CN111539936B - Mixed weight multispectral fusion method of lithium battery image - Google Patents

Mixed weight multispectral fusion method of lithium battery image Download PDF

Info

Publication number
CN111539936B
CN111539936B CN202010333856.6A CN202010333856A CN111539936B CN 111539936 B CN111539936 B CN 111539936B CN 202010333856 A CN202010333856 A CN 202010333856A CN 111539936 B CN111539936 B CN 111539936B
Authority
CN
China
Prior art keywords
image
channel
fusion
images
fused
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010333856.6A
Other languages
Chinese (zh)
Other versions
CN111539936A (en
Inventor
陈海永
唐毅强
刘卫朋
张建华
杨佳博
乞雨宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hebei University of Technology
Original Assignee
Hebei University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202010333856.6A priority Critical patent/CN111539936B/en
Publication of CN111539936A publication Critical patent/CN111539936A/en
Application granted granted Critical
Publication of CN111539936B publication Critical patent/CN111539936B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration by the use of local operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
    • G06T5/70—
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10036—Multispectral image; Hyperspectral image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20064—Wavelet transform [DWT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/10—Energy storage using batteries

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a mixed weight multispectral fusion method of a lithium battery image. Firstly, analyzing the difference of multiple defect information spectrums, collecting a group of multispectral images with specific wavelengths, and selecting the optimal five fusion wavebands and the optimal fusion space. And then, according to the difference of the representation forms of the three channels of the optimal fusion space of the multispectral defect image, each channel is independently processed, and the defect characteristics are highlighted and the complex background is restrained. The result shows that the fusion result of the method has smoother background and more prominent multi-type characteristics, enhances the defect part while inhibiting the complex background, accords with the identification of a human visual system on a defect region, has better performance, can reduce the interference of complex textures on the surface of the lithium battery piece on defect detection, and obtains clear multi-type surface defect images.

