An underwater target spectral reflectance reconstruction method based on difference calculation

Abstract

The application discloses a kind of underwater target spectral reflectivity reconstruction methods based on difference calculation, solve the technical problem that present underwater object spectral reflectivity acquisition method exists measurement inaccuracy or only applicable to close-up shooting, cannot be used for the calculation of long-distance object spectral reflectivity;The method of the application, without additional ranging equipment, only using a polarized spectral imager is moved to obtain two groups of images similar to binocular imaging, the distance, angle information of light source, imager and target point are calculated by binocular measurement algorithm and stereoscopic geometry model, combined with the water attenuation coefficient measured in situ by attenuator and Lambert-Beer law, the original incident and emission radiance value of target object can be calculated, so that the real spectral reflectivity of target is reconstructed.

Description

Technical Field

[0001] This invention relates to underwater spectral reflectance measurement methods, specifically to a method for reconstructing the spectral reflectance of an underwater target based on differential calculation. Background Technology

[0002] With the development of marine technology, spectral imagers now possess the capability to observe targets in deep waters. Many researchers are using manned or unmanned underwater vehicles equipped with spectral imagers to collect large-scale underwater spectral information. The measured spectral reflectance data can be applied to underwater ecological environment surveys, monitoring the growth and health of seabed flora and fauna, or assessing the distribution and abundance of underwater mineral or fishery resources. Therefore, obtaining effective and accurate spectral reflectance data of underwater targets can greatly facilitate exploration in other marine research fields.

[0003] However, there is limited research on underwater spectral reflectance measurement. This is because light propagating in air is less affected by atmospheric absorption and scattering, resulting in low attenuation. Furthermore, the light source characteristics are stable, and spectral reflectance can be obtained simply by measuring the emitted radiance of a standard reflectance plate and the emitted radiance of the object using a radiance meter, and then calculating the ratio between the two. However, underwater imaging is complex. Spectral imagers are affected by ambient sunlight, active light sources, scattered light emitted from suspended particles in the water, and attenuation of light emitted by the target object due to scattering and absorption during propagation. Additionally, water absorbs different wavelengths of light differently, leading to a significant difference between the spectral reflectance measured underwater and in air. Therefore, methods for measuring reflectance in air cannot be directly applied underwater.

[0004] Current research primarily employs two methods to obtain the spectral reflectance of underwater objects. One method, used by the Norwegian University of Science and Technology, involves placing a calibration plate with known reflectance within the image captured by a spectral imager. The radiance value of each pixel in the image is then compared to the emitted radiance value of the calibration plate, similar to the method used to measure the spectral reflectance of objects in air. However, this method has several drawbacks: the light field distribution in water is uneven, resulting in varying incident light intensities for each pixel, and the distances between each pixel, the calibration plate, and the imager also differ, leading to inaccurate spectral reflectance measurements. The other method involves establishing an underwater imaging model and using measuring instruments and algorithms to calculate and estimate the various components within the model, ultimately obtaining the object's true spectral reflectance. Zhejiang University, using a binocular spectral imager, an active light source, and a water attenuator, measures the light field distribution of the light source, the distance between the target point and the imager and light source, and the water attenuation coefficient in the calculation model to obtain the spectral reflectance of underwater objects. However, this method ignores the influence of scattered light from suspended particles on the image captured by the spectral imager, making it suitable only for close-range imaging and unsuitable for calculating the spectral reflectance of distant objects. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problem that existing methods for obtaining the spectral reflectance of underwater objects are either inaccurate in measurement or only applicable to close-range shooting and cannot be used for calculating the spectral reflectance of distant objects. In this invention, a method for reconstructing the spectral reflectance of underwater targets based on differential calculation is provided.

[0006] The concept of this invention is as follows: An underwater target imager is captured with the light source both on and off, and the influence of ambient solar light is removed using differential imaging. Then, the polarization information and Stokes vector differential imaging are used to remove the influence of backscattered light to obtain the target information light. By moving the polarization spectral imager and the light source to capture adjacent frames, calculations are performed to obtain the distance and angle information between the target point and the spectrometer and light source in the target scene. Finally, the true spectral reflectance of the underwater target is reconstructed using the known target information light, distance and angle information, light source irradiance information, and water attenuation coefficient.

[0007] In order to achieve the above-mentioned objectives and complete the above-mentioned inventive concept, the present invention adopts the following technical solution:

[0008] A method for reconstructing the spectral reflectance of underwater targets based on differential calculation is characterized by the following steps:

[0009] S1. Perform radiometric calibration on the polarization spectral imager to obtain the radiometric calibration coefficient C(λ), and use Zhang Zhengyou calibration method to calculate the intrinsic parameter matrix and distortion coefficient of the polarization spectral imager. At the same time, use a standard spectroradiometer to measure the irradiance E0(λ) at a distance of 1 meter from the light source.

