Solar cell panel surface quality detection method and device based on holographic imaging

Abstract

The invention provides a solar cell panel surface quality detection method and device based on holographic imaging, and relates to the technical field of solar cell panel invisible defect recognizing.The method comprises the steps that reference light emitted by a coherent laser light source and object light which is emitted by the coherent laser light source and passes through a to-be-detected sample are combined; obtaining an interference hologram of the to-be-detected sample; performing noise reduction processing and numerical reconstruction processing on the interference hologram to obtain a plurality of reproductive images at different reproductive distances from the surface invisible defects of the photovoltaic cell panel to be detected; inputting the plurality of reproductive images into a self-focusing holographic defect identification network, and detecting spatial distribution positions, defect types and depth information of invisible defects on the surface of the photovoltaic cell panel to be detected; the self-focusing holographic defect recognition network comprises a plurality of self-focusing attention units which are distributed in a U shape, and each self-focusing attention unit is used for extracting and fusing depth features and spatial features of a plurality of reproductive images; the technical problem that invisible defects of a cell panel cannot be accurately and comprehensively recognized is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of invisible defect recognition of solar panels, and in particular to a method and device for detecting surface quality of solar panels based on holographic imaging. Background Art

[0002] There are often various invisible defects on the surface and inside of silicon solar panels that are difficult for the human eye to identify, such as hidden cracks, fragments, broken grids and stress. At present, the detection of invisible defects of silicon solar panels mainly relies on electroluminescence (EL), photoluminescence (PL), infrared thermal imaging and other technologies.

[0003] However, electroluminescence technology is achieved by applying reverse bias to the solar panel, which may cause damage to the solar cell and has low detection accuracy in some practical application scenarios; photoluminescence technology has high requirements for application equipment and cannot accurately detect the type of cracks in some practical application scenarios; infrared thermal imaging technology detects infrared radiation on the surface of the solar panel to identify crack defects with significant temperature differences. This identification result is greatly affected by the ambient temperature and has low sensitivity for identifying micro cracks with unclear temperature differences.

[0004] Therefore, none of the above technologies can quickly and comprehensively identify invisible defects of solar panels, especially in the efficient detection of crack types, geometries and depths. Summary of the invention

[0005] The purpose of the present invention is to provide a method and device for detecting the surface quality of a solar panel based on holographic imaging, so as to alleviate the technical problem that invisible defects of the solar panel cannot be accurately and comprehensively identified.

[0006] In a first aspect, the present invention provides a method for detecting the surface quality of a solar panel based on holographic imaging, comprising:

[0007] The reference light emitted by the coherent laser light source and the object light emitted by the coherent laser light source and passed through the sample to be tested are combined to obtain an interference hologram of the sample to be tested; wherein the sample to be tested includes a photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested;

[0008] Performing noise reduction and numerical reconstruction on the interference hologram to obtain multiple reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested;

[0009] The multiple reproduced images are input into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested; wherein the self-focusing holographic defect recognition network includes a plurality of U-shaped distributed self-focusing attention units, each of which is used to extract and fuse the depth features and spatial features of the multiple reproduced images.

[0010] In an optional embodiment, the step of inputting the multiple reconstructed images into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of the invisible defects on the surface of the photovoltaic panel to be tested includes:

[0011] Input the multiple reconstructed images into a downsampling branch of a self-focusing holographic defect recognition network of a U-shaped structure, and obtain a first processing result after being processed in sequence by each self-focusing attention unit in the downsampling branch;

[0012] The first processing result is transmitted to the bottleneck layer of the U-shaped structure, and processed by the self-focusing attention unit of the bottleneck layer to obtain a second processing result;

[0013] The second processing result is transmitted to the up-sampling branch of the self-focusing holographic defect recognition network of the U-shaped structure, and the third processing result is obtained by sequentially processing each self-focusing attention unit in the up-sampling branch;

[0014] Based on the second processing result, detecting the defect types of invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type;

[0015] Based on the third processing result, the spatial distribution positions of invisible defects on the surface of the photovoltaic panel to be tested are detected.