Description

Mixed weight multispectral fusion method of lithium battery image
Technical Field
The invention belongs to the field of image fusion, and particularly relates to a mixed weight multispectral fusion method of a lithium battery image.
Background
Lithium batteries are a type of batteries using a nonaqueous electrolyte solution with lithium metal or a lithium alloy as a negative electrode material. Lithium batteries can be broadly divided into two categories: lithium metal batteries and lithium ion batteries, which are metal-free lithium and rechargeable. At present, lithium ion batteries are widely used in actual daily life of people, and are commonly called lithium batteries.
The surface of the lithium battery is damaged to different degrees due to the influence of reasons such as materials, machine equipment, manual misoperation and the like in factory production, and defects such as scratches, shrinkage cavities and the like are generated. The defect information features are highly similar to the complex background features of the lithium battery, so that the detection difficulty is increased. Quality control of lithium batteries has become a major issue to be addressed.
The lithium battery image has more defects, all defects cannot be clearly presented under a single condition, the subsequent detection difficulty is high, and Anwar et al (Anwar S A, abdullah M Z. Micro-crack detection of multicrystalline solar cells featuring an improved anisotropic diffusion filter and image segmentation technique [ J ]. Eurasip Journal on Image & Video Processing,2014 (1): 1-17) propose a detection method based on defect gradient characteristics, but the detection method is easy to be interfered by complex backgrounds. Zhao Peng et al (Zhao Peng, wang Nihong, pu Zhaobang. Image fusion based on morphological wavelet decomposition pyramid [ J ]. Photoelectron laser, 2008,19 (6): 814-817.) image fusion was performed by combining wavelet decomposition with image contrast pyramid using the morphological wavelet decomposition pyramid method, but the fused image failed to suppress local noise generation.
Therefore, the difficulty brought by the current method can be effectively improved and solved, and the method is an intensive research work of professionals. Under the condition, a mixed weight multispectral fusion algorithm of the lithium battery image is developed, the influence of a complex background is weakened, noise is suppressed, defect characteristics are enhanced, the contrast ratio of a defect part and a non-defect part is improved, the difficulty of subsequent detection and classification is reduced to the greatest extent, and the lithium battery image mixed weight multispectral fusion algorithm has high practical value.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a mixed weight multispectral fusion method of a lithium battery image. According to the method, a multispectral imaging mode is adopted, the multispectral image with fixed wavelength is obtained by utilizing the spectrum range difference of the defect type, the influence of a complex background is weakened by an image mixing weight fusion method, the defect characteristics are enhanced, and the detection difficulty is reduced to the greatest extent.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a mixed weight multispectral fusion method of lithium battery images comprises the following steps:
the first step: multispectral lithium battery image acquisition:
respectively acquiring images of the same lithium battery in 380nm-505nm wave bands, 530nm-630nm wave bands and 655nm-780nm wave bands to form a multispectral RGB battery source image, and acquiring at least one image of the lithium battery with corresponding wavelength in each wave band;
and a second step of: RGB conversion HSV:
converting the multispectral RGB battery source images with different wave bands selected in the first step from RGB space to HSV space, and then respectively processing H, S, V three-channel images independently;
the conversion relation from RGB to HSV is as follows:
Figure BDA0002465900940000021
wherein X, Y, Z is chromaticity coordinates, R, G, B is the color representing three channels of red, green and blue;
and a third step of: fusing the S channel images:
3-1) decomposing the S channel image obtained in the second step into a detail layer sub-image and an approximate layer sub-image by using empirical wavelet decomposition;
3-2) generating JND model-based saliency weights
Defining significance weights J (x, y) for the S channel images obtained in the second step by using a nonlinear additive JND model, and extracting pixels with good saturation and contrast in different spectrum images;
Figure BDA0002465900940000022
wherein
Figure BDA0002465900940000023
and />
Figure BDA0002465900940000024
Is background brightness masking and image visible texture masking, K l,t (0＜K l,t < 1) represents an overlap of masking effects; x, y represent chromaticity coordinates; i=1 to N, N being the number of input source image sequences;
wherein the background brightness is masked
Figure BDA0002465900940000025
The definition is as follows:
Figure BDA0002465900940000026
Figure BDA0002465900940000027
Figure BDA0002465900940000028
wherein I (x, y) is an intensity value, B (a, B) is a low-pass weight filter, and (a, B) represents the coordinate position of the low-pass weight filter B;
defining texture masking using edge differences
Figure BDA0002465900940000029
Figure BDA00024659009400000210
/>
Figure BDA00024659009400000211
wherein ,
Figure BDA00024659009400000212
is a convolution descriptor, g k (x, y) is the kth directional high pass filter, defined as:
Figure BDA0002465900940000031
Figure BDA0002465900940000032
3-3) generating saturation weight for the multispectral RGB battery source image obtained in the first step, and calculating saturation weight S according to standard deviation of R, G, B three channels i (x,y)：
Figure BDA0002465900940000033
wherein ,
Figure BDA0002465900940000034
rxy, gxy, bxy represents color saturation on three channels R, G, B, respectively;
