Image compression method and system for solar cell appearance inspection based on wavelet transform

Abstract

The present invention discloses a solar cell appearance inspection image compression method based on wavelet transform, comprising the following steps: (1) the present application adopts a haar wavelet function with good compression decomposition to decompose, and obtains a low-frequency component A, a horizontal high-frequency component H, a vertical high-frequency component V and a diagonal high-frequency component D; (2) since the important and useful information after wavelet decomposition is distributed in the low-frequency component, the high-frequency coefficient is subjected to threshold quantization processing each time; (3) after the signal obtained after denoising reduces the signal, and after reducing the interference of noise in the image, an inverse wavelet transform is adopted, which can greatly reduce the size of the image and is basically not distorted; (4) the compressed image is judged for appearance by an image analysis algorithm. The compressed image of the present application occupies less space, greatly reduces the storage space cost, and the compressed image has less loss than the original image, which can be basically ignored compared with the loss of traditional compression, and is convenient for later viewing and analysis.

Description

Technical Field

[0001] The invention relates to the field of image compression, and in particular to a solar cell appearance detection image compression method based on wavelet transform and a system thereof. Background Art

[0002] Product appearance inspection is an important part of the modern intelligent production process. Currently, the camera array basically takes pictures and then conducts appearance inspection. After the inspection is completed, the pictures are saved and processed, which not only takes up storage space, but also takes a lot of time.

[0003] AOI inspection of each cell on the AOI optical inspection machine must be carried out. China is actively building a 5GW factory, with a daily production of about 2.4 million cells. The AOI images of the segments alone require about 229TB of storage, and storing the AOI images of the finished product segments for one day consumes about 600,000 yuan of storage resources. Therefore, many factories only keep defect images, and even so, the monthly storage fee is 650,000 yuan. On this basis, people use lossy compression, but lossy compression cannot meet the storage needs. The lossy compression currently used has a compression rate of only 70% of the original, which will also result in a large amount of image data.

[0004] In summary, this application conducts analysis and discussion based on actual projects, conducts appearance inspection for new solar cell production based on Haar wavelet, processes images through transformation, and applies byond Compare to compare the original image and the restored image to verify the distortion of the compressed image, in order to break through the traditional IDE stereotype and provide a practical image compression method for solar cell appearance inspection. Summary of the invention

[0005] The present invention provides a solar cell appearance inspection image compression method and system based on wavelet transform. The method is based on the problem that the long-term storage of images generated in industrial production occupies too much storage space and the defects of existing compression methods. Through work and study experience, wavelet transform is used to compress and save images. Experiments have proved that the compression rate can be greatly improved by using wavelet transform to compress images. By comparison, it is found that the accuracy of the compressed images can reach 99.5%.

[0006] The present invention is achieved through the following technical solutions:

[0007] A solar cell appearance inspection image compression method based on wavelet transform comprises the following steps:

[0008] (1) This application uses the Haar wavelet function with better compression decomposition for decomposition, and after decomposition, a low-frequency component A, a horizontal high-frequency component H, a vertical high-frequency component V and a diagonal high-frequency component D are obtained;

[0009] (2) Since the important and useful information after wavelet decomposition is distributed in the low-frequency components, while the noise signal is distributed in the higher frequencies (horizontal high-frequency component H, vertical high-frequency component V and diagonal high-frequency component D), the high-frequency coefficients are threshold quantized each time;

[0010] (3) After the signal is reduced by noise reduction, the interference of noise in the image is reduced. Then, the inverse wavelet transform can greatly reduce the size of the image with little distortion;

[0011] (4) The compressed image is judged for its appearance through an image analysis algorithm, and feedback is given as OK / NG based on different set values. When it is detected that the image is judged to be NG, the image is saved.

[0012] As a preferred embodiment, the derivation process of the Haar wavelet function in step (1) is as follows:

[0013] Haar function For integer k, Graphics The result after shifting right or left by k,

[0014] Let V1 be the following linear combination

[0015]

[0016] V1 is a linear space with finite support and discontinuity points. The linear space composed of piecewise constants, the scaling function is The wavelet function is

[0017] The following conclusions will be drawn for the function through Haar wavelet transform:

[0018] (1) For a function f(x), if and only if f(2 n x)∈V n When f(x)∈V

[0019] (2) For a function f(x), if and only if f(2 -n x)∈V0, then f(x)∈V n

[0020] (3) For the function The standard orthogonal set that constitutes Vn,

[0021] Haar wavelet transform is a windowing process. Haar wavelet is a low-pass filter that will filter out high-frequency components.

[0022] As a preferred embodiment, in step (2), a suitable threshold is selected according to the amplitude values ​​of the coefficients of different high-frequency components.

[0023] As a preferred embodiment, the threshold used in step (2) of the present application is a soft threshold, and the threshold coefficient is selected as the mean square error of the high-frequency assignment, which ensures that there are enough valid signals passing the threshold each time, thus greatly reducing the presence of noise.

[0024] As a preferred embodiment, the inverse wavelet transform used in step (3) is Haar wavelet.

[0025] A solar cell appearance inspection image compression system based on wavelet transform comprises a visual inspection device and an acquisition system. The visual inspection device comprises a transmission mechanism which adopts a conveyor belt which is installed on a base. A gap is provided between silicon wafers to prevent the two silicon wafers from being too close to each other and affecting the imaging. The acquisition system comprises an industrial camera and a lens. The image compression system also comprises a Dell industrial computer, an M58 to C-port adapter ring, a 4-port 10 Gigabit network card, a USB3.0 hard disk extension cable, a power trigger cable, a stroboscopic light source extension cable, a power adapter, and a process light source.

[0026] As a preferred embodiment, the silicon wafer is a solar cell wafer, and the solar cell wafer is 166mm×166mm or 182mm×182mm, with an accuracy of 0.2mm, a detection speed of 6000 wafers per hour, and a spacing of 50mm between each wafer.

[0027] As a preferred embodiment, the industrial camera uses a Dushen GR4900M-M58-GH camera with a 25-megapixel camera, a focal length of 25mm lens, and a frame rate of 40MHz.

[0028] As a preferred embodiment, the Dell industrial computer adopts the 7080-ATS model, the M58 to C-port adapter ring adopts the AZURE-ADCM58 model, the 4-port 10 Gigabit network card is used to collect pictures, the USB3.0 hard disk extension cable is used to connect the hard disk, the power trigger cable adopts a 10M armored optical fiber cable, the stroboscopic light source extension cable adopts the Turing Microvision TVS-LL-HC-5 model, the power adapter adopts the Dushen ADS-26SGP-12-12024E, and the process light source adopts the Turing Microvision model BFL-296GB32-TVS.

[0029] As a preferred embodiment, the AOI image of the solar cell is 100M, and the image of the compressed finished AOI is used for analysis. The image analysis is performed in Python. CV2, numpy and pywt toolkits are referenced in Python. The CV2 toolkit includes the opening and closing functions of the image, the numpy toolkit contains the adaptive analysis function in the wavelet analysis, and the pywt toolkit contains the wavelet transform tool; the image input is subjected to wavelet transform through dwt2 in the pywt toolkit, and the haar wavelet is selected in dwt2. At this time, the haar wavelet transform in the wavelet transform is used. After the transformation, the low-frequency component, horizontal high-frequency component, vertical high-frequency component, and diagonal high-frequency component of the wavelet transform can be obtained. After obtaining the high-frequency component, noise reduction is performed according to the adaptive threshold method of the wavelet transform, and finally the horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component after noise reduction are obtained. The obtained high-frequency component has been subjected to high-frequency band filtering in the wavelet transform. The processing principle is to add a window in the wavelet transform and then obtain the compressed image through the inverse wavelet transform (idwt2), output and save or perform AOI analysis and finally save.

[0030] The invention principle of this application: This application is based on the problem of long-term storage of pictures generated in industrial production occupying too much storage space and the existing compression methods. Through work and study experience, wavelet transform is used to compress and save pictures. According to experiments, it is proved that the way of compressing pictures using wavelet transform can greatly improve the compression rate. Through comparison, it is found that the accuracy of compressed pictures can reach 99.5%. Based on this, it is proved that wavelet transform can be used in industrial compressed pictures. In summary, the method used in this article can effectively reduce the size of pictures, greatly reduce the hard disk space during the storage process, save production storage space, and reduce storage costs. The method described in this application not only compresses images but also can be used as a normal algorithm.

[0031] Beneficial effects of the present application: In view of the fact that the pictures are collected by the existing camera array, the present application adds the wavelet transform algorithm to the acquisition program to compress the images on the basis of the existing hardware without changing the hardware, thereby improving the storage utilization rate and reducing the storage cost. There is no need to change the hardware architecture, and the mature wavelet transform (haar wavelet transform) is used. The algorithm is mature and easy to implement, and the calculation process is simple and low in complexity. The compressed picture takes up less space, which greatly reduces the storage space cost. The compressed picture has less loss than the original picture, and the loss compared to the traditional compression is basically negligible, which is convenient for later viewing and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The schematic diagram of the hardware structure of a solar cell appearance inspection image compression system based on wavelet transform of the present invention.

[0033] Figure 2 The present invention is a schematic diagram of the operation flow of a solar cell appearance inspection image compression method based on wavelet transform.

[0034] Figure 3 Schematic diagram of Haar wavelet transform in the present invention.

[0035] Figure 4 The original image and the compressed image according to an embodiment of the present invention are placed on the left and right sides of the tool respectively, and a schematic diagram of the comparison result is shown.

[0036] Figure 5 It is a schematic diagram of comparing the original file and the compressed file of four-pass printing AOI image according to an embodiment of the present invention.

[0037] Figure 6 This is a schematic diagram of comparing abnormal pixels according to an embodiment of the present invention.

[0038] Figure 7 The diagram is a schematic diagram of a Beyond Compare report analysis according to an embodiment of the present invention.

[0039] Figure 8 The figure is a schematic diagram of the comparison between the original picture and the compressed picture according to an embodiment of the present invention.

[0040] Fig. 9 It is a schematic diagram of a printed AOI image source file and a compressed file according to an embodiment of the present invention.

[0041] Fig.10 FIG. 4 is a schematic diagram of pixel difference distribution according to an embodiment of the present invention.

[0042] Fig.11 This is a schematic diagram of pixel reporting after image compression according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0044] like Figure 1As shown, a solar cell appearance inspection image compression system based on wavelet transform includes visual inspection equipment and acquisition system. The visual inspection equipment includes a conveying mechanism. The conveying mechanism uses a conveyor belt 1, which is installed on a base. There is a certain gap between silicon wafers to prevent the two silicon wafers from being too close to each other and affecting the imaging. The industrial camera 2 and the lens are the most important units of the entire acquisition system and need to be selected according to the inspection requirements. At present, the solar cell is basically 166mm×166mm or 182mm×182mm, the required accuracy is 0.2mm, the inspection speed is 6000 pieces per hour, and the interval between each piece is about 50mm. Therefore, a 25 million pixel camera and a focal length 25mm lens are selected. Therefore, the Dushen GR4900M-M58-GH camera used in this system can reach a frame rate of 40MHz.

[0045] In this embodiment, a Dell industrial computer 7080-ATS is used, the M58 to C-port adapter ring is the AZURE-ADCM58 model, a 4-port 10 Gigabit network card is used to collect pictures, a USB3.0 hard disk extension cable is used to connect the hard disk, a 10M armored optical fiber cable and a 10M power trigger cable, a stroboscopic light source extension cable is the Turing Microvision TVS-LL-HC-5 model, a power adapter is the Dushen ADS-26SGP-12-12024E, and a process light source is the Turing Microvision model BFL-296GB32-TVS.

[0046] like Figure 2 As shown, a solar cell appearance inspection image compression method based on wavelet transform is characterized by comprising the following steps:

[0047] (1) This application uses the Haar wavelet function with better compression decomposition for decomposition, and after decomposition, a low-frequency component A, a horizontal high-frequency component H, a vertical high-frequency component V and a diagonal high-frequency component D are obtained;

[0048] (2) Since the important and useful information after wavelet decomposition is distributed in the low-frequency components, and the noise signal is distributed in the higher frequency, the high-frequency coefficients are subjected to threshold quantization processing each time; in the step (2), a suitable threshold is selected according to the amplitude values ​​of the coefficients of different high-frequency components. The threshold used in the present application in the step (2) is a soft threshold, and the threshold coefficient is selected as the mean square error of the high-frequency assignment, which ensures that there are enough effective signals passing the threshold each time, thus greatly reducing the presence of noise.

[0049] (3) After the signal is reduced by noise reduction, the interference of noise in the image is reduced, and then the inverse wavelet transform is used, which can greatly reduce the size of the image with basically no distortion; the inverse wavelet transform used in the step (3) is the Haar wavelet.

[0050] (4) The compressed image is judged for its appearance through an image analysis algorithm, and feedback is given as OK / NG based on different set values. When it is detected that the image is judged to be NG, the image is saved.

[0051] The main principle of wavelet denoising is to find the best mapping in the wavelet function space from the noisy signal, filter and analyze the signal, filter out the high-frequency interference signal, retain the useful signal as much as possible, and recombine the useful signal with the low-pass filtered signal.

[0052] The image input is transformed by wavelet transform through dwt2 in the pywt toolkit. Haar wavelet is selected in dwt2. At this time, the haar wavelet transform in the wavelet transform is used. After the transformation, the low-frequency component, horizontal high-frequency component, vertical high-frequency component, and diagonal high-frequency component of the wavelet transform can be obtained. After obtaining the high-frequency component, noise reduction is performed according to the adaptive threshold method of the wavelet transform. Finally, the horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component after noise reduction are obtained. The high-frequency component obtained has been filtered in the high-frequency band in the wavelet transform. The processing principle is to add a window in the wavelet transform and then obtain the compressed image through the inverse wavelet transform (idwt2), and the output is saved or AOI analysis is performed and finally saved.

[0053] like Figure 3 As shown in the figure, wavelet transform is a time-scale analysis method for a signal, with the characteristics of multi-resolution analysis, and the ability to characterize the local characteristics of the signal in both the time domain and the frequency domain. It is a method for analyzing local changes in time and frequency with a fixed window size but a variable shape, and both the time window and the frequency window can be changed. It has both a lower time resolution and a higher frequency resolution in the low-frequency part, and a higher time resolution and a lower frequency resolution in the high-frequency part. It is very suitable for analyzing non-stationary signals and extracting local characteristics of signals. When processing and analyzing signals, wavelet transform has the adaptability to the signal.

[0054] Wavelet decomposition can reduce the noise of the signal and improve the signal-to-noise ratio. Generally, after wavelet transformation, the useful signal is usually a low-frequency signal or a relatively stable signal, while the noise signal is generally a high-frequency signal. The wavelet signal used in this application is the Haar wavelet function. The wavelet function generates high-frequency information and low-frequency information through a high-pass filter and a low-pass filter. The derivation process of the Haar wavelet function is as follows:

[0055] Haar function For integer k, Graphics The result after shifting right or left by k,

[0056] Let V1 be the following linear combination

[0057]

[0058] V1 is a linear space with finite support and discontinuity points. The linear space composed of piecewise constants, the scaling function is The wavelet function is

[0059] The following conclusions will be drawn for the function through Haar wavelet transform:

[0060] (1) For a function f(x), if and only if f(2 n x)∈V n When f(x)∈V

[0061] (2) For a function f(x), if and only if f(2 -n x)∈V0, then f(x)∈V n

[0062] (3) For the function Constitute the standard orthogonal set of Vn.

[0063] Haar wavelet transform is a windowing process. Haar wavelet is a low-pass filter that will filter out high-frequency components.

[0064] In the above, X represents input data, hhigh represents high-pass filter, hlow represents low-pass filter, A, H, V, and D are output data after wavelet transform, respectively representing A low-frequency information, H horizontal high-frequency information, V vertical high-frequency information, and D diagonal high-frequency information.

[0065] After the wavelet transform is completed, the low-frequency components contain a lot of information, and the low-frequency components are basically not processed. The noise is basically distributed in the high-frequency components, so after obtaining the high-frequency components, the noise part in the high-frequency components is filtered out according to the adaptive wavelet threshold method, and the compressed image is obtained by performing inverse wavelet transform on the filtered high-frequency components and low-frequency components.

[0066] The finished AOI optical inspection picture of solar cells used in this application comes from the Changzhou Yijing project. Picture of the product front Figure 1 Zhang, finished product back Figure 1 There are 4 printing pictures on one sheet and 3 printing pictures on one sheet. The specifications are shown in Table 1 and Table 2.

[0067] Table 1 Original image size

[0068] picture 3 printing 4 printing Product front view Product back view size 24611KB 25641KB 102061KB 102561KB

[0069] Table 2 Image size after algorithm processing

[0070] picture 3 printing 4 printing Product front view Product back view size 9700KB 9755KB 9843KB 9840KB

[0071] From the above table, we can see that the compression rate of the compressed images is higher in the product area, and lower in printing. This means that the photos collected in the 3 / 4 printing process contain a lot of useful information, while the pictures collected in the product area have large noise and relatively less useful information. So we need to compare whether the finished product pictures can be used in normal inspection machines after compression or the loss after compression.

[0072] Example:

[0073] like Figure 4 , 5 As shown in , 6, 7, 8, 9, 10, and 11, the Beyond Compare tool is used to analyze the images. This tool can compare the differences in different areas. Figure 4 , 5 As shown, it can be seen that the two pictures almost overlap, but after zooming in, you can see that there are differences as follows Figure 6 , 7 As shown. Figure 5 It can be seen that the image size of the original document is about 100M. After being compressed by the algorithm, the image size is reduced to 10M of storage space, which greatly reduces the storage space requirements and achieves the compression effect. Figure 7 The tool is used to analyze the compressed image and compare it with the original image. There are 26214400 pixels in the whole image, of which 26205497 are the same pixels and 8903 are unimportant differences. After compression, it can be seen that the method used in this article does not cause the usual image loss. Figure 6 The dark dots can be seen in the framed part. These dots indicate that the compressed image is different from the original image. These differences are minor differences and will not affect the AOI determination. The 100M original image contains 26,214,400 pixels, of which 26,205,497 pixels are the same after compression and before compression, and 8,903 pixels are unimportant differences. This shows that the accuracy after compression is 99.966%, which is sufficient for storage and viewing, and can also perform AOI optical inspection normally. Figure 6 It can be seen that the basic overlap, now zoom in on the picture to see the differences. Fig. 9 It can be seen that the size of the picture before and after compression is about 25M before compression and about 9M after compression. By comparing the compressed pictures, it is found that the method used in this article can achieve lossless compression of pictures. Fig.10It can be seen that there are differences near the grid lines, and these differences are unimportant pixels. The number of pixels collected by the four-pass printing is 26,214,400, the number of overlapping pixels is 26,081,399, and the number of unimportant pixels is 133,001. Therefore, the overlap rate between the compressed image and the original image is 99.5% correct. Therefore, it can be saved and used normally.

[0074] This embodiment proves that wavelet transform can be used in industrial compression of images. The method of this embodiment can effectively reduce the size of images, greatly reduce the hard disk space during storage, save production storage space, and reduce storage costs. The method described in this application not only compresses images but can also be used as a normal algorithm.

[0075] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A solar cell appearance inspection image compression method based on wavelet transform, characterized in that: The steps include: (1) This application uses the Haar wavelet function with better compression decomposition for decomposition, and after decomposition, a low-frequency component A, a horizontal high-frequency component H, a vertical high-frequency component V and a diagonal high-frequency component D are obtained; (2) Since the important and useful information after wavelet decomposition is distributed in the low-frequency components, while the noise signal is distributed in the higher frequencies, the high-frequency coefficients are threshold quantized each time; (3) After the signal is reduced by noise reduction, the interference of noise in the image is reduced. Then, the inverse wavelet transform can greatly reduce the size of the image with little distortion; (4) The compressed image is judged by the image analysis algorithm, and feedback of OK / NG is given according to different set values. When the image is judged to be NG, the image is saved; The derivation process of the Haar wavelet function in step (1) is as follows: Haar function For integers Graphics The result after shifting right or left by k, Let V1 be the following linear combination V1 is a linear space with finite support and discontinuity points. The linear space composed of piecewise constants, the scaling function is The wavelet function is The following conclusions will be drawn for the function through Haar wavelet transform: (1) For a function f(x), if and only if f(2 n x)∈V n When f(x)∈V (2) For a function f(x), if and only if f(2 -n x)∈V0, then f(x)∈V n (3) For the function k∈z forms the standard orthogonal set of Vn, Haar wavelet transform is a windowing process. Haar wavelet is a low-pass filter that will filter out high-frequency components. The AOI image of the solar cell is 100M. The image of the compressed finished AOI is used for analysis. The Python language is used for image analysis. CV2, numpy and pywt toolkits are referenced in python. The CV2 toolkit contains the opening and closing functions of the image, the numpy toolkit contains the adaptive analysis function in wavelet analysis, and the pywt toolkit contains the wavelet transform tool. The image input is subjected to wavelet transform through dwt2 in the pywt toolkit. Haar wavelet is selected in dwt2. At this time, the haar wavelet transform in the wavelet transform is used. After the transformation, the low-frequency component, horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component of the wavelet transform can be obtained. After obtaining the high-frequency component, noise reduction is performed according to the adaptive threshold method of the wavelet transform. Finally, the horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component after noise reduction are obtained. The obtained high-frequency component has been subjected to high-frequency band filtering in the wavelet transform. The processing principle is to add a window in the wavelet transform and then obtain the compressed image through the inverse wavelet transform, and the output is saved or the AOI analysis is performed and finally saved.

2. According to the wavelet transform-based solar cell appearance inspection image compression method of claim 1, it is characterized in that: In the step (2), a suitable threshold is selected according to the amplitude values ​​of the coefficients of different high-frequency components.

3. The solar cell appearance inspection image compression method based on wavelet transform according to claim 1 is characterized in that: The threshold used in the present application in step (2) is a soft threshold, and the threshold coefficient is the mean square error of the high-frequency assignment, which ensures that there are enough valid signals passing the threshold each time, thus greatly reducing the presence of noise.

4. The solar cell appearance inspection image compression method based on wavelet transform according to claim 1 is characterized in that: The inverse wavelet transform used in step (3) is the Haar wavelet.

5. A solar cell appearance inspection image compression system based on wavelet transform, characterized in that: It includes visual inspection equipment and a collection system. The visual inspection equipment includes a transmission mechanism. The transmission mechanism uses a conveyor belt. The conveyor belt is installed on a base. There is a gap between the silicon wafers to prevent the two silicon wafers from being too close to each other and affecting the imaging. The collection system includes an industrial camera and a lens. The image compression system also includes a Dell industrial computer, an M58 to C-port adapter ring, a 4-port 10 Gigabit network card, a USB3.0 hard disk extension cable, a power trigger cable, a stroboscopic light source extension cable, a power adapter, and a process light source. The AOI image of the solar cell is 100M. The image of the compressed finished AOI is used for analysis. The Python language is used for image analysis. CV2, numpy and pywt toolkits are referenced in Python. The CV2 toolkit contains the opening and closing functions of the image. The numpy toolkit contains the adaptive analysis function in wavelet analysis. The pywt toolkit contains the wavelet transform tool. The image input is subjected to wavelet transform through dwt2 in the pywt toolkit. Haar wavelet is selected in dwt2. At this time, the haar wavelet transform in the wavelet transform is used. After the transformation, the low-frequency component, horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component of the wavelet transform can be obtained. After obtaining the high-frequency component, noise reduction is performed according to the adaptive threshold method of the wavelet transform. Finally, the horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component after noise reduction are obtained. The obtained high-frequency component has been subjected to high-frequency band filtering in the wavelet transform. The processing principle is to add a window in the wavelet transform and then obtain the compressed image through the inverse wavelet transform, and the output is saved or the AOI analysis is performed and finally saved.

6. The solar cell appearance inspection image compression system based on wavelet transform according to claim 5 is characterized in that: The silicon wafer is a solar cell wafer, which is 166mm×166mm or 182mm×182mm, with an accuracy of 0.2mm, a detection speed of 6000 pieces per hour, and a spacing of 50mm between each piece. The industrial camera uses a Dushen GR4900M-M58-GH camera with a 25-megapixel camera and a focal length of 25mm lens, and a frame rate of 40MHz.

7. The solar cell appearance inspection image compression system based on wavelet transform according to claim 5 is characterized in that: The Dell industrial computer adopts the 7080-ATS model, the M58 to C-port adapter ring adopts the AZURE-ADCM58 model, the 4-port 10 Gigabit network card is used to collect pictures, the USB3.0 hard disk extension cable is used to connect the hard disk, the power trigger cable adopts a 10M armored optical fiber cable, the stroboscopic light source extension cable adopts the Turing Microvision TVS-LL-HC-5 model, the power adapter adopts the Dushen ADS-26SGP-12-12024E, and the process light source adopts the Turing Microvision model BFL-296GB32-TVS.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images