Extremely low illumination tobacco image enhancement method and system based on multi-scale fusion

Abstract

The invention discloses an ultra-low illumination tobacco leaf image enhancement method and system based on multi-scale fusion, and belongs to the technical field of tobacco leaf image enhancement. The method comprises the following steps: S1, acquiring extremely low illumination tobacco leaf image data; s2, splitting the colored ultra-low illumination tobacco leaf image into R, G and B color channels; s3, enabling the fuzzy region to be clearer through a histogram equalization method, and obtaining a first image; s4, adapting to different illumination conditions by using a dual Gamma correction algorithm, and improving visual consistency to obtain a second image; s5, using an improved CLAHE algorithm to retain channels and collaterals of the flue-curing barn tobacco leaf image, and improving the image quality to obtain a third image; and S6, using a detail enhancement algorithm to improve specific area details of the curing barn tobacco leaf image, and retaining detail features to finally obtain a curing barn tobacco leaf curing enhancement image. Through the combination of various algorithms, the tobacco leaf image can be effectively enhanced even under the extremely low illumination conditions of serious degradation such as low signal-to-noise ratio, detail information loss and color distortion.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco leaf image enhancement, and in particular to an extremely low-light tobacco leaf image enhancement method and system based on multi-scale fusion. Background Art

[0002] At present, image enhancement technology is widely used in tobacco production, such as extremely low-light images for nighttime tobacco monitoring, low-light images for tobacco agricultural monitoring in rainy weather, extremely low-light images for nighttime cigar drying in air-drying rooms, and extremely low-light images for tobacco baking in flue-curing rooms, in order to enhance extremely low / low-light images in tobacco production. With the continuous breakthroughs in zero-sample or few-sample image enhancement technology, it has become feasible to achieve extremely low-light image enhancement for tobacco baking in flue-curing rooms under data scarcity conditions. However, the implementation of extremely low-light image enhancement in flue-curing rooms still faces many technical difficulties: the enhanced images have quality problems such as detail loss and blurring, low contrast, etc., and noise amplification is easily caused when improving image brightness. At the same time, color distortion and color cast will also occur. In addition, although the image enhancement algorithm based on deep learning has excellent results, it requires a large amount of high-quality training data, and it is difficult to obtain a matching training sample set in the actual flue-curing room environment. Therefore, this brings problems to the enhancement of extremely low-light images of tobacco baking in flue-curing rooms. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a method and system for extremely low-light tobacco leaf image enhancement based on multi-scale fusion.

[0004] The objective of the present invention is achieved through the following technical solutions: The first aspect of the present invention provides: a method for enhancing an extremely low-light tobacco leaf image based on multi-scale fusion, comprising the following steps: S1: Acquire extremely low light tobacco leaf image data; S2: Split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; S3: improving the global contrast of the extremely low-light tobacco leaf image by a histogram equalization method, enhancing the overall contrast of the image, and making the blurred area clearer to obtain a first image; S4: The second image is obtained by using a dual gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency; S5: using the improved CLAHE algorithm to adapt the local contrast of the second image, retaining the meridians of the tobacco leaf image in the flue-curing room, and improving the image quality to obtain the third image; S6: Use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features and finally obtain the flue-curing tobacco leaf baking enhanced image.

[0005] Preferably, the dual gamma correction algorithm comprises the following steps: Normalize the first image; then calculate the statistical features to obtain the average brightness and standard deviation; calculate the dark part Gamma value r based on the average brightness and standard deviation dark , bright part Gamma value r bright ; r dark = γ*(1-α d *μ+β d *σ); r bright = γ*(1+α b *μ +β b *σ); μ represents the average brightness of the image; σ represents the standard deviation of the image; α d Represents the dark brightness weight coefficient; β d Represents the dark contrast weight coefficient; α b Represents the brightness weight coefficient of the bright part; β b represents the bright contrast weight coefficient; γ is the global Gamma value; Next, perform dark mapping O dark (i) and highlight map O bright (i), then use the smooth transition function O(i) to fuse the smooth transition, generate a lookup table, and apply the lookup table to correct the second image; ; ; O(i)=w(i)*O dark (i)+(1−w(i))*O bright (i); where w(i) = 0.5 + 0.5 * cos (π * (it) / r), represents the output weight value; t represents the threshold point, that is, the dividing point between the dark and bright parts; r represents the transition area range, that is, the width of the smooth transition interval; i represents the input pixel value.