3-4) combining the saturation weight and the significance weight and normalizing the two to obtain a mixed weight W i (x, y), specifically defined as:
Figure BDA0002465900940000035
wherein Si (x, y) represents saturation weight, J i (x, y) represents JND saliency weight;
3-5) fusing the detail layer sub-images obtained in the step 3-1) by using a mixed weight according to the following rule to obtain fused detail layer sub-images,
Figure BDA0002465900940000036
wherein R (x, y) represents the pixel value of the coordinate position of the mixed weight fusion result, D i (x, y) represents a layer of detail layer pixel values of the empirical wavelet decomposition of the ith S-channel image;
3-6) carrying out absolute value large fusion on the approximate layer sub-image obtained in the step 3-1) to obtain a fused approximate layer sub-image;
3-7) fusing the fused approximate layer sub-image and the fused detail layer sub-image to obtain a fused S-channel image;
fourth step: fusion of H-channel images:
the H channel image contains tone information, gaussian filtering is firstly carried out on the H channel image to reduce the influence of noise, and then absolute value large fusion is carried out to obtain a fused H channel image;
fifth step: fusion of V-channel images:
discarding V channel images in 380nm-505nm wave bands and 530nm-630nm wave bands, only fusing V channel images in 655nm-780nm wave bands, firstly performing Gaussian filtering on the V channel images to reduce the influence of noise, and then performing selective absolute value large fusion to obtain fused V channel images;
sixth step: HSV image conversion after fusion:
and finally, fusing the fused V-channel image, the fused H-channel image and the fused S-channel image to obtain a fused HSV image, and converting the HSV image back to an RGB image to obtain a final fused lithium battery image.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, according to the difference of defect information spectrum presentation, the mixed weight multispectral fusion method of the lithium battery image is provided, each channel is independently processed according to the difference of the representation forms of three channels of the multispectral defect image optimal fusion space, the defect characteristics are highlighted, the complex background is restrained, the recognition of the defect area by the human visual system is met, and the method has good performance.
The multispectral mixed weight fusion method of the lithium battery image can meet the requirement of detecting the subsequent defects on the surface of the electrode of the lithium battery in the actual lithium battery industrial production, improve the accurate extraction efficiency of the surface defects, effectively prevent unqualified lithium batteries from flowing into the subsequent links, and further improve the quality and efficiency of quality inspection, and further improve the quality and production efficiency of the lithium battery module.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a color matching function.
Fig. 3 is an RGB chromaticity diagram.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention discloses a mixed weight multispectral fusion method of a lithium battery image, which comprises the following steps:
the first step: multispectral lithium battery image acquisition:
according to the invention, five wavelengths of 380nm, 605nm, 630nm, 655nm and 680nm are selected as the fusion source images of the lithium battery to be fused, the reflection of the 380nm-505nm wave band images is strong, and certain surface information is reserved; the image with the wave band of 530nm-630nm has higher brightness and saturation, the image layering property is better under the two wavelengths of 605nm and 630nm, and the defect characteristics are most obvious; the 655nm-780nm wave band image contains lower energy, weaker brightness and most obvious image defect characteristics at two wavelengths of 655nm and 680 nm.
And a second step of: RGB conversion HSV
Colorimetry is the science of researching visual laws of colors, color measurement theory and technology, and is based on the colorimetry system specified by the same international protocol to perform physical measurement on colors, and the international commission on illumination (CIE) has established a color matching function in 1931:
Figure BDA0002465900940000051
wherein c (lambda) is a color matching function, and the numbers of the three primary colors of red, green and blue corresponding to the isoelectric spectral color with the wavelength lambda are recorded as r, g and b.
The RGB to wavelength conversion relationship is:
Figure BDA0002465900940000052
wherein x, y and z are chromaticity coordinates, and R, G, B is the color representing three channels of red, green and blue.
According to the invention, an image quality evaluation index of SSIM (structural similarity) is selected to measure the relation among three channels of each color space, and compared with two spaces of RGB and YCbCr, the HSV space has the smallest similarity, that is, the difference of information contained in each channel is larger.
And a third step of: fusing the S channel images:
3-1) decomposing the S channel image obtained in the second step into a detail layer sub-image and an approximate layer sub-image by using empirical wavelet decomposition;
the empirical wavelet decomposition is a self-adaptive method, which generates a set of filters according to the characteristics of the signal itself, and filters the S-channel image of the multispectral lithium battery by using the generated filters to obtain detailed layer and approximate layer sub-images. The process of empirical wavelet decomposition is:
a constructing a line filter, performing fast Fourier transform on each line of the S channel image to obtain
Figure BDA0002465900940000053
Taking the mean F of the Fourier spectrum of all the rows of the current image row ：
Figure BDA0002465900940000054
wherein ,Nrow The number of lines of the S channel image;