[0016] In an optional embodiment, based on the second processing result, the step of detecting the defect type of the invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type includes:

[0017] The second processing result is input into the linear layer, and the output first output result is input into the activation function to calculate the defect probability value and depth information of each invisible defect type on the surface of the photovoltaic panel to be tested.

[0018] In an optional embodiment, based on the third processing result, the step of detecting the spatial distribution position of invisible defects on the surface of the photovoltaic panel to be tested includes:

[0019] The third processing result is input into the convolution layer, the output second output result is input into the activation function, and the output third output result is input into the binarization layer to obtain a position mask for characterizing the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested.

[0020] In an optional embodiment, each self-focusing attention unit processes input information, including:

[0021] The input information is subjected to a de-meaning process by the upstream branch of the self-focusing attention unit, and then subjected to a first depthwise separable convolutional layer and an activation function process to generate an attention matrix for characterizing the spatial distribution of invisible defects;

[0022] The input information is processed by a second depth-separable convolution layer and a third depth-separable convolution layer through the middle branch of the self-focusing attention unit to extract spatial features;

[0023] The input information is processed by Laplace autofocus algorithm and point convolution layer through the lower branch of the autofocus attention unit to extract deep features;

[0024] The deep feature is first multiplied by the preset weight of the lower branch, then multiplied by the dot product result of the attention matrix and the spatial feature, and then processed by the point convolution layer and the fourth deep convolution layer respectively to output the fusion information of the deep feature and the spatial feature in the input information.

[0025] In an optional embodiment, the step of performing noise reduction and numerical reconstruction on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested comprises:

[0026] Performing difference calculation on the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel to obtain a denoised hologram;

[0027] Performing Wiener filtering on the denoised hologram to obtain a denoised hologram;

[0028] The noise reduction hologram is numerically reconstructed according to the angular spectrum method to obtain a plurality of grayscale images of invisible defects at different distances from the surface of the photovoltaic panel to be tested.

[0029] In an optional embodiment, the step of combining the reference light emitted by a coherent laser light source and the object light emitted by the coherent laser light source and passing through the sample to be tested to obtain an interference hologram of the sample to be tested includes:

[0030] The light beam emitted by the coherent laser light source is divided into reference light and object light by a beam splitting lens;

[0031] The reference light that has been phase-delayed by the phase shift device is combined and interfered with the object light that has passed through the sample to be tested, so as to obtain and collect the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel.

[0032] In a second aspect, the present invention provides a solar panel surface quality detection device based on holographic imaging, comprising:

[0033] A holographic imaging module combines the reference light emitted by a coherent laser light source and the object light emitted by the coherent laser light source and passed through the sample to be tested, so as to obtain an interference hologram of the sample to be tested; wherein the sample to be tested includes a photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested;

[0034] An image reconstruction module performs noise reduction and numerical reconstruction on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested;

[0035] A defect recognition module inputs the multiple reproduced images into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested; wherein the self-focusing holographic defect recognition network includes a plurality of U-shaped distributed self-focusing attention units, each of which is used to extract and fuse the depth features and spatial features of the multiple reproduced images.

[0036] In a third aspect, the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, wherein the processor implements a method as described in any one of the aforementioned embodiments when executing the program.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the method described in any one of the aforementioned embodiments is implemented.

[0038] The embodiment of the present invention provides a method and device for detecting the surface quality of solar panels based on holographic imaging. It combines digital holographic reconstruction technology and deep learning technology to obtain three-dimensional image information of the surface of photovoltaic panels through holographic imaging. In order to realize efficient defect recognition in digital holographic technology, the hologram is numerically reconstructed based on the angular spectrum method to obtain multiple reproduced images at different reproduction distances. On this basis, a self-focusing holographic defect recognition network is designed by analyzing the detailed differences between the focused and defocused areas in the reproduced image; the network completes the reconstruction of the depth information in combination with the self-focusing algorithm, and finally realizes the reconstruction and classification of the defect distribution, thereby improving the detection level and efficiency of invisible defects of photovoltaic panels, and further realizing the surface quality detection of solar panels.