[0006] Preferably, the improved CLAHE algorithm comprises the following steps: Input the second image, convert the second image to LAB color space, separate L, A, and B channels, and calculate the L channel histogram; Performing a first block processing on the L channel histogram to obtain image blocks, setting an adaptive threshold for each image block, and performing a bilinear interpolation operation to eliminate the block effect to obtain a first image block; The first image block is divided into blocks again, and an adaptive threshold is set for each of the divided image blocks, and a bilinear interpolation operation is performed again to eliminate the block effect to obtain a third image.

[0007] Preferably, the detail enhancement algorithm comprises the following steps: Inputting a third image, and performing image preprocessing on the third image, including image normalization, and calculating image statistical features and adaptive parameters; Perform image detail extraction on the preprocessed image, including multi-scale Gaussian filtering, detail layer separation, weight calculation and detail enhancement; Perform image enhancement and optimization on the image after detail extraction, including local contrast enhancement, edge detection enhancement and adaptive fusion; The enhanced and optimized image is processed in detail, including bilateral filtering optimization, noise suppression and result normalization, to finally obtain the enhanced image of tobacco leaves baking in the flue-curing barn.

[0008] The second aspect of the present invention provides: an extremely low-light tobacco leaf image enhancement system based on multi-scale fusion, which is used to implement any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion, comprising: An image acquisition module, used to acquire extremely low-light tobacco leaf image data; An image splitting module is used to split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; The image enhancement module is used to improve the global contrast of extremely low-light tobacco leaf images through a histogram equalization method, enhance the overall contrast of the image, and make the blurred area clearer to obtain a first image; by using a dual Gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency to obtain a second image; use an improved CLAHE algorithm to adapt to the local contrast of the second image, retain the meridians of the flue-curing tobacco leaf image, and improve the image quality to obtain a third image; use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features, and finally obtain a flue-curing tobacco leaf baking enhanced image.

[0009] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion is implemented.

[0010] A fourth aspect of the present invention provides: a computer program product comprising instructions, which, when executed on a terminal, enables the terminal to execute any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion.

[0011] The beneficial effects of the present invention are: 1) Through the combination of improved histogram equalization algorithm, detail enhancement, adaptive gamma correction, and CLAHE contrast enhancement, the tobacco leaf image can be effectively enhanced even under extremely low light conditions with severe degradation such as low signal-to-noise ratio, loss of detail information, and color distortion, providing clear and reliable image data support for the automated monitoring of the tobacco leaf baking process.

[0012] 2) It can effectively solve the image quality problem caused by insufficient light in the baking room environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of the extremely low-light tobacco leaf image enhancement method based on multi-scale fusion; Figure 2 This is the flow chart of the dual gamma correction algorithm; Figure 3 It is the flow chart of the improved CLAHE algorithm; Figure 4 Enhance the algorithm flow chart for details. DETAILED DESCRIPTION

[0014] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0015] See also Figure 1-Figure 4 The first aspect of the present invention provides: a method for enhancing an extremely low-light tobacco leaf image based on multi-scale fusion, comprising the following steps: S1: Acquire extremely low light tobacco leaf image data; S2: Split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; S3: improving the global contrast of the extremely low-light tobacco leaf image by a histogram equalization method, enhancing the overall contrast of the image, and making the blurred area clearer to obtain a first image; S4: The second image is obtained by using a dual gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency; S5: using the improved CLAHE algorithm to adapt the local contrast of the second image, retaining the meridians of the tobacco leaf image in the flue-curing room, and improving the image quality to obtain the third image; S6: Use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features and finally obtain the flue-curing tobacco leaf baking enhanced image.

[0016] In this embodiment, the application limitations of traditional low-light image enhancement technology in the complex environment of a baking room are effectively overcome by adopting the technical route of a zero-sample adaptive enhancement algorithm. This method significantly improves the key indicators of image quality while maintaining computational efficiency: a higher signal-to-noise ratio is achieved to effectively suppress image noise, ensure the clear presentation of detail information, and ensure accurate color restoration. The enhanced tobacco leaf image not only has excellent intuitive effects, but also provides a high-quality data foundation for the subsequent intelligent monitoring and analysis of the tobacco leaf baking process. The image enhancement effect of this method under extremely low light conditions is significantly better than that of the existing technology, especially in an environment such as a baking room with complex and changeable lighting conditions, and can perform extremely low-light image enhancement, overcoming the defects of existing low-light image enhancement technology.