b determining the number of filters according to the designated No value by using an edge detection method, taking the median between two adjacent maximum values of the first No of the Fourier spectrum of the signal, and determining the No-1 edges omega of the divided spectrum n (1.ltoreq.n.ltoreq.No-1) while assuming ω 0 ＝0，ω No ＝π；
C constructing an empirical scale function
Figure BDA0002465900940000061
Figure BDA0002465900940000062
D constructing a group of empirical wavelets
Figure BDA0002465900940000063
/>
Figure BDA0002465900940000064
ω represents frequency and β can be any [0,1] function
E through empirical scale function
Figure BDA0002465900940000065
And empirical wavelet->
Figure BDA0002465900940000066
Calculating to obtain a line filter bank
Figure BDA0002465900940000067
wherein ,NR Row represents the row for the number of wavelet functions in the row filter bank, where no=n R 。
F repeating the steps 3-1 to 3-5, processing each column of the image, and constructing a column filter
Figure BDA0002465900940000068
wherein ,Nc For the number of wavelet functions in the column filter bank, col represents the column, where no=nc.
G filtering the image according to the use of the filter in the line filter bank to obtain N R +1 intermediate results, each of which is then filtered using a column filter to obtain the final (N R +1)*(N c +1) subband images.
For simplicity purposes it can be assumed that τ n ＝λω n (0 < lambda < 1) so that only one parameter lambda is required, instead of assigning tau to the scale function and each wavelet function n 。
3-2) JND model-based saliency weighting of S-channel image generation
JND refers to the Human Visual System (HVS) minimum perceived difference and defines a perception threshold to guide perceived image quality metric masking. It is an important visual saliency cue for the measurement of image quality. The present invention defines a saliency weight J (x, y) using a nonlinear addable JND model that is capable of extracting pixels with good saturation and contrast in different spectral images.
Figure BDA0002465900940000069
wherein
Figure BDA00024659009400000610
and />
Figure BDA00024659009400000611
Background luminance masking and image visible texture masking. K (K) l,t (0＜K l,t < 1) represents the overlap of masking effects, x, y represents chromaticity coordinates.
The invention uses two different functions to model brightness
Figure BDA0002465900940000071
And average background brightness. One is the root equation for low background luminance (less than 127) and the other is the approximately linear equation for the higher 127 portion, defined as:
Figure BDA0002465900940000072
Figure BDA0002465900940000073
/>
Figure BDA0002465900940000074
where I (x, y) is an intensity value, B (a, B) is a low-pass weight filter, and (a, B) represents the coordinate position of the low-pass weight filter B, and 1.ltoreq.a, b.ltoreq.5.
Texture masking is typically determined by the local spatial gradient around the pixel. To obtain a better alignmentThe edge and non-edge regions should be well distinguished for a deterministic JND estimation. The invention is defined taking into account edge differences, since the HVS draws more attention to the edge structure
Figure BDA0002465900940000075
Figure BDA0002465900940000076
Figure BDA0002465900940000077
wherein ,
Figure BDA0002465900940000078
is a convolution descriptor, g k (x, y) is the kth directional high pass filter, defined as:
Figure BDA0002465900940000079
Figure BDA00024659009400000710
3-3) generating saturation weights for multispectral lithium battery raw images
The saturation weight is considered to better preserve the saturation information of the source image sequence, so that the saturation of the fusion result is improved, the saturation weight is generated for the original image of the multispectral lithium battery, and the saturation weight can be calculated according to the standard deviation of R, G, B three channels:
Figure BDA0002465900940000081
wherein Si For the color saturation of the image,
Figure BDA0002465900940000082
rxy, gxy, bxy represents color saturation on three channels R, G, B, respectively;
3-4): mixing weights
The detail layer capacity is smaller, defect information is contained, and the mixed weight generated by the saturation weight and the saliency weight is used for fusion; in order to obtain a mixed weight of the ith Zhang Yuan image sequence considering the saturation weight and the significance weight, the method combines the two weights and normalizes the two weights to obtain the mixed weight:
Figure BDA0002465900940000083
wherein Wi (x, y) represents the mixed weight, S i (x, y) represents saturation weight, J i (x, y) represents JND saliency weight.
3-5) fusion rules
Figure BDA0002465900940000084
Wherein R (x, y) represents the pixel value of the coordinate position of the mixed weight fusion result, D i (x, y) represents a layer of detail layer pixel values of the empirical wavelet decomposition of the ith S-channel image; n is the number of input source image sequences.
3-6) carrying out absolute value large fusion on the sub-images of the approximate layer
The approximate layer capacity occupies a relatively large area, reflects the characteristics of the whole texture structure, and selects the absolute value to take the large fusion strategy, so that the edge strength after fusion is high, and the texture is clear.
Figure BDA0002465900940000085
wherein ,(mS ,n S ) For the S-channel coordinate position, c S (m S ,n S ) Representing the pixel values after the approximate layer fusion,
Figure BDA0002465900940000086
Figure BDA0002465900940000087
the pixel values of the five wavelength approximate layer sub-images respectively representing 380nm, 605nm, 630nm, 655nm and 680nm are fused at the corresponding positions of each picture;
3-7) fusing the fused approximate layer sub-image and the fused detail layer sub-image to obtain a fused S-channel image, wherein the fusion can be performed in a direct addition mode.
Fourth step: h channel fusion
The H channel contains tone information, gaussian filtering is firstly carried out on the tone information, the influence of noise is reduced, and then absolute value maximization fusion is selected.