[0039] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A flow chart of a method for detecting surface quality of a solar panel based on holographic imaging provided by an embodiment of the present invention;

[0043] Figure 2 An optical path structure diagram of a holographic imaging module provided by an embodiment of the present invention;

[0044] Figure 3 A diagram of an optical path structure of another holographic imaging module provided by an embodiment of the present invention;

[0045] Figure 4 A structural diagram of a self-focusing holographic defect recognition network provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of functional modules of a solar panel surface quality detection device based on holographic imaging provided by an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of the hardware architecture of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The inventors have found that cracks tend to grow and expand preferentially at the location of invisible cracks that already exist during the manufacturing stage. These initial cracks or microcracks may have little or no significant effect on power output in the initial stage; however, during long-term outdoor operation, these cracks will gradually expand, significantly affecting module performance, increasing the risk of power loss, and even causing hot spots, seriously threatening the service life and safety of solar cells. Therefore, in order to ensure the application reliability of solar panels, the accuracy of invisible defect detection on their surface is more critical.

[0050] Existing infrared thermal imaging, electroluminescent imaging, photoluminescent imaging, etc. are all macroscopic two-dimensional imaging, and cannot achieve efficient detection of some three-dimensional and tiny defects on the surface of photovoltaic panels, such as surface flatness and tiny cracks. They are all unable to quickly and comprehensively identify invisible defects of solar panels, especially in terms of efficient detection of crack types, geometric shapes and depths. The types of cracks (such as scratches, surface cracks and through cracks) and their geometric characteristics have a direct impact on the service life and performance of solar cells, and existing methods are difficult to provide accurate and detailed diagnostic information, as follows:

[0051] Electroluminescence technology applies a reverse bias to the solar cell to stimulate the recombination of electron-hole pairs. The defective area appears as a dark or bright area in the image due to different recombination rates. This technology has high sensitivity to microcracks and metallization defects and can clearly display the location and shape of the defects. However, since a reverse bias needs to be applied, it may cause a certain degree of damage to the solar cell. In addition, in a high humidity environment, the formation of a water film on the surface of the cell may affect the propagation and absorption of light. At the same time, temperature changes will significantly affect the recombination rate of carriers, thereby reducing the detection accuracy.

[0052] Photoluminescence technology uses light of a specific wavelength to excite materials to emit light, which can non-destructively detect invisible defects inside materials and perform high-sensitivity detection of solar panels. However, this technology has high requirements for light sources and detection equipment, expensive detection costs, and is easily interfered by background light, which limits its application in complex environments. In addition, although photoluminescence PL technology has certain advantages in crack detection, it is still insufficient to distinguish between types of scratches, surface cracks, and through cracks. These defects appear as continuous, V-shaped, and broken stripes, respectively, and existing technologies cannot provide detailed geometric information and accurate type distinction, limiting the practicality of the detection results.

[0053] Infrared thermal imaging technology detects infrared radiation on the surface of solar panels to locate areas where local temperature rises due to leakage. However, under bias conditions, excessive current in the leakage area will cause local temperature rise, and leakage defects can be identified by infrared imaging technology. However, the sensitivity of this technology is low for invisible defects that do not cause significant temperature differences (such as microcracks). In addition, changes in ambient temperature have a significant impact on the detection results. At the same time, the resolution of infrared thermal imaging technology is low, making it difficult to provide information on the geometry and type of cracks.

[0054] Based on this, an embodiment of the present invention provides a method and device for detecting surface quality of solar panels based on holographic imaging, which can achieve high-precision defect detection of solar panels by combining holographic imaging with a self-focusing attention mechanism.

[0055] To facilitate understanding of this embodiment, a solar panel surface quality detection method based on holographic imaging disclosed in an embodiment of the present invention is first introduced in detail. The holographic imaging of this method can be achieved through an optical path architecture, and the holographic image collected therefrom is input into a host computer, server, controller or other intelligent control device for subsequent processing such as image reconstruction and self-focusing holographic defect recognition.

