Visible light image defogging method based on short wave infrared geometric feature guidance

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a visible light image defogging method based on short-wave infrared geometric feature guidance, which comprises the following steps of: S1, acquiring a visible light image Ivis and a short-wave infrared image Iswitch in a foggy day environment in the same scene through a visible light camera and an infrared camera; respectively decomposing the acquired visible light image Ivis and the acquired short wave infrared image Iswitch into base layers Bvis and Bswitch and detail layers Dvis and Dswitch through weighted least squares WLS; s2, extracting a curvature geometric feature kswitch of the acquired short wave infrared image Iswitch, and carrying out weight mapping sigma (kswitch) according to the curvature geometric feature kswitch; the core innovation of the method is that the curvature geometric features and the gradient direction angle geometric features of the short-wave infrared image are combined, a layered optimization guiding mechanism is utilized, the purpose of defogging the visible light image is achieved, the edge structure protection capability and the texture detail recovery quality of a foggy day scene are remarkably improved, and the method has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method for dehazing visible light images based on short-wave infrared geometric features. Background Technology

[0002] In recent years, my country's computer vision field has developed rapidly, covering multiple areas such as autonomous driving and healthcare. Haze blurs distant scenes and causes loss of detail in images, affecting subsequent image analysis tasks. Image dehazing is an important task in computer vision, aiming to restore images distorted by adverse atmospheric conditions such as haze or smog.

[0003] Current research on image dehazing mainly faces the following problems: First, image processing using only a certain wavelength band contains too little information and often cannot fully recover all the details in a scene, especially under low visibility conditions. Second, in the fusion process of short-wave infrared and visible light, directly superimposing the infrared detail layer onto the visible light image with fixed fusion weights can lead to noise amplification in thin fog areas, insufficient detail compensation in dense fog areas, and may also cause spectral distortion. Third, while deep learning methods perform well in image dehazing, the training process is computationally complex and time-consuming, and is limited by hardware performance. They are also highly sensitive to noise, especially in strong haze environments, where the model may not be able to effectively suppress noise, resulting in unsatisfactory dehazing results.

[0004] Therefore, it is essential to combine the advantages of visible light images and infrared images and introduce a visible light image dehazing method guided by short-wave infrared geometric features to solve these problems. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a visible light image dehazing method guided by short-wave infrared geometric features, which solves the problems mentioned in the background section.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0009] A visible light image dehazing method based on short-wave infrared geometric features includes the following steps:

[0010] S1: Visible light images of a foggy environment in the same scene are acquired using a visible light camera and an infrared camera. vis With shortwave infrared image I swir Then, the acquired visible light image I vis With shortwave infrared image Iswir The basic layer B is decomposed using weighted least squares WLS. vis B swir With detail layer D vis D swir ;

[0011] S2: Extract the acquired shortwave infrared image I swir curvature geometry feature k swir And based on the curvature geometry feature k swir Perform weight mapping σ(k) swir );

[0012] S3: For the base layer B of the decomposed visible light image vis First, the transmittance map of the visible light base layer is optimized using the curvature regularization method. Second, the base layer weakening method is used to protect the edge structure and generate the optimized base layer B' for visible light images. vis Finally, a weight mapping based on σ(k) is performed. swir Image fusion yields the basic fusion layer B. fused ;

[0013] S4: For the detail layer D of the decomposed visible light image vis First, by extracting the detail layer D of the short-wave infrared image. swir The gradient direction angle geometric features θ are used to construct a direction-sensitive mask M(θ), and then combined with the fog density estimation ρ h A detail compensation method is used to adaptively adjust the injection intensity β to optimize texture details. Finally, image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused ;

[0014] S5: B fused and D fused The synthesized visible light image I is obtained after dehazing. out .

[0015] Furthermore, in step S1, the visible light image and the short-wave infrared image are respectively decomposed into a base layer B. vis B swir With detail layer D vis D swir The specific formula is as follows:

[0016]

[0017] Among them, I vis The acquired visible light image; I swir For the acquired shortwave infrared image; B vis This serves as the base layer for the decomposed visible light image; D vis B is the detail layer of the decomposed visible light image; swirThis forms the base layer for the decomposed shortwave infrared image; D swir λ represents the detail layer of the decomposed shortwave infrared image; λ is the filter intensity.