Figure BDA0002465900940000091
wherein ,(mH ,n H ) For H channel coordinate position, c H (m H ,n H ) Representing the pixel values of the H-channel after fusion,
Figure BDA0002465900940000092
Figure BDA0002465900940000093
representing pixel values for five wavelength H-channel images at 380nm, 605nm, 630nm, 655nm, and 680nm, respectively.
Fifth step: v channel fusion
Because the complex background exists in the form of brightness information, in order to weaken the influence of the complex background on subsequent defect detection, the V channel discards images at three wavelengths of 380nm, 605nm and 630nm, only fuses images at two wavelengths of 655nm and 680nm, firstly carries out Gaussian filtering on the images to reduce the influence of noise, and then carries out selective absolute value large fusion.
Figure BDA0002465900940000094
wherein ,,(mV ,n V ) For the V-channel coordinate position, c V (m v ,n v ) Representing the pixel values of the fused V-channel,
Figure BDA0002465900940000095
Figure BDA0002465900940000096
representing pixel values for two wavelength V-channel images at 655nm and 680nm, respectively.
Sixth step: HSV image conversion after fusion
And finally, fusing the fused V-channel image, the fused H-channel image and the fused S-channel image to obtain a fused HSV image, and converting the HSV image back to an RGB image to obtain a final fused lithium battery image.
Examples
The specific flowchart of the multi-resolution mixed weight fusion method of the multispectral lithium battery image in the embodiment is shown in fig. 1, and the method comprises the following steps:
a mixed weight multispectral fusion method of lithium battery images comprises the following steps:
the first step: multispectral lithium battery image acquisition:
according to the invention, five wavelengths of 380nm, 605nm, 630nm, 655nm and 680nm are selected as the fusion source images of the lithium battery to be fused, the reflection of the 380nm-505nm wave band images is strong, and certain surface information is reserved; the image with the wave band of 530nm-630nm has higher brightness and saturation, the image layering property is better under the two wavelengths of 605nm and 630nm, and the defect characteristics are most obvious; the 655nm-780nm wave band image contains lower energy, weaker brightness and most obvious image defect characteristics at two wavelengths of 655nm and 680 nm.
And a second step of: RGB conversion HSV
The commonly used color spaces are RGB, HSV and YCbCr, the invention selects the image quality evaluation index of SSIM (structural similarity) to measure the relation among three channels of each color space, and the comparison in the table 1 shows that the HSV space has the minimum similarity compared with the other two spaces, that is, the difference of the information contained in each channel is larger, so the HSV space is selected for the next fusion processing, the lithium battery multispectral images of different wave bands selected in the first step are converted into the HSV space from the RGB space, and then the H, S, V three-channel images are processed independently.
TABLE 1 comparison of structural similarity of three-channel images in the same color space
Figure BDA0002465900940000101
Colorimetry is the science of researching visual laws of colors, color measurement theory and technology, and is based on the colorimetry system specified by the same international protocol to perform physical measurement on colors, and the international commission on illumination (CIE) has established a color matching function in 1931:
Figure BDA0002465900940000102
wherein c (lambda) is a color matching function, and the numbers of the three primary colors of red, green and blue corresponding to the isoelectric spectral color with the wavelength lambda are recorded as r, g and b.
The RGB to wavelength conversion relationship is:
Figure BDA0002465900940000103
wherein x, y and z are chromaticity coordinates, and R, G, B is the color representing three channels of red, green and blue.
And a third step of: empirical wavelet decomposition of the S-channel image:
the S channel image is decomposed into a detail layer and an approximate layer by using empirical wavelet decomposition, the empirical wavelet decomposition is a self-adaptive method, a group of filters are generated according to the characteristics of the signal, and the generated filters are used for filtering the S channel image of the multispectral lithium battery to obtain the detail layer and the approximate layer sub-image.
Fourth step: generating JND model-based saliency weight for S-channel image
JND refers to the Human Visual System (HVS) minimum perceived difference and defines a perception threshold to guide perceived image quality metric masking. It is an important visual saliency cue for the measurement of image quality. The present invention defines a saliency weight J (x, y) using a nonlinear addable JND model that is capable of extracting pixels with good saturation and contrast in different spectral images.
Figure BDA0002465900940000104
wherein
Figure BDA0002465900940000105
and />
Figure BDA0002465900940000106
Is background brightness masking and image visible texture masking, K l,t (0＜K l,t < 1) represents an overlap of masking effects; x, y represent chromaticity coordinates; i=1 to N, N is the number of input source image sequences, n=5.
The invention uses two different functions to model brightness
Figure BDA0002465900940000111
And average background brightness. One is the root equation for low background luminance (less than 127) and the other is the approximately linear equation for the higher 127 portion, defined as:
Figure BDA0002465900940000112
Figure BDA0002465900940000113
Figure BDA0002465900940000114
where I (x, y) is the intensity value and B (a, B) is the low pass weight filter.
Texture masking is typically determined by the local spatial gradient around the pixel. In order to obtain a more accurate JND estimation, the edge and non-edge regions should be well distinguished. The invention is defined taking into account edge differences, since the HVS draws more attention to the edge structure