[0056] Figure 1 A flow chart of a method for detecting the surface quality of a solar panel based on holographic imaging provided by an embodiment of the present invention includes the following steps:

[0057] Step S102, combining the reference light emitted by the coherent laser light source and the object light emitted by the coherent laser light source and passing through the sample to be measured, to obtain an interference hologram of the sample to be measured.

[0058] Here, the light beam emitted by the coherent laser light source is divided into two paths, one of which does not pass through the sample to be tested is defined as the reference light, and the other light beam that passes through the sample to be tested is defined as the object light; wherein the sample to be tested includes the photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested;

[0059] In the embodiment of the present invention, the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background plate can be collected by a CCD camera or other equipment. It is understandable that the above hologram includes the amplitude and phase information of the sample to be tested, so that the subsequent steps can perform corresponding processing based on the hologram, which can achieve more accurate invisible defect detection.

[0060] Step S104, performing noise reduction and numerical reconstruction processing on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested.

[0061] Here, in order to further improve the detection accuracy of invisible defects on the surface of the photovoltaic panel, while reducing the noise of the hologram, the image display of invisible defects at various depths from the surface of the photovoltaic panel to be tested is reproduced; for example, the invisible defects at depths of 1 mm, 1.5 mm, 2 mm, etc. from the surface of the photovoltaic panel to be tested can be reproduced, and the reproduced image is essentially a grayscale image.

[0062] Step S106, inputting the multiple reconstructed images into the self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of the invisible defects on the surface of the photovoltaic panel to be tested.

[0063] Among them, the self-focusing holographic defect recognition network includes multiple self-focusing attention units distributed in a U shape. Each self-focusing attention unit is used to extract and fuse the depth features and spatial features of multiple reproduced images. Through the self-focusing holographic defect recognition network with this structure, the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested can be accurately identified.

[0064] In a preferred embodiment of practical application, firstly, the object light of the coherent laser light source irradiating the sample to be tested and the reference light not irradiated by the sample to be tested are combined for interference to obtain the interference hologram of the sample to be tested at this time; secondly, the interference hologram is subjected to noise reduction and numerical reconstruction processing to obtain multiple reproduced images of the surface invisible defects of the photovoltaic panel to be tested at different reproduction distances; then, the multiple reproduced images are input into a U-shaped self-focusing holographic defect recognition network, and the U-shaped structure is composed of multiple self-focusing attention units for extracting and fusing the depth features and spatial features of the multiple reproduced images, so as to accurately detect the spatial distribution position, defect type and depth information of the invisible defects on the surface of the photovoltaic panel to be tested.

[0065] It is understandable that the unique advantage of holographic imaging is that it can comprehensively detect various invisible defects including hidden cracks, fragments, broken grids and stress, especially in the fields of micro cracks and surface flatness. Step S102 of the embodiment of the present invention obtains an interference hologram of the sample to be tested to improve the accuracy of subsequent invisible defect identification, and the step includes:

[0066] Step 1.1), the light beam emitted by the coherent laser light source is split into reference light and object light by a beam splitting lens.

[0067] Here, the beam splitter lens is capable of splitting the light beam emitted by the coherent laser light source into two beams.

[0068] Step 1.2), combining and interfering the reference light that has been phase-delayed by the phase shift device with the object light of the sample to be tested, to obtain and collect the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel.

[0069] In practical applications, the laser holographic imaging system can first use a coherent laser light source to illuminate the photovoltaic panel to be tested placed on a background plate, and then use a CCD camera to collect the interference hologram of the photovoltaic panel to be tested; secondly, remove the photovoltaic panel to be tested from the object light path, and then use a coherent laser light source to illuminate only the background plate, and then combine the reference light and the object light to generate an interference holographic image of the background plate, and use a CCD camera to collect the interference hologram of the background plate.

[0070] It should be noted that the optical path architecture can use the amplitude division method to achieve optical path separation, dividing it into a reference optical path and an object optical path, and then achieve phase delay by adding a phase shift device in the reference optical path.

[0071] As an optional embodiment, the Figure 2 The optical path architecture shown in the figure realizes the generation of interference holograms; first, a coherent laser light source is selected, and the light emitted by it is collimated and expanded to form a plane light wave, which is divided into two paths after passing through the first 50:50 splitter lens. The transmitted light is the reference light path, which is reflected by the reflector and processed by the phase shift device to produce phase delay, and then enters the third splitter lens; the reflected light is the object light path, which is suitable for transmission objects. A transmission sample to be tested is placed in the object light path. After the light beam passes through the sample to be tested, it is reflected by the reflective film of the first splitter lens and then enters the third splitter lens. After passing through the sample to be tested, the object light carries the amplitude and phase information of the sample, and is combined with the reference light at the third splitter lens to produce interference. At this time, the transmitted light and reflected light of the third splitter lens are the interference holographic images to be collected, and then collected using a CCD camera.

[0072] As another optional embodiment, the Figure 3 The optical path architecture shown realizes the generation of interference hologram; first, a coherent laser light source is selected, and the light emitted by it is collimated and expanded to form a plane light wave, which is divided into two paths after passing through the first 50:50 splitter lens, wherein the transmitted light is the reference light path, and the transmitted light is reflected by the reflector and processed by the phase shift device to produce phase delay, and then enters the third splitter lens; the reflected light is the object light path, and the light beam is reflected by the reflector, transmits the sample to be tested, and then enters the third splitter lens; the object light path is suitable for transmission objects, and the object light carries the amplitude and phase information of the sample after passing through the sample to be tested, and is combined with the reference light at the third splitter lens to produce interference. At this time, the transmitted light and reflected light of the third splitter lens are the interference holographic images to be collected, and then a CCD camera is used to collect them.

[0073] In some embodiments, digital holographic reconstruction and angular spectrum method can be used to further process the three-dimensional image information of the surface of the photovoltaic panel to effectively achieve high-precision and three-dimensional detection of invisible defects in subsequent steps; exemplary step S104 includes:

[0074] Step 2.1), performing difference calculation between the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel to obtain a denoised hologram.

[0075] In the above-mentioned step embodiment, the photovoltaic panel and the background plate are imaged in the same imaging environment through the optical path architecture to generate an interference hologram; at this time, the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background plate are subtracted, and the influence of environmental noise such as the optical path and the light source can be subtracted to obtain a denoised hologram, which can be specifically achieved by the following formula:

[0076] I Fn =I 电池板 -I 背景

[0077] Among them, I Fn To denoise the hologram, I 电池板 is the interferometric hologram of the photovoltaic panel to be tested, I 背景 is the interference hologram of the background plate.

[0078] Step 2.2), the denoised hologram is subjected to Wiener filtering to further reduce random noise interference and obtain a denoised hologram, which can be specifically achieved by the following formula:

[0079] I Fm =E(I Fn )

[0080] Where E(·) represents Wiener filter denoising, I Fm For noise reduction holograms.

[0081] Step 2.3), numerically reconstructing the noise reduction hologram according to the angular spectrum method, and obtaining multiple grayscale images of invisible defects at different distances from the surface of the photovoltaic panel to be tested.

[0082] Here, the conventional angular spectrum method in the prior art is used to denoise the hologram I Fm Numerical reconstruction is performed to obtain multiple reconstructed images at different reconstruction distances.

[0083] In practical applications, such as Figure 4As shown, the self-focusing attention unit is embedded in the U-shaped structure as the basic module of the self-focusing holographic defect recognition network, and then the encoding and decoding functions are realized through the downsampling branch and the upsampling branch of the U-shaped structure. At the bottleneck layer of the network, a linear mapping layer is introduced to output the depth information of the defect type and the defect type. At the same time, a mask of the defect spatial distribution is generated in the decoding stage of the upsampling branch. Exemplarily, this step S106 can be implemented by the following steps, including:

[0084] Step 3.1), input multiple reproduced images into the downsampling branch of the U-shaped self-focusing holographic defect recognition network, and obtain the first processing result after being processed in sequence by each self-focusing attention unit in the downsampling branch.