[0010] S2. Place the polarization spectral imager and the light source at the same underwater altitude, with a distance of l between them. camera-light And in the same spatial plane, turn on the light source to illuminate the target scene, and use the water body attenuation coefficient measuring instrument to measure the current water body attenuation coefficient α(λ);

[0011] S3. Use a polarization spectral imager to capture images of the target scene when the light source is on and off, respectively, to obtain two sets of multispectral image sequences. The multispectral image sequence when the light source is on is... The multispectral image sequence when the light source is off is

[0012] S4. Move the polarization spectral imager and the light source horizontally a certain distance as a whole, and repeat S3 to obtain two sets of multispectral image sequences. The multispectral image sequence when the light source is turned on is... The multispectral image sequence when the light source is off is

[0013] S5. Using radiometric calibration coefficients, the two sets of multispectral image sequences in S3 are... The S4 and S4 sets of multispectral image sequences Converted into forward radiance values ​​respectively and radiance value after movement The target information radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source were calculated, and the spectral images after removing backscattered light before and after the overall movement were obtained respectively.

[0014] S6. Based on the turbidity of the water, select spectral images of specific bands from the spectral images after removing backscattered light before and after the overall movement. Combine the intrinsic parameter matrix and distortion coefficient of the polarization spectral imager to preprocess the two spectral images and determine the overlapping area of ​​the two images. Combine binocular ranging to calculate the distance of each pixel in the overlapping area from the target point in the scene to the optical center and light source of the polarization spectral imager before movement, as well as the incident angle of the light source.

[0015] S7. Based on the current water body attenuation coefficient α(λ), the irradiance E0(λ) at 1 meter from the light source, the target information light radiance value before the overall movement, the distance of each pixel in the overlapping area from the target point in the target scene to the optical center of the polarization spectral imager and the light source, and the incident angle of the light source, the spectral reflectance of the target scene is reconstructed to complete the underwater target spectral reflectance reconstruction.

[0016] Furthermore, in S1, the polarization mode of the polarization spectral imager is linear polarization, and the polarization angles are 0°, 45°, 90° and 135°.

[0017] Furthermore, the specific process of S5 is as follows:

[0018] S5.1, the two sets of multispectral image sequences S3 are represented as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... S4 represents the two sets of multispectral image sequences as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... Where x and y represent the abscissa and ordinate of the spectral image sequence, respectively, and λ represents the incident wavelength. Indicates the polarization angle;

[0019] S5.2. Convert the four multispectral image sequences in S5.1 into radiance values ​​using radiometric calibration coefficients, as follows: and

[0020] S5.3 Subtract the multispectral image sequences of the light source being turned on and off at the same location to obtain the active light source spectral image sequence at that location; remove the backscattered light from the active light source spectral image sequence, and calculate the target information light radiance value of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source, respectively, to obtain the spectral image after removing the backscattered light before and after the overall movement.

[0021] Furthermore, in S5.2, the formula for converting the four multispectral image sequences in S5.1 into radiance values ​​using radiometric calibration coefficients is as follows:

[0022]

[0023] Where x takes the value of camera-light-L, camera-L, camera-light-R, or camera-R; C(λ) represents the radiometric calibration coefficient.

[0024] Further, in S5.3, the subtraction of the multispectral image sequences of the light source being turned on and off at the same location to obtain the spectral image sequence of the active light source at that location specifically involves:

[0025] Subtracting the multispectral image sequences of the active light source at its original position before the polarization spectral imager and the light source were moved together, the spectral image sequence L of the active light source at that position is obtained. light-L (x,y,λ) is:

[0026] L light-L (x,y,λ)=L camera-light-L (x,y,λ)-L camera-L (x,y,λ)

[0027] Subtracting the multispectral image sequences of the active light source at the location where the polarization spectral imager and the light source are moved together yields the active light source spectral image sequence L at that location. light-R (x,y,λ) is:

[0028] L light-R (x,y,λ)=L camera-light-R (x,y,λ)-L camera-R (x,y,λ).

[0029] Furthermore, in S5.3, the calculation of the target information radiance value of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source is specifically as follows:

[0030] The Stokes vector representation of the target information light radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source is as follows:

[0031]

[0032] Among them, S m In (x,y,λ), m takes values ​​of light-L and light-R, representing the target information radiance value of the target scene entering the imaging spectrometer before the overall movement of the polarization spectral imager and the light source, and the target information radiance value of the target scene entering the imaging spectrometer after the overall movement of the polarization spectral imager and the light source; p scat (λ) represents the polarization degree of the backscattered light in the spectral image sequence of the active light source. The optimal target information light polarization degree p in the active light source spectral image sequence. obj (λ);

[0033] I m (x,y,λ)=L m (x,y,λ,90°)+L m (x,y,λ,0°);

[0034] Q m (x,y,λ)=L m (x,y,λ,90°)-L m (x,y,λ,0°);