[0018] Furthermore, based on the atmospheric scattering model, the transmittance can be estimated through the dark channel information of the image, thereby obtaining the fog density. In S2, the fog density value ρ... h The calculation formula for estimation is as follows:

[0019] The atmospheric scattering model can be expressed as:

[0020] I = J·t + A·(1-t)

[0021] Where I is the observed image; J is the fog-free image; t is the transmittance value t∈[0,1]; and A is the atmospheric light value.

[0022] In foggy areas, dark passages can be approximated as...

[0023]

[0024] And due to the fog-free area

[0025]

[0026] therefore

[0027]

[0028] Solve for transmittance

[0029]

[0030] Therefore, fog density ρ h Estimated as

[0031]

[0032] Where Ω refers to the global spatial domain of the image; ρ h This is an estimate of the fog density, ρ h ∈[0,1]; A is the global atmospheric light value, defined as the average value of the top 0.1% of pixels in the visible light image; I vis (c) is the acquired visible light image, c∈{r,g,b}, and t is the transmittance value, t∈[0,1].

[0033] Further, in S2, short-wave infrared image I is extracted. swir curvature geometry feature k swir The calculation formula is as follows:

[0034]

[0035] Expanded form:

[0036]

[0037] Where k is the original curvature feature; It is the image gradient. It is the unit normal vector. ∈ is a zero constant, ∈ = 10 -5 ; It is a divergence operator; It is the gradient magnitude.

[0038]

[0039] Where k is the original curvature feature; k swir The curvature geometric features of the extracted shortwave infrared image are normalized to [0,1]; τ is the curvature discrimination threshold, and τ = 0.25 is taken: curvature below τ is regarded as flat region, and curvature above τ is edge region.

[0040] Furthermore, the specific formula for the weight mapping calculation method based on the extracted short-wave infrared curvature geometric feature values ​​in step S2 is as follows:

[0041]

[0042] Where, σ(k) swir ) represents the visible light weighting function for curvature modulation of shortwave infrared images; σ represents the sigmoid function; k is the slope coefficient of the function; k0 is the center offset of the function; ρ h This is an estimated fog density value.

[0043] Furthermore, the calculation formula for improving the transmittance map of the visible light base layer using curvature regularization in S3 is as follows:

[0044]

[0045] Where t' is the optimized visible light base layer transmittance map; J is the haze-free image; t is the initial transmittance estimate; I vis The acquired visible light image; ||I vis -J·t|| 2 λ represents the data fidelity term; λ is the overall strength of regularization. The gradient of the haze-free image J; γ is the regularization weight coefficient; k swir Curvature geometry features of the extracted shortwave infrared image; For the curvature regularization term: using k swir Controlling the regularization strength, k swir Small, Suppress transmittance noise; k swir big, Protect the edges of the transmittance.

[0046] Furthermore, in S3, for the decomposed visible light base layer B... vis The formula for protecting edge structures using the base layer weakening method is as follows:

[0047]

[0048] Among them, B' vis B is the base optimization layer for visible light images. vis This serves as the base layer for the decomposed visible light image; Γ(B vis / t',k swir ) is the adaptive attenuation function; t' is the optimized visible light base layer transmittance map; k swir The curvature geometry features of the extracted shortwave infrared image; α is an adjustable attenuation coefficient that controls the influence of curvature on the attenuation intensity, α∈[0,1], with a default value of 0.3: in k swir Large regions, exponent 1-α·(1-k) swir Approaching 1 preserves edge details; while at k swir In small regions, the exponent is close to 1-α, which suppresses texture noise and achieves curvature-guided adaptive decay.

[0049] Furthermore, in S3, the weight mapping σ(k) swir Image fusion is performed to obtain the basic fusion layer B. fused The calculation formula is as follows:

[0050] B fused =σ(k) swir )·B′ vis

[0051] Among them, B fused This serves as the base layer for the fused visible light image; σ(k swir B' is the visible light weighting function for curvature modulation of shortwave infrared images; vis This is the base optimization layer for visible light images.

[0052] Furthermore, in S4, the detail layer D of the short-wave infrared image is extracted. swir The formula for calculating the orientation-sensitive mask M(θ) constructed from the gradient direction angle geometric feature θ is as follows:

[0053]

[0054] θ = arctan2(G y G x )

[0055]

[0056] Where * is defined as the convolution operation; D swir For the detailed layer of the decomposed shortwave infrared image; G x and G y The first and second matrices represent the gradient components of the short-wave infrared detail layer image in the horizontal and vertical directions, respectively; the first and second matrices represent the Sobel operator kernels in the x and y directions, respectively; θ is the gradient direction angle geometric feature of the short-wave infrared image detail layer; M(θ) is a direction-sensitive mask function used to enhance horizontal and vertical edges and suppress noise at diagonal edges; the three parameters 1.2, 0.8, and 0.3 represent the enhancement, moderate enhancement, and suppression factors required when θ is within the angle range, respectively, and are represented from top to bottom as: horizontal edge enhancement coefficient, vertical edge base coefficient, and diagonal texture suppression coefficient; 22.5° is the angle classification threshold to ensure that there is no overlap in classification between horizontal and vertical edges.

[0057] Furthermore, in S4, the fog density estimation ρ is combined h The calculation formula for using a detail compensation method to adaptively adjust the injection intensity β to optimize texture details is as follows:

[0058] β=0.1(1+0.5ρ h )

[0059] Where β is the gradient injection intensity, which controls the feature compensation intensity of the detail layer in the shortwave infrared image, and is determined according to the fog concentration ρ. h Adaptively adjust the injection intensity β, β∈[0.1,0.15]; ρ h This represents the estimated fog density value.

[0060] Furthermore, in step S4, image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused The calculation formula is as follows:

[0061] D′ vis =I vis -B′ vis

[0062] D fused =(1-σ(k) swir ))·D′ vis +β·M(θ)·D swir

[0063] Among them, D' vis A detail optimization layer for visible light images; D fused For the detail layer of the fused visible light image; σ(k swir ) represents the visible light weighting function for curvature modulation of shortwave infrared images; k swir The curvature geometry features of the extracted shortwave infrared image; β·M(θ)·Dswir For detailed compensation amount.

[0064] Furthermore, in S5, B fused and D fused The synthesized visible light image I is obtained after dehazing. out The calculation formula is as follows:

[0065] I out =B fused +D fused

[0066] Among them, I out B is the final output image for dehazing visible light. fused This serves as the base layer for the fused visible light image; D fused This is the detail layer of the fused visible light image.

[0067] (III) Beneficial Effects

[0068] Compared with existing technologies, this invention provides a visible light image dehazing method guided by short-wave infrared geometric features, which has the following beneficial effects:

[0069] This invention combines the unique advantages of visible light images and short-wave infrared images. Short-wave infrared images provide clear outlines and structural information, while visible light images provide rich color information and details. By utilizing the rich geometric features of short-wave infrared images, the structural information of visible light images can be effectively restored while ensuring clear details, thus significantly improving the dehazing effect.

[0070] The weighted fusion method proposed in this invention utilizes the curvature feature k of shortwave infrared images. swir With fog density value ρ h Adaptive generation of weight mapping solves the problems of insufficient detail in dense fog and noise in thin fog caused by fixed weights.

[0071] Compared to deep learning methods, combining visible light and shortwave infrared images for analysis and calculation can more realistically reproduce the scene blurring and contrast reduction caused by smog, providing a new approach for image processing research. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below:

[0073] Figure 1 This is a flowchart of the method of this invention patent;

[0074] Figure 2 This is a schematic diagram of the visible light image dehazing method based on short-wave infrared geometric features of the present invention.

[0075] Figure 3 These are dehazing effect diagrams of this invention patent, where Figure (a) is the acquired visible light image I. vis Figure (b) shows the acquired shortwave infrared image I. swir Figure (c) shows the final visible light dehazing image I output after processing by this method. out . Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0077] like Figure 3 As shown in the figure, an embodiment of the present invention proposes a visible light image dehazing method based on short-wave infrared geometric features, comprising:

[0078] S1: Visible light images of foggy urban street environments are acquired using visible light and infrared cameras. vis With shortwave infrared image I swir The acquired visible light image I vis With shortwave infrared image I swir WLS is used to decompose the data into a base layer B. vis B swir With detail layer D vis D swir .

[0079] S2: Extract the acquired shortwave infrared image I swir curvature geometry feature k swir And based on the curvature geometry feature k swir Perform weight mapping σ(k) swir ).

[0080] For fog density value ρ h Estimation is performed based on an atmospheric scattering model. Transmittance can be estimated using dark channel information from the image, and then fog density can be obtained. The specific process is as follows:

[0081] The atmospheric scattering model can be expressed as:

[0082] I = J·t + A·(1-t)

[0083] Where I is the observed image; J is the fog-free image; t is the transmittance value t∈[0,1]; and A is the atmospheric light value.

[0084] In foggy areas, dark passages can be approximated as...

[0085]

[0086] And due to the fog-free area

[0087]

[0088] therefore

[0089]

[0090] Solve for transmittance

[0091]

[0092] Therefore, fog density ρ h Estimated as

[0093]

[0094] Where Ω refers to the global spatial domain of the image; ρ h This is an estimate of the fog density, ρ h ∈[0,1]; A is the global atmospheric light value, defined as the average value of the top 0.1% of pixels in the visible light image, with a measured value of 235; I vis (c) is the acquired visible light image, c∈{r,g,b}, and t is the transmittance value, t∈[0,1].

[0095] Extracting shortwave infrared images I swir The curvature geometry characteristic is expressed as:

[0096]

[0097] Expanded form:

[0098]

[0099] Where k is the original curvature feature; It is the image gradient. It is the unit normal vector. ∈ is a zero constant, ∈ = 10 -5 ; It is a divergence operator; It is the gradient magnitude.

[0100]

[0101] Where k is the original curvature feature; k swir The curvature geometric features of the extracted shortwave infrared image are normalized to [0,1]; τ is the curvature discrimination threshold, taken as τ=0.25: road surface texture region k swir =0.15, 0.15 < 0.2, considered a flat area; traffic light area k swir=0.85, 0.85>0.8, are considered to be in the edge region.

[0102] Weight mapping is performed based on the extracted shortwave infrared curvature geometric feature values, expressed as follows:

[0103]

[0104] Where, σ(k) swir ) represents the visible light weighting function for curvature modulation of shortwave infrared images; σ represents the sigmoid function; k is the slope coefficient of the function; k0 is the center offset of the function; ρ h This is an estimated fog density value.

[0105] S3: For the base layer B of the decomposed visible light image vis First, the transmittance map of the visible light base layer is optimized using the curvature regularization method. Second, the base layer weakening method is used to protect edge structures, such as traffic lights at long distances, thus generating the base optimized layer B' for the visible light image. vis Finally, a weight mapping based on σ(k) is performed. swir Image fusion yields the basic fusion layer B. fused .

[0106] The transmittance map of the visible light base layer is improved using curvature regularization optimization. The calculation formula is as follows:

[0107]

[0108] Where t' is the optimized transmittance map of the visible light base layer; J is the haze-free image; t is the initial transmittance estimate; I vis The acquired visible light image; ||I vis -J·t|| 2 λ represents the data fidelity term; λ is the overall strength of regularization, taken as λ = 0.05; The gradient of the haze-free image J; γ is the regularization weight coefficient, taken as γ = 1.8; k swir Curvature geometry features of the extracted shortwave infrared image; For the curvature regularization term: using k swir Controlling the regularity intensity, in the road surface texture area, k swir =0.15, noise suppression rate is 62%, transmittance noise is suppressed; in the signal light area, k swir =0.85, structure retention 98.7%, protecting the edge of transmittance.

[0109] The calculation formula for the grassroots weakening method is:

[0110]

[0111] Among them, B'vis B is the base optimization layer for visible light images. vis This serves as the base layer for the decomposed visible light image; Γ(B vis / t',k swir ) is the adaptive attenuation function; t' is the optimized transmittance map of the visible light base layer; k swir The curvature geometry features of the extracted shortwave infrared image; α is an adjustable attenuation coefficient that controls the influence of curvature on the attenuation intensity, α∈[0,1], taking α=0.3: in the traffic light area, k swir =0.85, α=0.03, almost no attenuation, preserving edge details; while in the road surface texture area, k swir =0.15, α=0.28, texture noise is suppressed. Curvature-guided adaptive attenuation is achieved.

[0112] Based on weight mapping σ(k) swir The image fusion calculation formula is as follows:

[0113] B fused =σ(k) swir )·B′ vis

[0114] Among them, B fused This serves as the base layer for the fused visible light image; σ(k swir B' is the visible light weighting function for curvature modulation of shortwave infrared images; vis This is the base optimization layer for visible light images.

[0115] S4: For the detail layer D of the decomposed visible light image vis First, by extracting the detail layer D of the short-wave infrared image. swir The gradient direction angle geometric features θ are used to construct a direction-sensitive mask M(θ), and then combined with the fog density estimation ρ h A detail compensation method is used to adaptively adjust the injection intensity β to optimize texture details, such as road surface texture markings. Finally, image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused ;

[0116] By extracting the detail layer D of short-wave infrared images swir The formula for calculating the orientation-sensitive mask M(θ) constructed from the gradient direction angle geometric feature θ is as follows:

[0117]

[0118] θ = arctan2(G y G x )

[0119]

[0120] Where * is defined as the convolution operation; D swir For the detailed layer of the decomposed shortwave infrared image; G x and G y The first and second matrices represent the gradient components of the shortwave infrared detail layer image in the horizontal and vertical directions, respectively; the first and second matrices represent the Sobel operator kernels in the x and y directions, respectively; θ is the gradient direction angular geometric feature of the shortwave infrared image detail layer; M(θ) is a direction-sensitive mask function used to enhance horizontal and vertical edges and suppress noise at diagonal edges; the three parameters 1.2, 0.8, and 0.3 represent the enhancement, moderate enhancement, and suppression factors required when θ is within the angle range, respectively, and are represented from top to bottom as: horizontal edge enhancement coefficient, vertical edge base coefficient, and diagonal texture suppression coefficient; 22.5° is the angle classification threshold to ensure no overlap in classification between horizontal and vertical edges. In this embodiment, the horizontal traffic markings are enhanced by 1.2 times, and the diagonal leaf texture is suppressed by 0.3 times.

[0121] Combined with fog density estimation ρ h The injection intensity β is adaptively adjusted using a detail compensation method to optimize texture details, such as road surface texture markings. The calculation formula is as follows:

[0122] β=0.1(1+0.5ρ h ) = 0.135

[0123] Where β is the gradient injection intensity, which controls the feature compensation intensity of the detail layer in the shortwave infrared image, and is determined according to the fog concentration ρ. h Adaptively adjust the injection intensity β, β∈[0.1,0.15]; ρ h For the estimated fog density value, ρ h =0.7.

[0124] Image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused The calculation formula is as follows:

[0125] D′ vis =I vis -B′ vis

[0126] D fused =(1-σ(k) swir ))·D′ vis +0.135·M(θ)·D swir

[0127] Among them, D' vis A detail optimization layer for visible light images; D fused For the detail layer of the fused visible light image; σ(k swir) represents the visible light weighting function for curvature modulation of shortwave infrared images; k swir The curvature geometry features of the extracted shortwave infrared image; 0.135·M(θ)·D swir For detailed compensation amount.

[0128] S5: B fused and D fused The synthesized visible light image I is obtained after dehazing. out .

[0129] I out =B fused +D fused

[0130] Among them, I out B is the final output image for dehazing visible light. fused This serves as the base layer for the fused visible light image; D fused This is the detail layer of the fused visible light image.

[0131] Figure 3 This is a dehazing effect diagram of the present invention patent. Figure (a) shows the acquired visible light image I. vis Figure (b) shows the acquired shortwave infrared image I. swir Figure (c) shows the final visible light dehazing image I output after processing by the method of this patent. out .

[0132] This invention addresses the shortcomings of traditional dehazing techniques, such as detail loss and noise amplification. It innovatively proposes an image fusion method guided by the geometric features of short-wave infrared images, ultimately obtaining a visible light dehazed image. The key technological advantages are: an innovative fog density-adaptive weight control system that resolves the contradiction between detail loss in dense fog and noise in thin fog caused by fixed-weight fusion; and the combination of an atmospheric scattering model and direction-sensitive detail compensation, which restores scene contrast while preserving texture details. This provides a new approach for image processing research.

[0133] Finally, it should be noted that the above description is merely a preferred embodiment of this application, enabling those skilled in the art to understand or implement this application, and does not constitute a limitation on the present invention. Although this invention has been described in detail through this embodiment, those skilled in the art can still modify the technical solutions or make equivalent substitutions for some technical features as needed. The general principles described in this application can be applied to other embodiments without departing from the basic spirit and scope of this application. Therefore, this application is not limited to the embodiments described herein, but should cover the broadest scope consistent with the technical principles and innovative features of this application.

Claims

1. A visible light image dehazing method based on short-wave infrared geometric features, characterized in that: Specifically, the following steps are included: S1: Visible light images of a foggy environment in the same scene are acquired using a visible light camera and an infrared camera. vis With shortwave infrared image I swir Then, the acquired visible light image I vis With shortwave infrared image I swir The basic layer B is decomposed using weighted least squares WLS. vis B swir With detail layer D vis D swir ; S2: Extract the acquired shortwave infrared image I swir curvature geometry feature k swir And based on the curvature geometry feature k swir Perform weight mapping σ(k) swir ); S3: For the base layer B of the decomposed visible light image vis First, the transmittance map of the visible light base layer is optimized using the curvature regularization method. Second, the base layer weakening method is used to protect the edge structure and generate the optimized base layer B' for visible light images. vis Finally, a weight mapping based on σ(k) is performed. swir Image fusion yields the base fusion layer B. fused ; S4: For the detail layer D of the decomposed visible light image vis First, by extracting the detail layer D of the short-wave infrared image. swir The gradient direction angle geometric features θ are used to construct a direction-sensitive mask M(θ), and then combined with the fog density estimation ρ h A detail compensation method is used to adaptively adjust the injection intensity β to optimize texture details. Finally, image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused ; S5: B fused and D fused The synthesized visible light image I is obtained after dehazing. out .

2. The visible light image dehazing method based on short-wave infrared geometric features as described in claim 1, characterized in that: The shortwave infrared image I acquired in S2 is extracted. swir curvature geometry feature k swir Perform weight mapping σ(k) swir The process includes the following steps: Step 1, calculate the fog density value ρ h Estimation is performed based on an atmospheric scattering model. Transmittance can be estimated using dark channel information from the image, and then fog density can be obtained. The specific process is as follows: The atmospheric scattering model can be expressed as: I = J·t + A·(1-t) Where I is the observed image; J is the fog-free image; t is the transmittance value t∈[0,1]; and A is the atmospheric light value; In foggy areas, dark passages can be approximated as... And due to the fog-free area therefore Solve for transmittance Therefore, fog density ρ h Estimated as Where Ω refers to the global spatial domain of the image; ρ h This is an estimate of the fog density, ρ h ∈[0,1]; A is the global atmospheric light value, defined as the average value of the top 0.1% of pixels in the visible light image; I vis (c) is the acquired visible light image, c∈{r,g,b}, t is the transmittance value, t∈[0,1]; Step 2, extract shortwave infrared image I swir curvature geometry feature k swir The expression is: Expanded form: Where k is the original curvature feature; It is the image gradient. It is the unit normal vector. ∈ is a zero constant, ∈ = 10 -5 ; It is a divergence operator; It is the gradient magnitude; Where k is the original curvature feature; k swir The curvature geometric features of the extracted shortwave infrared image are normalized to [0,1]; τ is the curvature discrimination threshold, and τ = 0.25 is taken: curvature below τ is regarded as flat region, and curvature above τ is edge region; Step 3: Based on the extracted shortwave infrared curvature geometric feature value k swir Perform weight mapping σ(k) swir The expression is: Where, σ(k) swir ) represents the visible light weighting function for curvature modulation of shortwave infrared images; σ represents the sigmoid function; k is the slope coefficient of the function; k0 is the center offset of the function; ρ h This is an estimated fog density value.

3. The visible light image dehazing method based on short-wave infrared geometric features as described in claim 1, characterized in that: The basic fusion layer B is obtained in S3. fused The specific operating steps are as follows: The transmittance map of the visible light base layer is improved by curvature regularization optimization. The calculation formula is as follows: Where t' is the optimized transmittance map of the visible light base layer; J is the haze-free image; t is the initial transmittance estimate; I vis The acquired visible light image; ||I vis -J·t|| 2 λ represents the data fidelity term; λ is the overall strength of regularization. The gradient of the haze-free image J; γ is the regularization weight coefficient; k swir Curvature geometry features of the extracted shortwave infrared image; For the curvature regularization term: using k swir Controlling the regularization strength, k swir Small, Suppress transmittance noise; k swir big, Protect the edges of the transmittance; The calculation formula for the grassroots weakening method is: Among them, B' vis B is the base optimization layer for visible light images. vis This serves as the base layer for the decomposed visible light image; Γ(B vis / t',k swir ) is the adaptive attenuation function; t' is the optimized transmittance map of the visible light base layer; k swir The curvature geometry features of the extracted shortwave infrared image; α is an adjustable attenuation coefficient that controls the influence of curvature on the attenuation intensity, α∈[0,1], with a default value of 0.3: in k swir Large regions, exponent 1-α·(1-k) swir Approaching 1 preserves edge details; while at k swir In small regions, the exponent is close to 1-α, which suppresses texture noise and achieves curvature-guided adaptive decay; Based on weight mapping σ(k) swir The image fusion calculation formula is as follows: B fused =σ(k swir )·B' vis Among them, B fused This serves as the base layer for the fused visible light image; σ(k swir B' is the visible light weighting function for curvature modulation of shortwave infrared images; vis This is the base optimization layer for visible light images.

4. The visible light image dehazing method based on short-wave infrared geometric features as described in claim 1, characterized in that: The detail blending layer D is obtained in S4. fused The specific operating steps are as follows: By extracting the detail layer D of short-wave infrared images swir The formula for calculating the orientation-sensitive mask M(θ) constructed from the gradient direction angle geometric feature θ is as follows: θ=arctan2(G y ,G x ) Where * is defined as the convolution operation; D swir For the detailed layer of the decomposed shortwave infrared image; G x and G y The first and second matrices represent the gradient components of the shortwave infrared detail layer image in the horizontal and vertical directions, respectively; the first and second matrices represent the Sobel operator kernels in the x and y directions, respectively; θ is the gradient direction angular geometric feature of the shortwave infrared image detail layer; M(θ) is a direction-sensitive mask function used to enhance horizontal and vertical edges and suppress noise at diagonal edges; the three parameters 1.2, 0.8, and 0.3 represent the enhancement, moderate enhancement, and suppression factors required when θ is within the angle range, respectively, and are represented from top to bottom as: horizontal edge enhancement coefficient, vertical edge base coefficient, and diagonal texture suppression coefficient; 22.5° is the angle classification threshold to ensure that there is no overlap in classification between horizontal and vertical edges; Combined with fog density estimation ρ h The calculation formula for using a detail compensation method to adaptively adjust the injection intensity β to optimize texture details is as follows: β=0.1(1+0.5ρ h ) Where β is the gradient injection intensity, which controls the feature compensation intensity of the detail layer in the shortwave infrared image, and is determined according to the fog concentration ρ. h Adaptively adjust the injection intensity β, β∈[0.1,0.15]; ρ h The value is the estimated fog density. Image fusion based on weight mapping is performed to obtain the detail fusion layer D. fused The calculation formula is as follows: D’ vis =I vis -B’ vis D fused =(1-σ(k swir ))·D' vis +β·M(θ)·D swir Among them, D' vis A detail optimization layer for visible light images; D fused For the detail layer of the fused visible light image; σ(k swir ) represents the visible light weighting function for curvature modulation of shortwave infrared images; k swir The curvature geometry features of the extracted shortwave infrared image; β·M(θ)·D swir For detailed compensation amount.

5. The visible light image dehazing method based on short-wave infrared geometric features as described in claim 1, characterized in that: In S5, B will be... fused and D fused The synthesized visible light image I is obtained after dehazing. out The calculation formula is as follows: I out =B fused +D fused Among them, I out B is the final output image for dehazing visible light. fused This serves as the base layer for the fused visible light image; D fused This is the detail layer of the fused visible light image.

Images