[0085] Among them, the downsampling branch of the U-shaped structure may include n self-focusing attention units; Figure 4 Only an example in which a downsampling branch consists of four self-focusing attention units is shown, but it is not limited to this; it can be understood that each self-focusing attention unit has the same processing steps for the input information, and the output information of the previous self-focusing attention unit is used as the input information of the next self-focusing attention unit, until each self-focusing attention unit in the downsampling branch of the U-shaped structure is traversed and the first processing result is output.

[0086] Step 3.2), the first processing result is passed to the bottleneck layer of the U-shaped structure, and processed by the self-focusing attention unit of the bottleneck layer to obtain the second processing result.

[0087] The bottleneck layer of the U-shaped structure is a structure connecting the upsampling branch and the downsampling branch, which consists of a single self-focusing attention unit.

[0088] Step 3.3), passing the second processing result to the upsampling branch of the U-shaped self-focusing holographic defect recognition network, and obtaining the third processing result after being processed in sequence by each self-focusing attention unit in the upsampling branch;

[0089] Among them, the upsampling branch of the U-shaped structure may include n self-focusing attention units; Figure 4 Only an example in which an upsampling branch consists of four self-focusing attention units is shown, but it is not limited to this; it can be understood that each self-focusing attention unit has the same processing steps for the input information, and the output information of the previous self-focusing attention unit is used as the input information of the next self-focusing attention unit, until each self-focusing attention unit in the upsampling branch of the U-shaped structure is traversed and the third processing result is output.

[0090] Step 3.4), based on the second processing result, detect the defect types of invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type.

[0091] Exemplarily, the second processing result is input into the linear layer, and the output first output result is input into the activation function to calculate the defect probability value and depth information of each invisible defect type on the surface of the photovoltaic panel to be tested.

[0092] The linear layer is combined with the Softmax activation function to generate the type and depth information of the defect, which is specifically achieved through the following formula:

[0093] D=Sofmax(Linear(F out_2 ))

[0094] Among them, F out_2 is the second processing result, D is used to characterize the probability and depth information of each invisible defect type, Linear(·) represents the linear layer, and Sofmax(·) is the activation function.

[0095] Step 3.5), based on the third processing result, detecting the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested.

[0096] Exemplarily, the third processing result is sequentially input into the first convolution layer and the second convolution layer, the output second output result is input into the activation function, and the output third output result is input into the binarization layer to obtain a position mask for characterizing the spatial distribution position of invisible defects on the surface of the photovoltaic panel to be tested.

[0097] The second processing result of the output of the bottleneck layer is processed by each self-focusing attention unit in the upsampling branch to complete feature decoding, and the third processing unit is output. Then, a position mask for characterizing the spatial distribution of invisible defects is generated through an additional prediction head (composed of the first convolution layer, the second convolution layer and the binarization layer). The mask can reflect a two-dimensional matrix, and the spatial position of the defect is known according to the value of each element in the matrix; for example, if an element A is 0, there is no defect at the position of element A, and if an element B is 1, there is a defect at the position of element B; the specific position mask is calculated by the following formula:

[0098] Fmask=Binary(Softmax(Conv(Conv(Fout_3))))

[0099] Among them, F mask is the position mask, F out_3 is the third processing result, Sofmax(·) is the activation function, Conv(·) is the convolution layer, and Binary(·) is the binarization layer.

[0100] The self-focusing attention unit in the above embodiment embeds the depth information reconstruction and defect position recognition of the self-focusing algorithm into a learnable self-focusing holographic defect recognition network. The network module is divided into three paths. The input information entering the self-focusing attention unit will pass through the upper branch F u , Middle Branch F m and the lower branch F d Three branches. The self-focusing attention holographic defect recognition network composed of self-focusing attention units focuses on the defect area and extracts depth information and defect features. Specifically, each self-focusing attention unit processes the input information in the following steps:

[0101] Step 4.1), the input information is de-meaned by the upstream branch of the self-focusing attention unit, and then processed by the first depth-separable convolutional layer and the activation function to generate an attention matrix for characterizing the spatial distribution of invisible defects;

[0102] The upper branch includes a learnable self-aggregation operator. Based on this custom learnable operator, the approximate gradient distribution of the spatial scale can be obtained through the mean removal operation, and the learned weights of the first depth-separable convolutional layer are used to further focus on the key areas of the image and generate an attention matrix.

[0103] Specifically, the ground branch calculates the approximate gradient distribution after removing the mean through deep separable convolution (DwConv) and generates the attention matrix, which is as follows:

[0104] F u =Sofmax(DwConv(I F -mean(I F )))

[0105] Among them, F u is the attention matrix, DwConv(·) is the depth-wise separable convolutional layer, and I F is the input information. If the input at this time is the self-focusing attention unit of the first execution order of the downsampling branch, then I F are multiple reconstructed images, and Sofmax(·) is the activation function.

[0106] Step 4.2), the input information is processed by the second depth-separable convolution layer and the third depth-separable convolution layer through the middle branch of the self-focusing attention unit to extract spatial features;

[0107] The middle branch is used to extract the features of the spatial dimension. The spatial features can be directly extracted by applying the depthwise separable convolution to the input information. The spatial features are calculated by the following formula:

[0108] F m =DwConv(IF )

[0109] Among them, F m is the spatial feature, DwConv(·) is the depth-wise separable convolutional layer, and I F is the input information. If the input at this time is the self-focusing attention unit of the first execution order of the downsampling branch, then I F Multiple reproduced images.

[0110] Step 4.3), the input information is processed by the Laplace autofocus algorithm and the point convolution layer through the lower branch of the autofocus attention unit to extract the deep features;

[0111] The Laplace operator L based self-focus gating of the lower branch is used to realize the extraction of depth information features, and the weight w is used to strengthen or weaken the influence of the lower branch; the lower branch is inspired by the peak judgment in the Laplace self-focusing algorithm, and the Laplace self-focusing algorithm L(·) is used in the lower branch to represent I F The channel dimension of the 8-neighborhood Laplacian operation is performed, and the feature extraction of the channel dimension is performed through the point convolution layer PwConv(·), and the main peak feature is enhanced to obtain the deep feature F d .

[0112] F d =PwConv(L(I F ))

[0113] In step 4.4), the deep feature is first multiplied by the preset weight of the lower branch, and then multiplied by the dot product result of the attention matrix and the spatial feature, and then processed by the point convolution layer and the fourth deep convolution layer respectively, and the fusion information of the deep feature and the spatial feature in the input information is output.

[0114] The spatial dimension features output by the middle branch are multiplied point by point with the attention matrix to integrate the depth information and spatial features. Subsequently, the three branches are further fused in the spatial dimension and the depth dimension respectively. The specific calculation can be done by the following formula:

[0115] F out_1 =DwConv(PwConv(F u F m (w·F d )))

[0116] Among them, F out_1 It is the first processing result output by the self-focusing attention unit.

[0117] It should be noted that the designed self-focusing holographic defect recognition network is a multi-task network, including a defect space position extraction module and a depth information and defect type determination module. The cross entropy loss function is used to train the depth information and defect type determination module, and the mean square error (MSE) loss function is used to train the defect space position extraction module. By constructing a holographic defect data set containing different defect types and distributions, the network is supervised and trained to obtain the optimized network training weight W, which exists in each depth-separable convolution layer DWconv of the trained network. In the holographic defect detection application, the optimized network weight W is first loaded into the trained self-focusing holographic defect recognition network, and then the trained self-focusing holographic defect recognition network is used to execute steps S102-S106 of the aforementioned embodiment to realize invisible defect detection of photovoltaic panels based on holographic imaging, and output defect distribution maps, depth information and defect types.

[0118] The embodiments of the present invention can avoid damage to solar cells through non-destructive detection methods without applying voltage or physical contact; at the same time, based on interference holography technology, it can simultaneously obtain the amplitude and phase information of the sample, and combine the depth information reconstruction and spatial feature extraction in the self-focusing holographic defect recognition network to achieve accurate recognition of the position, depth and distribution of invisible defects. This method significantly improves the sensitivity and resolution of invisible defect detection.

