A hyperspectral image defogging method

Abstract

The present invention relates to the field of image processing technology, and in particular to a method for defogging a hyperspectral image, comprising: collecting a hyperspectral image and decomposing it into multiple single-band images; segmenting any single-band image to obtain a single-band low-frequency image, including: converting the single-band image from the spatial domain to the frequency domain to obtain a single-band spectrum diagram; obtaining two annular regions in the spectrum diagram, calculating the energy gradient between the annular regions, and determining a segmentation threshold based on the energy gradient; segmenting the single-band image based on the segmentation threshold to obtain a single-band low-frequency image; processing the single-band low-frequency image using an atmospheric scattering model to obtain a single-band defogging image; and combining multiple single-band defogging images to obtain a defogging hyperspectral image. The present invention can fully utilize the spectral information of the image through a segmentation method based on spectral characteristics, remove high-frequency noise and details, retain effective information, and improve the defogging effect.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a hyperspectral image defogging method. Background Art

[0002] Hyperspectral images are image data acquired by an imaging spectrometer across dozens or even hundreds of consecutive narrow wavelengths. Each pixel contains not only spatial information but also its reflectivity or radiation intensity at different wavelengths, forming a complete spectral curve.

[0003] Fog in hyperspectral images typically refers to image degradation caused by atmospheric scattering, such as mist and haze, manifesting as decreased contrast and spectral distortion. Hyperspectral image defogging methods are commonly used in fields such as satellite remote sensing, environmental monitoring, military reconnaissance, agricultural monitoring, and disaster warning. However, hyperspectral images have numerous bands and large data volumes; the noise in each band is complex, and the spatial and spectral distribution of fog is uneven, making defogging extremely difficult. Consequently, practical applications of hyperspectral image defogging technology suffer from low defogging accuracy, poor spectral fidelity, and insufficient adaptability to complex scenes.

[0004] In the existing technology, in order to effectively remove the interference of haze in hyperspectral images, traditional methods driven by physical models and hybrid pixel decomposition technology are usually adopted. Physical model-driven methods have poor adaptability in complex scenes and have difficulty in accurately separating the spectral information of the fog layer and the ground objects. Hybrid pixel decomposition technology is based on a spectral mixing model and can directly calculate the endmember abundance from the mixed spectrum of haze interference, achieving mathematical separation of the fog layer and the ground object spectrum, and avoiding the strong dependence of traditional methods on physical parameters. However, hybrid pixel decomposition technology usually processes directly in the spatial domain and is easily interfered by high-frequency noise. Therefore, it will lead to limited endmember extraction accuracy, which further leads to poor hyperspectral image defogging effect.

[0005] It can be seen that the hyperspectral image defogging method in the prior art still has the problem of poor defogging effect. Summary of the Invention

[0006] In order to solve the technical problem that the above-mentioned hyperspectral image defogging method has poor defogging effect, the present invention provides a method, which includes: collecting a hyperspectral image and decomposing it into multiple single-band images; segmenting any of the single-band images to obtain the single-band low-frequency image, including: converting the single-band image from the spatial domain to the frequency domain to obtain the single-band spectrum diagram; the spectrum diagram displays the amplitude spectrum of the spectrum; centering the spectrum diagram to obtain the center of the spectrum diagram, obtaining at least two annular areas in the spectrum diagram, calculating the energy gradient between the two annular areas, and determining a segmentation threshold according to the energy gradient; segmenting the single-band image based on the segmentation threshold to obtain the single-band low-frequency image; the center point of the annular area is the center of the spectrum diagram; using an atmospheric scattering model to process the single-band low-frequency image to obtain the single-band defogging image; combining the multiple single-band defogging images to obtain a defogged hyperspectral image.

[0007] By decomposing the hyperspectral image into single-band images, processing each single-band image, and using the atmospheric scattering model to process the single-band low-frequency image, the impact of fog on the image can be removed more accurately. In the process of segmenting the single-band image to obtain the low-frequency image, the image is first converted from the spatial domain to the frequency domain to obtain the spectral amplitude spectrum, and then centered. By calculating the energy gradient between the annular regions to determine the segmentation threshold, this spectral characteristic-based segmentation method can fully utilize the spectral information of the image. The low-frequency image contains the main structure and contour information of the image, removes high-frequency noise and details, retains more valuable information for dehazing, and helps to perform dehazing operations more accurately while reducing unnecessary calculations.