Figure BDA0002465900940000115
/>
Figure BDA0002465900940000116
Figure BDA0002465900940000117
wherein ,
Figure BDA0002465900940000118
is a convolution descriptor, g k (x, y) is the kth directional high pass filter, defined as:
Figure BDA0002465900940000119
Figure BDA00024659009400001110
fifth step: generating saturation weights for multispectral lithium battery raw images
The saturation weight is considered to better preserve the saturation information of the source image sequence, so that the saturation of the fusion result is improved, the saturation weight is generated for the original image of the multispectral lithium battery, and the saturation weight can be calculated according to the standard deviation of R, G, B three channels:
Figure BDA00024659009400001111
wherein Si For the color saturation weight of the image,
Figure BDA0002465900940000121
sixth step: mixing weights and fusion rules
The capability of the 6-1 detail layer is smaller, defect information is contained, and the mixed weight generated by the saturation weight and the saliency weight is used for fusion; in order to obtain a mixed weight of the ith Zhang Yuan image sequence considering the saturation weight and the significance weight, the method combines the two weights and normalizes the two weights to obtain the mixed weight:
Figure BDA0002465900940000122
wherein Wi (x, y) represents the mixed weight, S i (x, y) represents saturation weight, J i (x, y) represents JND saliency weight.
6-2 fusion rules
Figure BDA0002465900940000123
Wherein R (x, y) represents the pixel value of the coordinate position of the fusion result, W i (x, y) represents a mixed weight, D i (x, y) represents a layer of detail layer pixel values of the empirical wavelet decomposition of the ith S-channel image, N being the number of input source image sequences.
Seventh step: the absolute value of the sub-images of the approximate layer is taken and fused
The approximate layer capacity occupies a relatively large area, reflects the characteristics of the whole texture structure, and selects the absolute value to take the large fusion strategy, so that the edge strength after fusion is high, and the texture is clear.
Figure BDA0002465900940000124
wherein ,cS (m S ,n S ) Representing the value of the pixel after the fusion,
Figure BDA0002465900940000125
Figure BDA0002465900940000126
pixel values representing five wavelength approximation layer sub-images of 380nm, 605nm, 630nm, 655nm, and 680nm, respectively
And fusing the fused approximate layer sub-image and the fused detail layer sub-image to obtain a fused S-channel image.
Eighth step: h channel fusion
The H channel contains tone information, gaussian filtering is firstly carried out on the tone information, the influence of noise is reduced, and then absolute value maximization fusion is selected.
Figure BDA0002465900940000127
wherein ,cH (m H ,n H ) Representing the value of the pixel after the fusion,
Figure BDA0002465900940000131
Figure BDA0002465900940000132
representing pixel values for five wavelength H-channel images at 380nm, 605nm, 630nm, 655nm, and 680nm, respectively.
Ninth step: v channel fusion
Because the complex background exists in the form of brightness information, in order to weaken the influence of the complex background on subsequent defect detection, the V channel discards images at three wavelengths of 380nm, 605nm and 630nm, only fuses images at two wavelengths of 655nm and 680nm, firstly carries out Gaussian filtering on the images to reduce the influence of noise, and then carries out selective absolute value large fusion
Figure BDA0002465900940000133
wherein ,cV (m, n) represents the pixel value of the V channel after fusion,
Figure BDA0002465900940000134
representing pixel values for two wavelength V-channel images at 655nm and 680nm, respectively.
Tenth step: HSV image conversion after fusion
And finally, converting the fused HSV image back to an RGB image to obtain a final fused lithium battery image.
The invention analyzes the fusion result of the lithium battery image and measures the performance of the method from four indexes of edge strength, average gradient, image definition and information entropy. As can be seen from Table 2, the mixed weight multispectral fusion algorithm of the lithium battery image provided by the invention is higher than other methods in four index aspects compared with other methods, has higher values than the original five wavelengths, enhances the defect part while inhibiting the complex background, and accords with the identification of the human visual system on the defect region.
Table II, method index measurement Performance
Edge strength Average gradient Image definition Information entropy
The algorithm of the invention 55.651 5.183 7.771 7.189
Laplacian pyramid 49.943 4.855 6.874 6.374
Gradient domain method 53.881 5.180 7.619 6.693
380nm 26.966 2.726 3.311 5.625
605nm 55.343 5.151 7.153 6.869
630nm 50.325 4.849 6.161 6.869
655nm 34.896 3.352 3.905 5.951
680nm 34.957 3.358 3.913 5.955
The mixed weight multispectral fusion algorithm of the lithium battery image can meet the requirement of detecting the subsequent defects on the surface of the electrode of the lithium battery in the actual lithium battery industrial production, improve the accurate extraction efficiency of the surface defects, effectively prevent unqualified lithium batteries from flowing into the subsequent links, and further improve the quality and efficiency of quality inspection, and further improve the quality and production efficiency of the lithium battery module.
While the present invention has been described with reference to the above-described embodiments, it is to be understood that the above-described embodiments are illustrative only and not limiting, and that many forms may be made by those skilled in the art without departing from the spirit of the invention and the scope of the appended claims, which are to be construed as embodying the present invention.
The invention is applicable to the prior art where it is not described.