[0017] In some embodiments, the dual gamma correction algorithm includes the following steps: Normalize the first image; then calculate the statistical features to obtain the average brightness and standard deviation; calculate the dark part Gamma value r based on the average brightness and standard deviation dark , bright part Gamma value r bright ; r dark = γ*(1-α d *μ+β d *σ); r bright = γ*(1+α b *μ +β b *σ); μ represents the average brightness of the image; σ represents the standard deviation of the image; α d Represents the dark brightness weight coefficient; β d Represents the dark contrast weight coefficient; α b Represents the brightness weight coefficient of the bright part; β b represents the bright contrast weight coefficient; γ is the global Gamma value; Next, perform dark mapping O dark (i) and highlight map O bright (i), then use the smooth transition function O(i) to fuse the smooth transition, generate a lookup table, and apply the lookup table to correct the second image; ; ; O(i)=w(i)*O dark (i)+(1−w(i))*O bright (i); where w(i) = 0.5 + 0.5 * cos (π * (it) / r), represents the output weight value; t represents the threshold point, that is, the dividing point between the dark and bright parts; r represents the transition area range, that is, the width of the smooth transition interval; i represents the input pixel value.

[0018] In some embodiments, the improved CLAHE algorithm comprises the following steps: Input the second image, convert the second image to LAB color space, separate L, A, and B channels, and calculate the L channel histogram; Performing a first block processing on the L channel histogram to obtain image blocks, setting an adaptive threshold for each image block, and performing a bilinear interpolation operation to eliminate the block effect to obtain a first image block; The first image block is divided into blocks again, and an adaptive threshold is set for each of the divided image blocks, and a bilinear interpolation operation is performed again to eliminate the block effect to obtain a third image.

[0019] In this embodiment, the specific calculation formula includes: 1) Converting the RGB color space into the LAB color space: Converting the night image of the cigar I gamma (x,y) is split into I gamma_B (x,y),I gamma_G (x,y),I gamma_R (x,y) three channels. gamma (x,y)=[I gamma_B (x,y), I gamma_G (x,y), I gamma_R (x,y)].

[0020] 2) RGB Normalization: I gamma_B' (x,y)=I gamma_B (x,y) / 255; I gamma_G' (x,y)= I gamma_G (x,y) / 255; I gamma_R' (x,y)= I gamma_R (x,y) / 255; 3) Non-linear RGB to linear RGB conversion: ; Where cgamrgb is represented by {gamma_R, gamma_G, gamma_B}; Represents linear RGB values.

[0021] 4) Direct conversion matrix from linear RGB to LAB: ; Nonlinear transfer function f(t): ; fx = f(X'); fy = f(Y'); fz = f(Z'); Calculate LAB: L = 116×fy – 16; a = 500×(fx - fy); b = 200×(fy - fz).

[0022] 5) Separate the Lab channel into L channel, a channel, and b channel.

[0023] 6) Use local histogram equalization for the L channel.

[0024] Among them, L lab_out1 (x, y) represents the equalization result of the image using the local histogram algorithm; L lab_in (x, y) represents the local value of the input L channel map; max(L lab_in (x,y)) and min(L lab_in (x, y)) represent the maximum L channel value and the minimum L channel value in the current local area respectively.

[0025] 7) Use the Laplacian operator on the L channel to calculate the variance of the Laplacian value of the L channel in the previous step: ; Among them, lap var represents the variance of the Laplace operator; ∇² represents the Laplace operator; L lab_out1 (x, y) represents the image of the L channel local histogram equalization in the previous step; Var represents the variance calculation function.

[0026] Laplace operator calculation formula: ;in, Represents the Laplace operator calculation formula.

[0027] Laplace variance calculation formula: ; Among them, N lab It represents the total number of pixels of the image with local histogram equalization of the L channel; μ represents the mean of the Laplacian operator.

[0028] 8) Limit processing: cliplimit=min(40.0,2.0+α lab ×ln(1 + lap var )); among them, clip limit Indicates the contrast limit value; α lab Indicates the scaling factor; lap var represents the Laplace operator variance.

[0029] 9) Improved CLAHE: Using local histogram equalization for the L channel: Among them, L lab_out2(x,y) represents the pair L lab_out1 Image equalization result; max(L lab_out1 (x,y)) and min(L lab_out1 (x, y)) represent the maximum L channel value and the minimum L channel value in the current local area respectively.

[0030] 10) Use bilinear interpolation to eliminate block effects on the L channel: ;in, x' and y' Respectively represent L lab_out2 (x,y) the x-coordinate and y-coordinate after affine transformation; represents the affine transformation matrix; x and y Respectively represent L lab_out2 The x- and y-coordinates of (x,y).

[0031] 11) Merger Channel, a channel, b channel, convert the Lab color channel into an image of the RGB color channel I enhance_clahe (x,y).

[0032] In some embodiments, the detail enhancement algorithm comprises the following steps: Inputting a third image, and performing image preprocessing on the third image, including image normalization, and calculating image statistical features and adaptive parameters; Perform image detail extraction on the preprocessed image, including multi-scale Gaussian filtering, detail layer separation, weight calculation and detail enhancement; Perform image enhancement and optimization on the image after detail extraction, including local contrast enhancement, edge detection enhancement and adaptive fusion; The enhanced and optimized image is processed in detail, including bilateral filtering optimization, noise suppression and result normalization, to finally obtain the enhanced image of tobacco leaves baking in the flue-curing barn.

[0033] In this embodiment, the specific calculation formula includes: (1) the output image I of the previous step is enhance_clahe (x,y) is normalized to obtain image I img_normal (x,y).

[0034] (2) Calculation I img_normal The mean μ of the (x,y) image img_normal and standard deviation σ img_normal .

[0035] (3) Gaussian filtering: ; Among them, I gauss (x,y,σi mg_normal) indicates that in the original normalized I img_normal (x, y) is the image after Gaussian transformation; σ img_normal I img_normal The standard deviation of the (x,y) image.

[0036] (4) Adaptive weight, difference calculation formula: ; Mean calculation formula: ; where μ diff Represents the difference image I diff The mean of (x,y); N diff Represents the difference image I diff The total number of pixels at (x,y).

[0037] Local standard deviation formula: ; Among them, σ diff Represents the difference image I diff (x, y) is the standard deviation of the local area; M and N represent the difference image I diff The width and height of (x,y).

[0038] Adaptive weight formula: ; Among them, ω detail Represents the difference image I diff Adaptive weight of (x,y); σ diff Represents the difference image I diff (x,y) standard deviation of the local area; σ img_normal I img_normal The standard deviation of the (x,y) image.

[0039] (5) Multi-scale detail extraction, fine-scale detail calculation formula: Basic scale detail calculation formula: Coarse scale details: Among them, I detail_f (x,y),I detail_b (x,y),I detail_c (x, y) represent the fine scale detail map, basic scale detail map, and coarse scale detail map respectively; I gauss_f (x,y,σ img_normal ), I gauss_b (x,y,σ img_normal ), I gauss_c (x,y,σ img_normal ) represent the img_normalThe (x,y) image is Gaussian blurred.

[0040] (6) Edge strength, edge strength calculation formula: ; Among them, E edge (x,y) represents the edge strength; G x I img_normal (x,y) is the gradient of the image in the x direction; G y I img_normal (x,y) The gradient of the image in the y direction.

[0041] Edge strength weight calculation formula: ; Among them, ω edge (x,y) represents the edge weight; ε represents the numerical stability factor.

[0042] (7) Local contrast enhancement, the enhancement coefficient calculation formula is: k = (w-1) / 2, where w and h are the mean filter window sizes (15, 15) respectively. Local mean calculation: ; Among them, I local_contrast (x,y) means The local mean of the image; N local Represents the number of pixels in the w×h window.

[0043] Local difference calculation: ; Contrast enhancement calculation formula: Among them, I local_contrast (x,y) represents I img_normal The local mean of the (x,y) image; ω edge (x,y) represents the edge strength map.

[0044] (8) Detail fusion, detail fusion calculation formula: Among them, ω detail_1 ,ω detail_2 ,ω detail_3 Respectively represent the fine scale weight, basic scale weight, and coarse scale weight, ω detail_1 +ω detail_2 +ω detail_3 =1.

[0045] (9) Image enhancement: ; Among them, ω contrast represents the contrast edge weight; ω detail Represents detail edge weights.

[0046] The second aspect of the present invention provides: an extremely low-light tobacco leaf image enhancement system based on multi-scale fusion, which is used to implement any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion, comprising: An image acquisition module, used to acquire extremely low-light tobacco leaf image data; An image splitting module is used to split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; The image enhancement module is used to improve the global contrast of extremely low-light tobacco leaf images through a histogram equalization method, enhance the overall contrast of the image, and make the blurred area clearer to obtain a first image; by using a dual Gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency to obtain a second image; use an improved CLAHE algorithm to adapt to the local contrast of the second image, retain the meridians of the flue-curing tobacco leaf image, and improve the image quality to obtain a third image; use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features, and finally obtain a flue-curing tobacco leaf baking enhanced image.

[0047] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion is implemented.

[0048] A fourth aspect of the present invention provides: a computer program product comprising instructions, which, when executed on a terminal, enables the terminal to execute any of the above-mentioned extremely low-light tobacco leaf image enhancement methods based on multi-scale fusion.

[0049] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A method for enhancing extremely low-light tobacco leaf images based on multi-scale fusion, characterized by: The following steps are involved: S1: Acquire extremely low light tobacco leaf image data; S2: Split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; S3: improving the global contrast of the extremely low-light tobacco leaf image by a histogram equalization method, enhancing the overall contrast of the image, and making the blurred area clearer to obtain a first image; S4: The second image is obtained by using a dual gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency; S5: using the improved CLAHE algorithm to adapt the local contrast of the second image, retaining the meridians of the tobacco leaf image in the flue-curing room, and improving the image quality to obtain the third image; S6: Use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features and finally obtain the flue-curing tobacco leaf baking enhanced image.

2. The method for extremely low-light tobacco leaf image enhancement based on multi-scale fusion according to claim 1, characterized in that: The dual gamma correction algorithm comprises the following steps: Normalize the first image; then calculate the statistical features to obtain the average brightness and standard deviation; calculate the dark part Gamma value r based on the average brightness and standard deviation dark , bright part Gamma value r bright ; r dark = γ*(1-α d *μ+β d *σ); r bright = γ*(1+α b *μ +β b *σ); μ represents the average brightness of the image; σ represents the standard deviation of the image; α d Represents the dark brightness weight coefficient; β d Represents the dark contrast weight coefficient; α b Represents the brightness weight coefficient of the bright part; β b represents the bright contrast weight coefficient; γ is the global Gamma value; Next, perform dark mapping O dark (i) and highlight map O bright (i), then use the smooth transition function O(i) to fuse the smooth transition, generate a lookup table, and apply the lookup table to correct the second image; ； ； O(i)=w(i)*O dark (i)+(1−w(i))*O bright (i); where w(i) = 0.5 + 0.5 * cos (π * (it) / r), represents the output weight value; t represents the threshold point, that is, the dividing point between the dark and bright parts; r represents the transition area range, that is, the width of the smooth transition interval; i represents the input pixel value.

3. The method for extremely low-light tobacco leaf image enhancement based on multi-scale fusion according to claim 1, characterized in that: The improved CLAHE algorithm comprises the following steps: Input the second image, convert the second image to LAB color space, separate L, A, and B channels, and calculate the L channel histogram; Performing a first block processing on the L channel histogram to obtain image blocks, setting an adaptive threshold for each image block, and performing a bilinear interpolation operation to eliminate the block effect to obtain a first image block; The first image block is divided into blocks again, and an adaptive threshold is set for each of the divided image blocks, and a bilinear interpolation operation is performed again to eliminate the block effect to obtain a third image.

4. The method for enhancing tobacco leaf images at extremely low illumination levels based on multi-scale fusion according to claim 1, characterized in that: The detail enhancement algorithm comprises the following steps: Inputting a third image, and performing image preprocessing on the third image, including image normalization, and calculating image statistical features and adaptive parameters; Perform image detail extraction on the preprocessed image, including multi-scale Gaussian filtering, detail layer separation, weight calculation and detail enhancement; Perform image enhancement and optimization on the image after detail extraction, including local contrast enhancement, edge detection enhancement and adaptive fusion; The enhanced and optimized image is processed in detail, including bilateral filtering optimization, noise suppression and result normalization, to finally obtain the enhanced image of tobacco leaves baking in the flue-curing barn.

5. An extremely low-light tobacco leaf image enhancement system based on multi-scale fusion, characterized by: The method for realizing the extremely low-light tobacco leaf image enhancement method based on multi-scale fusion as claimed in any one of claims 1 to 4 comprises: An image acquisition module, used to acquire extremely low-light tobacco leaf image data; An image splitting module is used to split the color extremely low-light tobacco leaf image into three color channels: R, G, and B; The image enhancement module is used to improve the global contrast of extremely low-light tobacco leaf images through a histogram equalization method, enhance the overall contrast of the image, and make the blurred area clearer to obtain a first image; by using a dual Gamma correction algorithm to enhance the brightness and contrast of the first image, adapt to different lighting conditions, and improve visual consistency to obtain a second image; use an improved CLAHE algorithm to adapt to the local contrast of the second image, retain the meridians of the flue-curing tobacco leaf image, and improve the image quality to obtain a third image; use high-frequency filtering in the detail enhancement algorithm to enhance the leaf meridians of the third image, improve the details of specific areas of the flue-curing tobacco leaf image, retain the detail features, and finally obtain a flue-curing tobacco leaf baking enhanced image.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by the processor, the extremely low-light tobacco leaf image enhancement method based on multi-scale fusion as described in any one of claims 1 to 4 is implemented.

7. A computer program product comprising instructions, characterized in that: When the computer program product is run on a terminal, the terminal executes the extremely low-light tobacco leaf image enhancement method based on multi-scale fusion as described in any one of claims 1 to 4.

Images