[0035] U m (x,y,λ)=L m (x,y,λ,45°)-L m (x,y,λ,135°);

[0036] in, This represents the spectral image sequence L of an active light source. m (x,y,λ) are processed Image obtained by polarization of an angle;

[0037] The calculation process for the backscattered light polarization degree of the active light source spectral image sequence is as follows:

[0038] The region σ without the target object is outlined in the active light source spectral image sequence corresponding to each wavelength. void (λ), the backscattered polarization degree of the pixels in the active light source spectral image sequence is estimated as:

[0039]

[0040] B max (x,y,λ) is σ void The maximum backscattered radiance in (λ), B min (x,y,λ) is σ void (λ) represents the minimum backscattered radiance; n is σ voidThe number of pixels in (λ);

[0041] The target information optical polarization degree p in the active light source spectral image sequence obj (λ) takes values ​​from 0 to 1 in steps of 0.01, and outlines the region containing a clear single target in the active light source spectral image sequence for each wavelength. clear (λ), the optimal polarization degree of the target information light is calculated using the mutual information between the target information light radiance value and the backscattered light radiance value, and the formula is:

[0042]

[0043] In the formula, MI[B,S] represents B m (x,y,λ) and S m The mutual information value of (x,y,λ), B m (x,y,λ) represents the backscattered light radiance value. Describing the optimal p obj (λ).

[0044] Furthermore, S6 specifically refers to:

[0045] S6.1. Under a certain water turbidity, select the two multispectral image sequences with the highest contrast from the spectra before and after the overall movement, after removing backscattered light. Use the spectra before the movement as the left image and the spectra after the movement as the right image.

[0046] S6.2. Using the intrinsic parameter matrix and distortion coefficients of the polarization spectral imager, as well as the pose information of the target object scene captured by the polarization spectral imager, polarization correction is performed on the two left and right images after removing backscattered light. The SIFT operator is used to perform stereo matching of the two images to determine the overlapping area of ​​the two spectral images.

[0047] S6.3 Calculate the depth Z of the target point P in the scene corresponding to each pixel in the overlapping region. p ;

[0048] S6.4 Perform binocular visual ranging on the left and right images after removing backscattered light, combined with depth Z. p The 3D geometric model calculates the distance l from the optical center of the overall moving polarization spectral imager to the target point P in the scene corresponding to each pixel in the overlapping region. camera (x, y), the distance l of the target point P in the scene corresponding to each pixel in the overlapping region from the light source before the overall movement. light (x,y) and the incident angle of the light source.

[0049] Furthermore, S6.3 specifically refers to:

[0050] With the optical center of the moving pre-polarization spectral imager as the origin O Cl The origin O is taken as the optical center of the moving polarization spectral imager. Cr Establish two coordinate systems: the direction of movement as the x-axis, the vertically upward direction as the z-axis, and any direction perpendicular to both the x-axis and z-axis as the y-axis.

[0051] The horizontal movement distance l of the polarization spectral imager is known. L-R The x-coordinates of target point P in the scene are respectively the x-coordinates of the images formed in the overlapping areas of the left and right images. l x r The focal length f of the polarization spectral imager is used to calculate the depth Z of the target point P along the z-axis. P for:

[0052]

[0053] Furthermore, in S6.4, the incident angle of the light source is the line connecting the light source and the target point P in the target scene before the overall movement, and the target point P in the target scene at xO. Cl The angle formed by the lines connecting the projections onto the y-plane.

[0054] Furthermore, the specific process of S7 is as follows:

[0055] S7.1, From the Lambert-Beer Law, we can obtain:

[0056]

[0057] Among them, S light-L (x,y,λ) represents the target information radiance value before overall movement, L reflect (x,y,λ) represents the radiance reflected by the target object in the overlapping area under active light illumination before the overall movement, α(λ) represents the current water body attenuation coefficient, l camera (x,y) represents the distance between each pixel in the overlapping region and the optical center of the polarization spectral imager;

[0058] S7.2 When a light source illuminates a sphere at a distance of 1m, the irradiance received by all points is the same, and the irradiance received by a distant target is related to the square of the illumination distance. Combining Lambert's law of cosines and the Lambert-Beer law, the irradiance E of the target scene in the overlapping area can be obtained. incidence (x,y,λ) is:

[0059]

[0060] Where E0(λ) represents the irradiance at a distance of 1 meter from the light source, θ i (x,y) represents the incident angle of the light source; llight (x,y) represents the distance between the target point on the target object and the light source corresponding to each pixel in the overlapping region;

[0061] S7.3 Calculate the spectral reflectance of the reconstructed target scene using the following formula:

[0062]

[0063] The beneficial effects of this invention are:

[0064] 1. This invention provides a method for reconstructing the spectral reflectance of underwater targets based on differential calculation. It does not require additional ranging equipment. It only uses a polarization spectral imager to take moving pictures to obtain two sets of images similar to binocular imaging. Through binocular measurement algorithms and solid geometric models, the distance and angle information between the light source, the imager and the target point are calculated. At the same time, combined with the water attenuation coefficient measured in situ by the attenuator and the Lambert-Beer law, the original incident and outgoing radiance values ​​of the target can be calculated, thereby reconstructing the true spectral reflectance of the target.