Claims (3)

1. A mixed weight multispectral fusion method of lithium battery images comprises the following steps:
the first step: multispectral lithium battery image acquisition:
respectively acquiring images of the same lithium battery in 380nm-505nm wave bands, 530nm-630nm wave bands and 655nm-780nm wave bands to form a multispectral RGB battery source image, and acquiring at least one image of the lithium battery with corresponding wavelength in each wave band;
and a second step of: RGB conversion HSV:
converting the multispectral RGB battery source images with different wave bands selected in the first step from RGB space to HSV space, and then respectively processing H, S, V three-channel images independently;
the conversion relation from RGB to HSV is as follows:
Figure QLYQS_1
wherein X, Y, Z is chromaticity coordinates, R, G, B is the color representing three channels of red, green and blue;
and a third step of: fusing the S channel images:
3-1) decomposing the S channel image obtained in the second step into a detail layer sub-image and an approximate layer sub-image by using empirical wavelet decomposition;
3-2) generating JND model-based saliency weights
Defining significance weights J (x, y) for the S channel images obtained in the second step by using a nonlinear additive JND model, and extracting pixels with good saturation and contrast in different spectrum images;
Figure QLYQS_2
wherein
Figure QLYQS_3
and />
Figure QLYQS_4
Is background brightness masking and image visible texture masking, K l,t Representing overlap of masking effects, 0 < K l,t < 1; x, y represent chromaticity coordinates; i=1 to N, N being the number of input source image sequences;
wherein, the backgroundLuminance masking
Figure QLYQS_5
The definition is as follows:
Figure QLYQS_6
Figure QLYQS_7
Figure QLYQS_8
wherein I (x, y) is an intensity value, B (a, B) is a low-pass weight filter, and (a, B) represents the coordinate position of the low-pass weight filter B;
defining texture masking using edge differences
Figure QLYQS_9
/>
Figure QLYQS_10
Figure QLYQS_11
wherein ,
Figure QLYQS_12
is a convolution descriptor, g k (x, y) is the kth directional high pass filter, defined as:
Figure QLYQS_13
Figure QLYQS_14
3-3) generating saturation weight for the multispectral RGB battery source image obtained in the first step, and calculating saturation weight S according to standard deviation of R, G, B three channels i (x,y)：
Figure QLYQS_15
wherein ,
Figure QLYQS_16
rxy, gxy, bxy represents color saturation on three channels R, G, B, respectively;
3-4) combining the saturation weight and the significance weight and normalizing the two to obtain a mixed weight W i (x, y), specifically defined as:
Figure QLYQS_17
wherein Si (x, y) represents saturation weight, J i (x, y) represents JND saliency weight;
3-5) fusing the detail layer sub-images obtained in the step 3-1) by using a mixed weight according to the following rule to obtain fused detail layer sub-images,
Figure QLYQS_18
wherein R (x, y) represents the pixel value of the coordinate position of the mixed weight fusion result, D i (x, y) represents a layer of detail layer pixel values of the empirical wavelet decomposition of the ith S-channel image;
3-6) carrying out absolute value large fusion on the approximate layer sub-image obtained in the step 3-1) to obtain a fused approximate layer sub-image;
3-7) fusing the fused approximate layer sub-image and the fused detail layer sub-image to obtain a fused S-channel image;
fourth step: fusion of H-channel images:
the H channel image contains tone information, gaussian filtering is firstly carried out on the H channel image to reduce the influence of noise, and then absolute value large fusion is carried out to obtain a fused H channel image;
fifth step: fusion of V-channel images:
discarding V channel images in 380nm-505nm wave bands and 530nm-630nm wave bands, only fusing V channel images in 655nm-780nm wave bands, firstly performing Gaussian filtering on the V channel images to reduce the influence of noise, and then performing selective absolute value large fusion to obtain fused V channel images;
sixth step: HSV image conversion after fusion:
and finally, fusing the fused V-channel image, the fused H-channel image and the fused S-channel image to obtain a fused HSV image, and converting the HSV image back to an RGB image to obtain a final fused lithium battery image.
2. The method of claim 1, wherein in the first step, the lithium battery image is selected at five wavelengths of 380nm, 605nm, 630nm, 655nm and 680 nm; n=5.
3. The method for multi-spectral fusion of mixed weights of lithium battery images according to claim 2, wherein the rule of the absolute value maximization fusion in the step 3-6) is as follows:
Figure QLYQS_19
wherein ,(mS ,n S ) For the S-channel coordinate position, c S (m S ,n S ) Representing the pixel values after the approximate layer fusion,
Figure QLYQS_20
Figure QLYQS_21
respectively represent 380nFive wavelength approximations of the sub-images at m, 605nm, 630nm, 655nm and 680nm,
the rule of the selective absolute value large fusion in the fifth step is as follows:
Figure QLYQS_22
wherein ,(mV ,n V ) For the V-channel coordinate position, c V (m v ,n v ) Representing the pixel values of the fused V-channel,
Figure QLYQS_23
Figure QLYQS_24
pixel values representing two wavelength V-channel images of 655nm and 680nm, respectively;
the rule of the absolute value large fusion in the fourth step is as follows:
Figure QLYQS_25
wherein ,(mH ,n H ) For H channel coordinate position, c H (m H ,n H ) Representing the pixel values of the H-channel after fusion,
Figure QLYQS_26
Figure QLYQS_27
representing pixel values for five wavelength H-channel images at 380nm, 605nm, 630nm, 655nm, and 680nm, respectively. />
CN202010333856.6A 2020-04-24 2020-04-24 Mixed weight multispectral fusion method of lithium battery image Active CN111539936B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010333856.6A CN111539936B (en) 2020-04-24 2020-04-24 Mixed weight multispectral fusion method of lithium battery image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010333856.6A CN111539936B (en) 2020-04-24 2020-04-24 Mixed weight multispectral fusion method of lithium battery image