[0008] As a further improvement of the method of the present invention, the energy gradient between the two annular regions is calculated, comprising: calculating the energy of the two annular regions respectively; calculating the energy gradient between the two annular regions ;in, The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, is the distance between the inner circle and the outer circle of the annular region, is the step size when calculating the energy gradient.

[0009] By calculating the energy of the two annular regions separately and then calculating the energy gradient between them, we can accurately reflect the energy variations between different regions in the spectrum. Accurate energy gradient calculation provides a reliable basis for determining the segmentation threshold. When segmenting a single-band image to obtain a low-frequency image, properly determining the segmentation threshold is crucial. Using the energy gradient to determine the threshold ensures that the segmentation result is more consistent with the image's spectral characteristics.

[0010] As a further improvement of the method of the present invention, the energy of the annular region satisfies the relationship: ;in, The frequency points in the spectrum are in polar coordinates. The complex value of the spectrum of The frequency points in the spectrum are in polar coordinates. The amplitude spectrum value of is the polar angle of the frequency point in the spectrum graph in the polar coordinate system, It is a modulo operation on the complex number of the spectrum.

[0011] Calculating the energy of an annular region based on polar coordinates and amplitude spectrum values ​​can more effectively reflect the characteristics of the spectrum. The polar coordinate system can clearly locate the position of the frequency point in the spectrum graph, and combined with the amplitude spectrum values, it can comprehensively describe the spatial distribution and energy distribution of the spectrum. In the spectrum analysis of hyperspectral images, different ground objects or image features may exhibit specific energy distribution patterns in different regions of the spectrum. This precise energy calculation method can better identify and analyze these features, for example, distinguishing the spectral energy characteristics corresponding to different structures such as edges and textures in the image.

[0012] As a further improvement of the method of the present invention, determining the segmentation threshold according to the energy gradient includes: obtaining the outer circle radius of the annular area that maximizes the energy gradient, and the segmentation threshold is the outer circle radius.

[0013] As a further improvement to the method of the present invention, the segmentation of the single-band image based on the segmentation threshold to obtain the single-band low-frequency image includes: marking the frequency components in the spectrum graph whose radius is smaller than the segmentation threshold as low-frequency components, and marking the remaining frequency components as high-frequency components; setting the complex values ​​of the frequency points of the high-frequency components to zero to obtain a low-frequency spectrum; and converting the low-frequency spectrum from the frequency domain to the spatial domain to obtain the single-band low-frequency image.

[0014] As another improvement to the method of the present invention, the atmospheric scattering model includes transmittance calculation, and the transmittance calculation includes: performing mixed pixel decomposition on the low-frequency image to obtain abundance coefficients corresponding to pixels in the low-frequency image; establishing a linear mapping relationship between the abundance coefficient of the pixel and the initial transmittance value, and calculating the initial transmittance value of the pixel based on the mapping relationship; processing the initial transmittance value based on a dark channel prior method to obtain the target transmittance.

[0015] As another improvement of the method of the present invention, the mixed pixel decomposition of the low-frequency image includes: extracting the endmembers of the single-band low-frequency image to obtain fog layer endmembers and spectral endmembers, and constructing a fog layer endmember spectral matrix and ground object endmember spectral matrix For each pixel in the low-frequency image, a mist mixed pixel model is applied, i.e. ;in, is the pixel in the low-frequency image The spectral value of It is a pixel The value of the end-member spectral matrix in the fog layer, It is a pixel The abundance coefficient It is a pixel The objective function is defined, and the constraint conditions are determined. When the constraint conditions are met, the objective function is solved to obtain the abundance coefficient.

[0016] The haze-mixed pixel model considers the different contributions of fog layer endmembers and ground object endmembers to the pixel spectral values. By defining an objective function and constraints, and using an optimization algorithm to solve the abundance coefficient, the relative proportions of fog and ground objects in each pixel can be more accurately calculated. This makes the recovered ground object spectral values ​​closer to reality, improves the accuracy of the spectral information, and provides a more reliable data foundation for subsequent spectral-based analyses, such as material composition identification and ground object classification.

[0017] As another improvement of the method of the present invention, the objective function is ; in, is to minimize the error, are the coordinates of the pixels in the low-frequency image, are all the pixels in the low-frequency image.

[0018] As another improvement of the method of the present invention, the establishment of the linear mapping relationship between the abundance coefficient of the pixel and the initial transmittance value includes: the linear mapping relationship is ,in, It is a pixel The initial transmittance value, is the preset adjustment factor, It is a pixel abundance coefficient.

[0019] As another improvement to the method of the present invention, the combining of the multiple single-band defogging images to obtain the defogging hyperspectral image includes: stacking the multiple single-band defogging images in the order of their corresponding spectral wavelengths to form a defogging hyperspectral image, wherein each pixel corresponds to the same spatial position in all single bands.

[0020] By accurately restoring the hyperspectral data structure, ensuring spatial information consistency and maintaining spectral information continuity, the application value of the dehazed hyperspectral image is ultimately enhanced.

[0021] Beneficial effects of the present invention:

[0022] The present invention decomposes the hyperspectral image into single-band images and uses the atmospheric scattering model to process the single-band low-frequency image to accurately remove fog interference. When acquiring the low-frequency image, the image is converted from the spatial domain to the frequency domain, and the image is centered based on the spectral amplitude spectrum. The segmentation threshold is determined by calculating the energy gradient between the annular regions, making full use of the spectral information, retaining the main structure and contour of the image, and removing high-frequency noise, thereby improving the defogging accuracy and reducing the amount of calculation. At the same time, the energy and energy gradient of the annular region are accurately calculated to provide a reliable basis for determining the segmentation threshold, so that the segmentation result is consistent with the image spectral characteristics, ensuring the efficiency and accuracy of the hyperspectral image defogging process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a hyperspectral image defogging method provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of spectrum centralization provided by an embodiment of the present invention;

[0025] Figure 3 A schematic diagram of dividing annular areas on a spectrum diagram provided by an embodiment of the present invention.

[0026] In the figure, 1, first quadrant; 2, second quadrant; 3, third quadrant; 4, fourth quadrant; 51, center of the spectrum graph; 52, inner circle; 53, outer circle. DETAILED DESCRIPTION

[0027] The embodiment of the present invention provides a hyperspectral image defogging method, such as Figure 1 As shown, the method includes steps S100 to S400:

[0028] Step S100: Acquire a hyperspectral image and decompose it into multiple single-band images.

[0029] To elaborate, there are two main ways to collect and decompose hyperspectral images. One is based on scanning, and the other is based on snapshots.

[0030] The scanning method uses a sensor to scan in the spatial or spectral dimension, acquiring spectral information of the surface in multiple continuous narrow single bands line by line or point by point, ultimately forming three-dimensional hyperspectral data containing all the single bands. The corresponding decomposition method reads the stored hyperspectral data file, usually a three-dimensional array, and obtains each independent single band image by extracting two-dimensional slices of the third dimension of the data cube, the single band dimension.

[0031] The snapshot method uses specially designed sensors, such as filter arrays or computational spectral imaging systems, to simultaneously capture image information for the entire field of view across multiple spectral bands in a single exposure, generating raw data containing multiple bands of information. The snapshot method's corresponding decomposition involves reading the raw snapshot data, typically through a demultiplexing or reconstruction process, to convert it into a standard three-dimensional hyperspectral data cube arranged by band. Each independent band image is then obtained by extracting two-dimensional slices of the third dimension, the band dimension.

[0032] Hyperspectral images typically contain a large amount of single-band data, which increases the difficulty of data processing and analysis. Decomposing them into single-band images can remove redundant information and achieve data dimensionality reduction. This not only reduces data processing time and computing resources, but also improves the efficiency and accuracy of data analysis. For example, in face recognition applications, key single-band features can be extracted from the numerous single-band images in hyperspectral images to build more effective recognition models.

[0033] Step S200: Segment any single-band image to obtain a single-band low-frequency image.

[0034] Specifically, in the process of dehazing hyperspectral images, the noise of each single band is significantly different. For example, the noise of the short-wave infrared single band is high, and the noise of the visible light single band is low. The traditional fixed threshold strategy cannot adapt to this difference, resulting in the loss of high-frequency ground object details or the residual low-frequency fog layer. Specifically, in a single band with high noise (such as short-wave infrared), the fixed threshold will mistakenly retain the high-frequency noise as a valid signal, resulting in image blur; while in a single band with low noise (such as visible light), the fixed threshold over-suppresses the high-frequency component, resulting in the loss of ground object texture. In order to solve this problem, the low-frequency image of the single band is obtained by segmenting the image of each single band. The process includes steps S210-S250:

[0035] Step S210: convert the single-band image from the spatial domain to the frequency domain to obtain a single-band spectrogram; the spectrogram displays the amplitude spectrum of the spectrum.

[0036] To elaborate, for any single-band image, converting it from the spatial domain to the frequency domain yields the image's spectrum. The spectrum is often visualized as a spectrogram to help us understand the image's frequency components. A spectrogram typically displays the amplitude of the spectrum, and sometimes also the phase spectrum. In this disclosure, the amplitude spectrum is specifically displayed.

[0037] In addition, the methods for converting the single-band image from the spatial domain to the frequency domain and having reversibility include Fourier transform, discrete cosine transform, and wavelet transform. In actual use, you can choose any one of these methods, and these methods are all existing technologies and will not be explained in detail here.

[0038] Step S220: performing centralization processing on the spectrum graph to obtain the center of the spectrum graph.

[0039] To elaborate, after converting this single-band image from the spatial domain to the frequency domain, the zero-frequency component in the uncentered spectrogram is located in the upper left corner. To facilitate observation and analysis of the frequency distribution of the spectrum, particularly the low- and high-frequency components, it is often necessary to move the zero-frequency component in the spectrogram to the center.

[0040] like Figure 2 As shown, the spectrum graph consists of four quadrants. The centering operation rearranges the four quadrants of the spectrum, swapping the first quadrant 1 in the upper left corner (containing the zero-frequency component) with the third quadrant 3 in the lower right corner, and swapping the second quadrant 2 in the upper right corner with the fourth quadrant 4 in the lower left corner. After the centering operation, the center of the spectrum graph corresponds to the zero-frequency component, with lower-frequency components closer to the center and higher-frequency components farther away from the center.

[0041] Step S230: Acquire at least two annular regions in the spectrum diagram.

[0042] To expand, after getting the center of the spectrum, take it as the center of the circle and is the inner circle and radius is the outer circle, and the area between the two circles is the annular area. It is the distance between the annular areas, which can be set according to the actual situation and is usually 1.

[0043] like Figure 3 As shown, the background in the figure is a spectrum diagram, wherein the center point of the annular area is the center 51 of the spectrum diagram, circle 52 is the inner circle of the smallest annular area, and circle 53 is the outer circle of the smallest annular area.

[0044] It should be noted that the outer radius and inner radius To be smaller than the maximum effective radius of the image:

[0045] ;

[0046] in, is the maximum effective radius, , are the length and width of the spectrum graph, is the minimum function.

[0047] Step S240: Calculate the energy gradient between the two annular regions, and determine the segmentation threshold according to the energy gradient.

[0048] To expand on this, to calculate the energy gradient between two annular regions, we first need to calculate the energy of each of the two annular regions separately. The energy calculation formula for any annular region is:

[0049] ;

[0050] in, The frequency points in the spectrum are in polar coordinates. The complex value of the spectrum of The frequency points in the spectrum are in polar coordinates. The amplitude spectrum value of is the polar angle of the frequency point in the spectrum graph in the polar coordinate system, It is a modulo operation on the complex number of the spectrum.

[0051] It should be noted that , It is the distance between the inner circle and the outer circle of a ring area.

[0052] In this formula, the energy is calculated by calculating the square sum of the amplitudes of all points in the annular area corresponding to each radius. In the frequency domain, the low-frequency area is characterized by a gentle energy change, while the high-frequency area shows a trend of violent energy fluctuations. For this single band, if it is within the inner radius The higher the energy in the corresponding annular area, the more likely it is that the annular area is in the high-frequency area and contains rich ground feature details or noise. On the contrary, when the energy value of the annular area is low, it is likely that the area is in the fog layer and the low-frequency area corresponding to the illumination, and the signal performance is more stable.

[0053] Secondly, the energy gradient of the two annular regions is calculated based on their energies:

[0054] ;

[0055] in The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, is the distance between the inner circle and the outer circle of the annular region, is the step size when calculating the energy gradient.

[0056] In the actual processing, not only the energy gradient of the two annular regions is calculated, but also Traverse from 0 to the maximum effective radius Inner radius A low calculated energy gradient indicates that the energy difference between the two annular regions is small. A higher energy gradient indicates a significant energy transition in the frequency domain at that radius. Therefore, the location corresponding to the maximum energy gradient reflects the transition from stable to drastic energy changes and can be used as the dividing point between low-frequency (stable region) and high-frequency (noise region).

[0057] It should be noted that when calculating the energy gradient , which can be selected and adjusted according to actual conditions. Figure 3 The given annular area.

[0058] Finally, the outer radius of the annular area corresponding to the maximum energy gradient is taken as the gradient mutation radius in the spectrum of the single band, that is, the segmentation threshold between high frequency and low frequency.

[0059] Specifically, an energy accumulation curve can be drawn, and by analyzing the gradient change of the energy accumulation curve, the mutation point of the frequency domain energy distribution can be located and used as the segmentation threshold.

[0060] Step S250: Segment the single-band image based on the segmentation threshold to obtain a single-band low-frequency image.

[0061] To elaborate, the frequency components in the spectrum graph whose radius is smaller than the segmentation threshold are marked as low-frequency components, and the remaining frequency components are marked as high-frequency components; the spectrum values ​​and phase values ​​of the high-frequency components are set to zero to obtain the low-frequency spectrum; the low-frequency spectrum is converted from the frequency domain to the spatial domain to obtain the low-frequency image of the single band.

[0062] Furthermore, it should be noted that the method used to convert the low-frequency spectrum from the frequency domain to the spatial domain must correspond to the method used to convert the single-band image from the spatial domain to the frequency domain in step S210, that is, it must be its inverse transform. For example, if the Fourier transform was used in step S210, then the inverse Fourier transform should be used here for the next conversion.

[0063] Step S300: Processing a single-band low-frequency image using an atmospheric scattering model to obtain a single-band defogging image.

[0064] To elaborate, the atmospheric scattering model describes the physical phenomena of light scattering and absorption as it propagates through the atmosphere due to its interactions with various atmospheric particles, such as air molecules, aerosols, and water vapor. The core goal of dehazing methods based on the atmospheric scattering model is to estimate the atmospheric light value and transmittance from an observed foggy image, and then use the atmospheric scattering model to invert the image to produce a clear, fog-free image.

[0065] The present invention improves the calculation of transmittance value, including steps S310 to S340:

[0066] Step S310: Decompose the low-frequency image into mixed pixels to obtain the abundance coefficients corresponding to the pixels in the low-frequency image.

[0067] To elaborate, first, the endmember extraction method such as N-FINDR algorithm and minimum volume transformation is used to extract the endmembers of the single-band low-frequency image to obtain the fog layer endmembers and spectral endmembers. The endmember extraction method such as N-FINDR algorithm and minimum volume transformation method are existing technologies and will not be described here. After the endmember extraction is completed, the spectral curves of the extracted fog layer endmembers and ground object endmembers are arranged in rows or columns to construct the fog layer endmember spectral matrix respectively. and ground object endmember spectral matrix .

[0068] Secondly, given the original spectral image of the single band, the spectral matrix of the fog layer endmembers, and the spectral matrix of the ground object endmembers, the objective function is set to minimize the error of the mixed pixel model:

[0069] ;

[0070] in, is to minimize the error, are the coordinates of the pixels in the low-frequency image, are all pixels in the low-frequency image, is the pixel in the low-frequency image The spectral value of It is a pixel The value of the end-member spectral matrix in the fog layer, It is a pixel The abundance coefficient It is a pixel The value of the end-member spectral matrix of the ground object.

[0071] This formula indicates that in this single band, the square of the difference between the spectral value of each pixel in the low-frequency image and the spectral value corresponding to the haze mixed pixel model is calculated, and then the square of the difference of all pixels in the single band is accumulated and summed, and finally the error value between the value obtained by fitting the haze mixed pixel model and the spectral value of the low-frequency image is obtained.

[0072] This error value quantifies the degree of deviation between the predicted value of the mist mixed pixel model and the true value. The smaller the value, the more accurate the mist mixed pixel model is in simulating the actual spectrum. Conversely, the larger the value, the lower the fit between the mist mixed pixel model and the actual scene.

[0073] Next, determine the constraints as ,When the constraints are met, the optimization algorithm is used to solve the objective function and obtain the abundance coefficient.

[0074] Specifically, for this single band, the gradient descent method is used to solve the optimal fog layer abundance coefficient map.

[0075] Step S320: Establish a linear mapping relationship between the abundance coefficient of the pixel and the initial transmittance value, and calculate the initial transmittance value of the pixel based on the mapping relationship:

[0076] ;

[0077] in, It is a pixel The initial transmittance value, It is a pixel The abundance coefficient of It is a preset adjustment factor with an empirical value of 0.8. It dynamically adjusts the weight of the fog layer abundance on the transmittance correction to balance the defogging effect and detail preservation.

[0078] The abundance coefficient of the end member indicates the proportion of different substances in the mixed pixel. The calculation of the initial transmittance by the above formula can assist in analyzing the relationship between the distribution of different substances and image clarity and light propagation characteristics. It plays a role in adjusting the sensitivity of the initial transmittance to the end member abundance coefficient. The value will make the initial transmittance more sensitive to the change of the end-member abundance coefficient, that is, a small change in the end-member abundance coefficient will cause a large change in the initial transmittance; a smaller value will cause a small change in the end-member abundance coefficient. The value makes the change of initial transmittance relatively gentle and less sensitive to the change of end member abundance coefficient. The value is used to optimize the calculation of the initial transmittance to achieve better analysis results.

[0079] Step S330: Generate an initial transmittance map according to the initial transmittance value.

[0080] Step S340: Process the initial transmittance value based on the dark channel prior method to obtain the target transmittance.

[0081] Specifically, the transmittance of the sky or highlights in the initial transmittance map is often underestimated, resulting in over-enhancement or distortion in the dehazed image. Therefore, a dark channel is used to correct the visible light single band, optimizing its transmittance map to better align with the actual physical scene and avoid over-correction.

[0082] The above describes how to calculate the transmittance value in the atmospheric scattering model. Next, we will explain how to use the atmospheric scattering model to invert a clear, fog-free image based on the atmospheric light value and transmittance.

[0083] Specifically, calculate the pixels in the low-frequency image The spectral value after dehazing in this single band:

[0084] ;

[0085] in, is the pixel in the low-frequency image The spectral value after dehazing in a single band, is the pixel in the hyperspectral image The spectral value in this single band, is the atmospheric light value of the low-frequency image, is the maximum function, is the pixel in the low-frequency image In this single-band corrected transmittance value, It is a preset small value, and the experience value is , the purpose is to avoid transmittance When it is close to zero, the denominator is too small.

[0086] By calculating the defogging spectral values ​​of pixels in a low-frequency image in a single band, we can effectively reduce the impact of fog on the image, making object edges and details more distinct. Fog scatters and absorbs light, causing spectral deviations in the image. Calculating the defogging spectral values ​​can correct these deviations, making the pixel spectral values ​​closer to reality.

[0087] The atmospheric light value is calculated as follows: the pixels with brightness in the top 0.1% in the fog layer abundance coefficient map of the single band are selected, and their average brightness is calculated as the atmospheric light value.

[0088] The above operation is performed on all single bands to obtain multiple single-band dehazed images.

[0089] Step S400: combining multiple single-band defogging images to obtain a defogging hyperspectral image.

[0090] To elaborate, after obtaining the dehazed images of each single band, these single band images need to be combined and stacked to form a dehazed hyperspectral image.

[0091] Specifically, the grayscale values ​​of each single band are arranged in order of wavelength to form a 3D data cube. A feature point matching algorithm is used to ensure that the images of each single band are spatially aligned, that is, the position of each pixel in different single bands is consistent.

[0092] After obtaining the dehazed hyperspectral image, a verification step can be added to use the spectral angle matching method to verify whether the dehazed image is distorted.

[0093] Specifically, a reference spectrum of the target object is obtained from a spectral library or field measurements. For each pixel in the dehazed hyperspectral image, the angle between it and the reference spectrum, i.e., the spectral angle, is calculated.

[0094] The spectral angle threshold is set according to the actual situation, and the empirical value is 0.2 radians. For each pixel in the dehazed hyperspectral image, if the angle between it and the reference spectrum is less than the threshold, the pixel is judged to be true, otherwise the pixel is considered distorted.

[0095] The spectral angles of all pixels in the dehazed hyperspectral image are counted to determine the ratio of distorted pixels to the total pixels. A threshold is set at 7%. If the ratio of distorted pixels to the total pixels exceeds the threshold, the dehazed hyperspectral image is considered distorted. The threshold can be set based on actual needs and empirical values.

[0096] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A hyperspectral image defogging method, characterized in that: include: Collect hyperspectral images and decompose them into multiple single-band images; Segmenting any of the single-band images to obtain the single-band low-frequency image includes: converting the single-band image from a spatial domain to a frequency domain to obtain a frequency spectrum of the single band; the frequency spectrum displays an amplitude spectrum of the frequency spectrum; performing centering processing on the frequency spectrum to obtain a center of the frequency spectrum; obtaining at least two annular regions in the frequency spectrum, calculating an energy gradient between the two annular regions, and determining a segmentation threshold based on the energy gradient; segmenting the single-band image based on the segmentation threshold to obtain the single-band low-frequency image; the center point of the annular region is the center of the frequency spectrum; Processing the single-band low-frequency image using an atmospheric scattering model to obtain the single-band defogging image; Combining the multiple single-band defogging images to obtain a defogging hyperspectral image; Calculating the energy gradient between the two annular regions includes: Calculating the energies of the two annular regions respectively; Calculate the energy gradient between the two annular regions ; in, The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, The radius of the inner circle in the spectrum is , the outer radius is The energy of the annular region, is the distance between the inner circle and the outer circle of the annular region, is the step size when calculating the energy gradient.

2. The hyperspectral image defogging method according to claim 1, characterized in that: The energy of the annular region satisfies the relationship: ; in, The frequency points in the spectrum are in polar coordinates. The complex value of the spectrum of The frequency points in the spectrum are in polar coordinates. The amplitude spectrum value of is the polar angle of the frequency point in the spectrum graph in the polar coordinate system, It is a modulo operation on the complex number of the spectrum.

3. The hyperspectral image defogging method according to claim 1, characterized in that: Determining the segmentation threshold according to the energy gradient includes: The outer circle radius of the annular region that maximizes the energy gradient is obtained, and the segmentation threshold is the outer circle radius.

4. The hyperspectral image defogging method according to claim 1, characterized in that: Segmenting the single-band image based on the segmentation threshold to obtain the single-band low-frequency image includes: Mark the frequency components in the spectrum graph whose radius is smaller than the segmentation threshold as low-frequency components, and mark the remaining frequency components as high-frequency components; Setting the complex value of the frequency point of the high-frequency component to zero to obtain a low-frequency spectrum; The low-frequency spectrum is converted from the frequency domain to the spatial domain to obtain the single-band low-frequency image.

5. The hyperspectral image defogging method according to claim 1, characterized in that: The atmospheric scattering model includes calculation of transmittance, and the calculation of transmittance includes: Performing mixed pixel decomposition on the low-frequency image to obtain abundance coefficients corresponding to pixels in the low-frequency image; Establishing a linear mapping relationship between the abundance coefficient of the pixel and the initial transmittance value, and calculating the initial transmittance value of the pixel based on the mapping relationship; The initial transmittance value is processed based on a dark channel prior method to obtain a target transmittance.

6. The hyperspectral image defogging method according to claim 5, characterized in that: The mixed pixel decomposition of the low-frequency image includes: extracting endmembers from the single-band low-frequency image to obtain fog layer endmembers and spectral endmembers, and constructing a fog layer endmember spectral matrix. and ground object endmember spectral matrix ; For each pixel in the low-frequency image, a mist mixed pixel model is applied, namely: ;in, is the pixel in the low-frequency image calculated by the mist mixed pixel model The spectral value of It is a pixel The value of the end-member spectral matrix in the fog layer, It is a pixel The abundance coefficient of It is a pixel The objective function is defined, and the constraint conditions are determined. When the constraint conditions are met, the objective function is solved to obtain the abundance coefficient.

7. The hyperspectral image defogging method according to claim 6, characterized in that: The objective function is ; in, is to minimize the error, are the coordinates of the pixels in the low-frequency image, are all pixels in the low-frequency image; the formula indicates that in this single band, the square of the difference between the spectral value of each pixel in the low-frequency image and the spectral value corresponding to the haze mixed pixel model is calculated, and then the square of the difference of all pixels in the single band is accumulated and summed, and finally the error value between the value obtained by fitting the haze mixed pixel model and the spectral value of the low-frequency image is obtained.

8. The hyperspectral image defogging method according to claim 5, characterized in that: The establishment of the linear mapping relationship between the abundance coefficient of the pixel and the initial transmittance value includes: the linear mapping relationship is ;in, It is a pixel The initial transmittance value, is the preset adjustment factor, It is a pixel abundance coefficient.

9. The hyperspectral image defogging method according to claim 1, characterized in that: Combining the multiple single-band defogging images to obtain a defogging hyperspectral image includes: The multiple single-band defogging images are stacked in order of their corresponding spectral wavelengths to form a defogging hyperspectral image, in which each pixel corresponds to the same spatial position in all single bands.

Images