[0065] 2. The present invention provides a method for reconstructing the spectral reflectance of underwater targets based on differential calculation. It utilizes an active light source scintillation differential method to remove the portion of reflected light generated by sunlight irradiation in the target information light received by the spectrometer, ensuring that the target object is only irradiated by the active light source in the reflectance calculation, without interference from other light sources.

[0066] 3. This invention provides a method for reconstructing the spectral reflectance of underwater targets based on differential calculation. It utilizes image polarization information and Stokes vector difference to remove the influence of backscattered light. Backscattered light is a major cause of reduced image quality and interference with the spectral characteristics of the object itself. By removing backscattered light through polarization information, the target information light is obtained, improving the accuracy of target reflectance reconstruction. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating an embodiment of an underwater target spectral reflectance reconstruction method based on differential calculation according to the present invention;

[0068] Figure 2 This is a schematic diagram of the binocular ranging principle in embodiment S6.5 of the underwater target spectral reflectance reconstruction method based on differential calculation of the present invention;

[0069] Figure 3 This is a schematic diagram of the three-dimensional geometric model in embodiment S6.5 of the present invention, which is a method for reconstructing the spectral reflectance of an underwater target based on differential calculation.

[0070] Figure 4 This is a light source radiation distribution diagram in an embodiment of the underwater target spectral reflectance reconstruction method based on differential calculation according to the present invention;

[0071] Figure 5 This is a schematic diagram of reflectance reconstruction in an embodiment of the underwater target spectral reflectance reconstruction method based on differential calculation according to the present invention. Detailed Implementation

[0072] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] This invention provides a method for reconstructing the spectral reflectance of underwater targets based on differential calculation, such as... Figure 1 As shown, it includes the following steps:

[0074] S1. Radiometric calibration of the polarization spectral imager is performed on water using a radiometric calibration device (standard integrating sphere light source system, spectral radiance standard lamp system, standard spectroradiometer). The pixel output values ​​of the spectrometer are then obtained. With spectral radiance The radiation calibration coefficient C(λ), where The linear polarization direction angles are 0°, 45°, 90° and 135°, respectively. The intrinsic parameter matrix and distortion coefficient of the polarization spectral imager are analyzed using the Zhang Zhengyou calibration method. At the same time, the irradiance E0(λ) at a distance of 1 meter from the light source is measured using a standard spectroradiometer.

[0075] S2. Place the polarization spectral imager and the light source at the same underwater altitude, with a distance of l between them. camera-light Furthermore, both objects are in the same spatial plane. The light source is turned on to illuminate the target scene, and the current water attenuation coefficient α(λ) is measured using a water attenuation coefficient measuring instrument.

[0076] S3. Use a polarization spectral imager to capture images of the target scene, obtaining a sequence of multispectral images. The light source was quickly turned off, and the imaging spectrometer was used again to photograph the target object, obtaining a series of multispectral images.

[0077] S4, Moving polarization spectral imager and light source assembly L-R Repeat step S3 under the condition that the distance, light source, and shooting angle remain unchanged to obtain two sets of multispectral image sequences.

[0078] S5. Using radiometric calibration coefficients, convert the two sets of multispectral image sequences in S3 and S4 into radiance values ​​before and after movement, respectively. Then, calculate the target information radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source, obtaining the spectral images after removing backscattered light before and after the overall movement. The specific process is as follows:

[0079] S5.1, the two sets of multispectral image sequences S3 are represented as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... S4 represents the two sets of multispectral image sequences as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... Where x and y represent the abscissa and ordinate of the spectral image sequence, respectively, and λ represents the incident wavelength. Indicates the polarization angle;

[0080] S5.2. Convert the four multispectral image sequences in S51 into radiance values ​​using radiometric calibration coefficients, as follows: and

[0081] The formula for converting the four multispectral image sequences in S5.1 into radiance values ​​using radiometric calibration coefficients is as follows:

[0082]

[0083] Where x takes the value of camera-light-L, camera-L, camera-light-R, or camera-R; C(λ) represents the radiometric calibration coefficient.

[0084] The radiance values ​​of the spectral image before and after the device is moved, with the light source on, can both be expressed as:

[0085]

[0086] The radiance values ​​of the spectral image before and after the device is moved, with the light source off, can both be expressed as:

[0087]

[0088] Among them, S sun (x,y,λ) represents the target information light radiance value reflected to the imager by ambient light illuminating the object, B. sun (x,y,λ) represents the backscattered light radiance value caused by ambient light.

[0089] S5.3. Subtract the multispectral image sequences of the light source being turned on and off at the same location respectively to obtain the active light source spectral image sequence at that location; specifically:

[0090] Subtracting the multispectral image sequences of the active light source at its original position before the polarization spectral imager and the light source were moved together, with the light source turned on and off respectively, yields the active light source spectral image sequence L at that position. light-L (x,y,λ) is:

[0091]

[0092] Subtracting the multispectral image sequences of the active light source at its original position before the polarization spectral imager and the light source were moved together, with the light source turned on and off respectively, yields the active light source spectral image sequence L at that position. light-R (x,y,λ) is:

[0093]

[0094] Among them, S light-L (x,y,λ),S light-R (x, y, λ) represent the radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source, respectively; B light-L (x,y,λ), B light-R (x, y, λ) represent the backscattered light radiance values ​​before and after the overall movement.

[0095] The active light source spectral image sequence can be represented as:

[0096] L(x,y,λ)=L max (x,y,λ)+L min (x,y,λ)(6)

[0097] Among them, L max (x,y,λ) and L min (x, y, λ) represent the imaging results when the backscattered light radiance is at its maximum and minimum during the imaging process, respectively, and can be expressed as:

[0098] L max (x,y,λ)=S max (x,y,λ)+B max (x,y,λ)(7)

[0099] L min (x,y,λ)=S min (x,y,λ)+B min (x,y,λ)(8)

[0100] The degree of polarization p is represented by the Stokes vector.

[0101]

[0102] in:

[0103]

[0104] L max (x,y,λ)-L min (x,y,λ)=p scat (λ)B(x,y,λ)+p obj (λ)S(x,y,λ)(13)

[0105] The Stokes vector representation of the target information light radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source is as follows:

[0106]

[0107] Among them, S m In (x,y,λ), m takes values ​​of light-L and light-R, representing the target information radiance value of the target scene entering the imaging spectrometer before the overall movement of the polarization spectral imager and the light source, and the target information radiance value of the target scene entering the imaging spectrometer after the overall movement of the polarization spectral imager and the light source; p scat (λ) represents the polarization degree of the backscattered light in the spectral image sequence of the active light source. The optimal target information light polarization degree p in the active light source spectral image sequence. obj (λ);

[0108] I m (x,y,λ)=L m (x,y,λ,90°)+L m (x,y,λ,0°);

[0109] Q m (x,y,λ)=L m (x,y,λ,90°)-L m (x,y,λ,0°);

[0110] U m (x,y,λ)=L m (x,y,λ,45°)-L m (x,y,λ,135°);

[0111] in, This represents the spectral image sequence L of an active light source. m (x,y,λ) are processed Image obtained by polarization of an angle;

[0112] The region σ without the target object is outlined in the active light source spectral image sequence corresponding to each wavelength. void (λ), where each pixel in its region has L max (x,y,λ)≈B max (x,y,λ),L min (x,y,λ)≈B min (x,y,λ), the global backscattered light polarization degree is estimated as:

[0113]

[0114] B max (x,y,λ) is σ void The maximum backscattered radiance in (λ), B min (x,y,λ) is σ void (λ) represents the minimum backscattered radiance; n is σ void The number of pixels in (λ);

[0115] The mutual information between the target information light and the backscattered light is

[0116]

[0117] The target information optical polarization degree p in the active light source spectral image sequence obj (λ) takes values ​​from 0 to 1 in steps of 0.01, and outlines the region containing a clear single target in the active light source spectral image sequence for each wavelength. clear (λ), the optimal polarization degree of the target information light is calculated using the mutual information between the target information light radiance value and the backscattered light radiance value, and the formula is:

[0118]

[0119] In the formula, MI[B,S] represents B m (x,y,λ) and S m The mutual information value of (x,y,λ), B m (x,y,λ) represents the backscattered light radiance value. Describing the optimal p obj (λ).

[0120] The target information radiance value S of the target scene entering the imaging spectrometer before and after the overall movement is calculated using the above formula. light-L (x,y,λ) and S light-R (x,y,λ).

[0121] Based on the target scene before and after overall movement, the target information light radiance value S enters the imaging spectrometer. light-L (x,y,λ) and Slight-R (x,y,λ) yields the spectral images after removing backscattered light before and after the overall movement.

[0122] S6. Based on the water turbidity, select spectral images of specific bands from the spectral images after removing backscattered light before and after the overall movement. Preprocess the two spectral images using the intrinsic parameter matrix and distortion coefficients of the polarization spectral imager, and determine the overlapping area of ​​the two images. Calculate the distance from each pixel in the overlapping area to the target point in the scene before movement, the optical center of the polarization spectral imager, and the light source, as well as the incident angle of the light source; specifically:

[0123] S6.1. Under a certain water turbidity, select the two multispectral image sequences with the highest contrast from the spectra before and after the overall movement, after removing backscattered light. Use the spectra before the movement as the left image and the spectra after the movement as the right image.

[0124] S6.2. Using the intrinsic parameter matrix and distortion coefficients of the polarization spectral imager obtained in S1, as well as the pose information of the polarization spectral imager when capturing the target scene, polarization correction is performed on the two left and right images after removing backscattered light. The SIFT operator is used to perform stereo matching of the two images to determine the overlapping area of ​​the two spectral images.

[0125] S6.3 Calculate the depth Z of the target point P in the scene corresponding to each pixel in the overlapping region. p ;

[0126] With the optical center of the moving pre-polarization spectral imager as the origin O Cl The origin O is taken as the optical center of the moving polarization spectral imager. Cr Establish two coordinate systems: the direction of movement as the x-axis, the vertically upward direction as the z-axis, and any direction perpendicular to both the x-axis and z-axis as the y-axis.

[0127] The horizontal movement distance l of the polarization spectral imager is known. L-R The x-coordinates of target point P in the scene are respectively the x-coordinates of the images formed in the overlapping areas of the left and right images. l x r The focal length f of the polarization spectral imager, i.e., by Figure 2 And the known distance l between the optical centers of the two images L-R The x-coordinate of the target point in the left and right images. l x r The depth Z can be calculated using the focal length f of the imaging spectrometer and the following formula (19). P :

[0128]

[0129] S6.4 Perform binocular visual ranging on the left and right images after removing backscattered light, combined with depth Z. p The 3D geometric model calculates the distance l from the optical center of the overall moving polarization spectral imager to the target point P in the scene corresponding to each pixel in the overlapping region. camera (x, y), the distance l of the target point P in the scene corresponding to each pixel in the overlapping region from the light source before the overall movement. light (x, y) and the incident angle of the light source: by Figure 2 Two pairs of similar triangles ΔO C AH~ΔO C OC and ΔO C HP~ΔO C Cp can be obtained from the following relationship

[0130]

[0131] The coordinates of the target point P in the camera coordinate system can be calculated using equations (19) and (20), and then the distance of the target point P from the optical center O of the imaging spectrometer can be calculated. C Distance l camera and distance from light source O L Distance l light for

[0132]

[0133] Simultaneously, the angle θ between the incident ray from the light source to the target point P and the plane containing the target point can be calculated. i for:

[0134]

[0135] According to equations (19)(20)(21)(22)(23), the distance l from each pixel in the overlapping region to the optical center of the imaging spectrometer and the light source can be calculated. camera (x,y),l light (x,y) and the incident angle θ of the light source i (x,y).

[0136] S7. Based on the current water attenuation coefficient, irradiance at 1 meter from the light source, target information radiance value before overall movement, distance of each pixel in the overlapping area from the target point in the scene to the optical center of the polarization spectral imager and the light source, and the incident angle of the light source, reconstruct the spectral reflectance of the target scene to complete the underwater target spectral reflectance reconstruction. The specific process is as follows:

[0137] S7.1, From the Lambert-Beer Law, we can obtain:

[0138]

[0139] Among them, S light-L (x,y,λ) represents the target information radiance value before overall movement, L reflect (x,y,λ) represents the radiance reflected by the target object in the overlapping area under active light illumination before the overall movement, α(λ) represents the current water body attenuation coefficient, l camera (x,y) represents the distance between each pixel in the overlapping region and the optical center of the polarization spectral imager;

[0140] Therefore, the radiance reflected by the target object scene in the overlapping area under active light illumination can be obtained as follows:

[0141]

[0142] S7.2 When a light source illuminates a sphere at a distance of 1m, the irradiance received by all points is the same, and the irradiance received by a distant target is related to the square of the illumination distance. Combining Lambert's law of cosines and the Lambert-Beer law, the irradiance E of the target scene in the overlapping area can be obtained. incidence (x,y,λ) is:

[0143]

[0144] Where E0(λ) represents the irradiance at a distance of 1 meter from the light source, θ i (x,y) represents the incident angle of the light source; l light (x,y) represents the distance between the target point on the target object and the light source corresponding to each pixel in the overlapping region;

[0145] S7.3 Calculate the spectral reflectance of the reconstructed target scene using the following formula:

[0146]

[0147] This invention discloses an underwater target spectral reflectance reconstruction method based on differential calculation. It removes the influence of backscattered light during imaging by using polarization information and Stokes vectors, calculates the distance between the device (imager, light source) and all target points in the image using a neighboring frame image difference method, and removes the influence of ambient light on data inversion using a light field difference method. This avoids the shortcomings of existing reflectance calculation methods (which require the object and the whiteboard to be at the same distance from the imager to retrieve the true reflectance data of the object, and the backscattered light caused by the active light source at long distance imaging cannot be ignored). Using an underwater imaging model and the Lambert-Beer law, it accurately calculates the incident light radiance value and the received information light radiance value of the active light source at target points at different distances within the measurement range in the image, thereby retrieving the true spectral reflectance of the object. This reduces the impact of uneven illumination and ambient light on the accuracy of data inversion, improving the accuracy of spectral reflectance inversion.

[0148] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for reconstructing the spectral reflectance of underwater targets based on differential calculation, characterized in that, Includes the following steps: S1. Perform radiometric calibration on the polarization spectral imager to obtain the radiometric calibration coefficients, and use Zhang Zhengyou calibration method to calculate the intrinsic parameter matrix and distortion coefficients of the polarization spectral imager. At the same time, use a standard spectroradiometer to measure the irradiance at a distance of 1 meter from the light source. S2. Place the polarization spectral imager and light source at the same height underwater, turn on the light source to illuminate the target scene, and use the water attenuation coefficient measuring instrument to measure the current water attenuation coefficient. S3. Use a polarization spectral imager to capture the target scene when the light source is on and off, respectively, to obtain two sets of multispectral image sequences; S4. Move the polarization spectral imager and the light source horizontally a certain distance as a whole, repeat S3, and obtain two sets of multispectral image sequences. S5. Using radiometric calibration coefficients, convert the two sets of multispectral image sequences in S3 and S4 into radiance values ​​before and after movement, respectively. Calculate the radiance values ​​of the target information light entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source, and obtain the spectral images after removing backscattered light before and after the overall movement. S6. Based on the turbidity of the water, select spectral images of specific bands from the spectral images after removing backscattered light before and after the overall movement. Combine the intrinsic parameter matrix and distortion coefficient of the polarization spectral imager to preprocess the two spectral images and determine the overlapping area of ​​the two images. Combine binocular ranging to calculate the distance of each pixel in the overlapping area from the target point in the scene to the optical center and light source of the polarization spectral imager before movement, as well as the incident angle of the light source. S7. Based on the current water body attenuation coefficient, irradiance at 1 meter from the light source, target information radiance value before overall movement, distance of each pixel in the overlapping area from the target point in the target scene to the optical center of the polarization spectral imager and the light source, and the incident angle of the light source, the spectral reflectance of the target scene is reconstructed to complete the underwater target spectral reflectance reconstruction.

2. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 1, characterized in that, In S1, the polarization mode of the polarization spectral imager is linear polarization, and the polarization angles are 0°, 45°, 90° and 135°.

3. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 1, characterized in that, The specific process of S5 is as follows: S5.1, the two sets of multispectral image sequences S3 are represented as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... S4 represents the two sets of multispectral image sequences as follows: with the light source turned on, the multispectral image sequence is... With the light source off, the multispectral image sequence is... Where x and y represent the abscissa and ordinate of the spectral image sequence, respectively, and λ represents the incident wavelength. Indicates the polarization angle; S5.

2. Convert the four multispectral image sequences in S5.1 into radiance values ​​using radiometric calibration coefficients, as follows: and S5.3 Subtract the multispectral image sequences of the light source being turned on and off at the same location respectively to obtain the active light source spectral image sequence at that location; By removing backscattered light from the active light source spectral image sequence, a new spectral image is obtained. The target information light radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source are calculated respectively. This yields the spectral images after removing backscattered light before and after the overall movement.

4. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 3, characterized in that, In S5.2, the formula for converting the four multispectral image sequences in S5.1 into radiance values ​​using radiometric calibration coefficients is as follows: Where x takes the value of camera-light-L, camera-L, camera-light-R, or camera-R; C(λ) represents the radiometric calibration coefficient.

5. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 4, characterized in that, In S5.3, the specific steps of subtracting the multispectral image sequences of the light source being turned on and off at the same location to obtain the spectral image sequence of the active light source at that location are as follows: Subtracting the multispectral image sequences of the active light source at its original position before the polarization spectral imager and the light source were moved together, with the light source turned on and off respectively, yields the active light source spectral image sequence L at that position. light-L (x,y,λ) is: L light-L (x,y,λ)=L camera-light-L (x,y,λ)-L camera-L (x,y,λ) Subtracting the multispectral image sequences of the active light source at the location where the polarization spectral imager and the light source are moved together yields the active light source spectral image sequence L at that location. light-R (x,y,λ) is: L light-R (x,y,λ)=L camera-light-R (x,y,λ)-L camera-R (x,y,λ).

6. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 5, characterized in that, In S5.3, the specific steps for calculating the target information radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source are as follows: The Stokes vector representation of the target information light radiance values ​​of the target scene entering the imaging spectrometer before and after the overall movement of the polarization spectral imager and the light source is as follows: Among them, S m In (x,y,λ), m takes values ​​of light-L and light-R, representing the target information radiance value of the target scene entering the imaging spectrometer before the overall movement of the polarization spectral imager and the light source, and the target information radiance value of the target scene entering the imaging spectrometer after the overall movement of the polarization spectral imager and the light source; p scat (λ) represents the polarization degree of the backscattered light in the spectral image sequence of the active light source. The optimal target information light polarization degree p in the active light source spectral image sequence. obj (λ); I m (x,y,λ)=L m (x,y,λ,90°)+L m (x,y,λ,0°); Q m (x,y,λ)=L m (x,y,λ,90°)-L m (x,y,λ,0°)? U m (x,y,λ)=L m (x,y,λ,45°)-L m (x,y,λ,135°); in, This represents the spectral image sequence L of an active light source. m (x,y,λ) are processed Image obtained by polarization of an angle; The calculation process for the backscattered light polarization degree of the active light source spectral image sequence is as follows: The region σ without the target object is outlined in the active light source spectral image sequence corresponding to each wavelength. void (λ), the backscattered polarization degree of the pixels in the active light source spectral image sequence is estimated as: B max (x,y,λ) is σ void The maximum backscattered radiance in (λ), B min (x,y,λ) is σ void (λ) represents the minimum backscattered radiance; n is σ void The number of pixels in (λ); The target information optical polarization degree p in the active light source spectral image sequence obj (λ) takes values ​​from 0 to 1 in steps of 0.01, and outlines the region containing a clear single target in the active light source spectral image sequence for each wavelength. clear (λ), the optimal polarization degree of the target information light is calculated using the mutual information between the target information light radiance value and the backscattered light radiance value, and the formula is: In the formula, MI[B,S] represents B m (x,y,λ) and S m The mutual information value of (x,y,λ), B m (x,y,λ) represents the backscattered light radiance value. Describing the optimal p obj (λ).

7. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 3, characterized in that, S6 specifically refers to: S6.

1. Under a certain water turbidity, select the two multispectral image sequences with the highest contrast from the spectra before and after the overall movement, after removing backscattered light. Use the spectra before the movement as the left image and the spectra after the movement as the right image. S6.

2. Using the intrinsic parameter matrix and distortion coefficients of the polarization spectral imager, as well as the pose information of the target object scene captured by the polarization spectral imager, polarization correction is performed on the two left and right images after removing backscattered light. The SIFT operator is used to perform stereo matching of the two images to determine the overlapping area of ​​the two spectral images. S6.3 Calculate the depth Z of the target point P in the scene corresponding to each pixel in the overlapping region. p ; S6.4 Perform binocular visual ranging on the left and right images after removing backscattered light, combined with depth Z. p The 3D geometric model calculates the distance l from the optical center of the overall moving polarization spectral imager to the target point P in the scene corresponding to each pixel in the overlapping region. camera (x, y), the distance l of the target point P in the scene corresponding to each pixel in the overlapping region from the light source before the overall movement. light (x,y) and the incident angle of the light source.

8. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 7, characterized in that, S6.3 specifically refers to: With the optical center of the moving pre-polarization spectral imager as the origin O Cl The origin O is taken as the optical center of the moving polarization spectral imager. Cr Establish two coordinate systems: the direction of movement as the x-axis, the vertically upward direction as the z-axis, and any direction perpendicular to both the x-axis and z-axis as the y-axis. The horizontal movement distance l of the polarization spectral imager is known. L-R The x-coordinates of target point P in the scene are respectively the x-coordinates of the images formed in the overlapping areas of the left and right images. l x r The focal length f of the polarization spectral imager is used to calculate the depth Z of the target point P along the z-axis. P for:

9. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 8, characterized in that, In S6.4, the incident angle of the light source is the line connecting the light source and the target point P in the target scene before the overall movement, and the target point P in the target scene at xO. Cl The angle formed by the lines connecting the projections onto the y-plane.

10. The underwater target spectral reflectance reconstruction method based on differential calculation according to claim 7, characterized in that, The specific process for S7 is as follows: S7.1, From the Lambert-Beer Law, we can obtain: Among them, S light-L (x,y,λ) represents the target information radiance value before overall movement, L reflect (x,y,λ) represents the radiance reflected by the target object in the overlapping area under active light illumination before the overall movement, α(λ) represents the current water body attenuation coefficient, l camera (x,y) represents the distance between each pixel in the overlapping region and the optical center of the polarization spectral imager; S7.2 When a light source illuminates a sphere at a distance of 1m, the irradiance received by all points is the same, and the irradiance received by a distant target is related to the square of the illumination distance. Combining Lambert's law of cosines and the Lambert-Beer law, the irradiance E of the target scene in the overlapping area can be obtained. incidence (x,y,λ) is: Where E0(λ) represents the irradiance at a distance of 1 meter from the light source, θ i (x,y) represents the incident angle of the light source; l light (x,y) represents the distance between the target point on the target object and the light source corresponding to each pixel in the overlapping region; S7.3 Calculate the spectral reflectance of the reconstructed target scene using the following formula:

Images