Publications (2)

Publication Number Publication Date
CN111539936A CN111539936A (en) 2020-08-14
CN111539936B true CN111539936B (en) 2023-06-09

Family

ID=71975371

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010333856.6A Active CN111539936B (en) 2020-04-24 2020-04-24 Mixed weight multispectral fusion method of lithium battery image

Country Status (1)

Country Link
CN (1) CN111539936B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109215008A (en) * 2018-08-02 2019-01-15 上海海洋大学 A kind of multispectral and panchromatic image fusion method of entirety two generations Bandelet transformation
CN110517210A (en) * 2019-07-08 2019-11-29 河北工业大学 More exposure welding zone image interfusion methods based on Haar small echo Gradient Reconstruction
AU2020100199A4 (en) * 2020-02-08 2020-03-19 Cao, Sihua MR A medical image fusion method based on two-layer decomposition and improved spatial frequency

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8411938B2 (en) * 2007-11-29 2013-04-02 Sri International Multi-scale multi-camera adaptive fusion with contrast normalization
CN108537756B (en) * 2018-04-12 2020-08-25 大连理工大学 Single image defogging method based on image fusion

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109215008A (en) * 2018-08-02 2019-01-15 上海海洋大学 A kind of multispectral and panchromatic image fusion method of entirety two generations Bandelet transformation
CN110517210A (en) * 2019-07-08 2019-11-29 河北工业大学 More exposure welding zone image interfusion methods based on Haar small echo Gradient Reconstruction
AU2020100199A4 (en) * 2020-02-08 2020-03-19 Cao, Sihua MR A medical image fusion method based on two-layer decomposition and improved spatial frequency

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Multi-exposure fusion for welding region bassed on multi-scale transform and hybrid weight;Haiyong Chen,et al;The International Journal of Advanced Manufacturing Technology;第101卷;105-117 *
非均匀纹理复杂表面缺陷多光谱图像显著性融合;陈海永;杨佳博;陈鹏;刘坤;;北京工业大学学报;46(第01期);10-16 *

Also Published As

Publication number Publication date
CN111539936A (en) 2020-08-14

Similar Documents

Publication Publication Date Title
Martinkauppi et al. Behavior of skin color under varying illumination seen by different cameras at different color spaces
CN105046653B (en) A kind of video raindrop minimizing technology and system
CN106997584A (en) A kind of haze weather image enchancing method
CN101650833B (en) Color image quality evaluation method
CN106780417A (en) A kind of Enhancement Method and system of uneven illumination image
CN111489346B (en) Full-reference image quality evaluation method and system
CN109215042B (en) Photovoltaic cell panel hot spot effect detection system based on computer vision and calculation method thereof
CN105205792A (en) Underwater image enhancement method based on brightness and chrominance separation
CN108965647A (en) A kind of foreground image preparation method and device
CN110674884A (en) Image identification method based on feature fusion
CN112419210A (en) Underwater image enhancement method based on color correction and three-interval histogram stretching
CN111738931A (en) Shadow removal algorithm for aerial image of photovoltaic array unmanned aerial vehicle
CN111539936B (en) Mixed weight multispectral fusion method of lithium battery image
Hsieh et al. Objective haze removal assessment based on two-objective optimization
CN111563866B (en) Multisource remote sensing image fusion method
CN116503259B (en) Mosaic interpolation method and system
WO2010128683A1 (en) Blue sky color detection technique
CN107424134A (en) Image processing method, device, computer-readable recording medium and computer equipment
CN114677384B (en) Solar cell coating defect detection method
Pele et al. Improving perceptual color difference using basic color terms
CN111985492B (en) Cloud identification method
Tang et al. Sky-preserved image dehazing and enhancement for outdoor scenes
CN111667446B (en) image processing method
CN108961258A (en) A kind of foreground image preparation method and device
CN112258463A (en) Photovoltaic defect feature acquisition method and system based on camera parameter adjustment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant