L, move the sliding window L according to the step size Lstep, and obtain the noise data DN of the original image n (i, j) and the background scan strip data HdarkDN under each step Lstep sliding window L The difference HD of (i,j) h (i,j), that is: HD h (i,j)=DN n (i,j)-HdarkDN L (i,j) Among them, the background scan strip data under each step Lstep sliding window is: HdarkDN L (i,j)=PdarkDN h (i,L+j-1) 3).3 pairs of HD h (i,j) performs a two-dimensional Fourier transform to obtain the transform function FFT2 h (i,j), take the transformation function FFT2 h (i,j) is the maximum value FMAX1 between the scan columns 1 to (1 / 2*j) h (i) and the maximum value FMAX2 between the scan column number (1 / 2*j+1) to the j column h (i); 3).4According to FMAX1 h (i) and FMAX2 h (i) Calculate the matching function stdH(h): 3).5 Take the background scanning window L corresponding to the minimum value of the matching function stdH(h), which is the noise curve data darkDN of the background fitting that is most consistent with the noise data n : 3).6 Obtain the noise curve data darkDN n The noise data DN of the original image n (i, j) establish a one-to-one correspondence, perform periodic noise removal on the original image, and obtain the denoised image: in, The noise curve data darkDN obtained in step 3.5) is the background fitting data that is most consistent with the noise data n Uniform value of .

5. The method for adaptive denoising of thermal infrared remote sensing images based on background information according to claim 1 or 4, characterized in that: The step 5 specifically includes the following: 5).1 The denoised image newDN n Perform a two-dimensional Fourier transform to get the mean: And the maximum value: W max =max(FFT2 new (i,j) 5).2 Determine when FMAX1 h (i) and FMAX2 h When the value of (i) is greater than 1.1×W, or FMAX1 h (i) and FMAX2 h The value of (i) is equal to W max When , it indicates that there is still residual periodic noise in the data; 5).3 Take the denoised image as the original image and repeat steps 1.-4. until FMAX1 is satisfied. h (i) and FMAX2 h When the values ​​of (i) are all less than 1.1×W, the periodic noise of the image is completely removed.

Images