[0119] In some embodiments, Figure 5 As shown, an embodiment of the present invention provides a solar panel surface quality detection device based on holographic imaging, comprising:

[0120] A holographic imaging module combines the reference light emitted by a coherent laser light source and the object light emitted by the coherent laser light source and passed through the sample to be tested, so as to obtain an interference hologram of the sample to be tested; wherein the sample to be tested includes a photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested;

[0121] An image reconstruction module performs noise reduction and numerical reconstruction on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested;

[0122] A defect recognition module inputs the multiple reproduced images into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested; wherein the self-focusing holographic defect recognition network includes a plurality of U-shaped distributed self-focusing attention units, each of which is used to extract and fuse the depth features and spatial features of the multiple reproduced images.

[0123] Furthermore, the defect recognition module is specifically used to input the multiple reproduced images into the downsampling branch of the self-focusing holographic defect recognition network of the U-shaped structure, and obtain a first processing result through sequential processing by each self-focusing attention unit in the downsampling branch; transfer the first processing result to the bottleneck layer of the U-shaped structure, and obtain a second processing result through processing by the self-focusing attention unit in the bottleneck layer; transfer the second processing result to the upsampling branch of the self-focusing holographic defect recognition network of the U-shaped structure, and obtain a third processing result through sequential processing by each self-focusing attention unit in the upsampling branch; based on the second processing result, detect the defect type of the invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type; based on the third processing result, detect the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested.

[0124] Furthermore, the defect recognition module is specifically used to input the second processing result into the linear layer, and then input the output first output result into the activation function, so as to calculate the defect probability value and depth information of each invisible defect type on the surface of the photovoltaic panel to be tested.

[0125] Furthermore, the defect recognition module is specifically used to input the third processing result into the convolution layer, input the output second output result into the activation function, and then input the output third output result into the binarization layer to obtain a position mask for characterizing the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested.

[0126] Furthermore, the defect recognition module is specifically used to perform de-meaning processing on the input information through the upper branch of the self-focusing attention unit, and then perform processing through the first depth-separable convolution layer and the activation function to generate an attention matrix for characterizing the spatial distribution of invisible defects; perform a second depth-separable convolution layer and a third depth-separable convolution layer processing on the input information through the middle branch of the self-focusing attention unit to extract spatial features; perform Laplace self-focusing algorithm processing and point convolution layer processing on the input information through the lower branch of the self-focusing attention unit to extract deep features; first multiply the deep features and the preset weights of the lower branch, and then multiply them with the dot product result of the attention matrix and the spatial features, and then process them through the point convolution layer and the fourth depth convolution layer respectively to output the fusion information of the deep features and spatial features in the input information.

[0127] Furthermore, the image reconstruction module is specifically used to perform difference calculation on the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel to obtain a denoised hologram; perform Wiener filtering on the denoised hologram to obtain a reduced-noise hologram; and numerically reconstruct the reduced-noise hologram according to the angular spectrum method to obtain multiple grayscale images of invisible defects at different distances from the surface of the photovoltaic panel to be tested.

[0128] Furthermore, the holographic imaging module is specifically used to split the light beam emitted by the coherent laser light source into reference light and object light through a splitter lens; combine and interfere the reference light that has been phase-delayed through a phase shift device with the object light that has passed through the sample to be tested, so as to obtain and collect the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel.

[0129] An embodiment of the present invention provides an electronic device for implementing an electronic device. In this embodiment, the electronic device may be, but is not limited to, a personal computer (PC), a laptop computer, a monitoring device, a server, or other computer device with analysis and processing capabilities.

[0130] As an exemplary embodiment, see Figure 6 The electronic device 110 includes a communication interface 111, a processor 112, a memory 113 and a bus 114. The processor 112, the communication interface 111 and the memory 113 are connected via the bus 114. The memory 113 is used to store a computer program that supports the processor 112 to execute the method. The processor 112 is configured to execute the program stored in the memory 113.

[0131] The machine-readable storage medium mentioned in this article can be any electronic, magnetic, optical or other physical storage device that can contain or store information, such as executable instructions, data, etc. For example, the machine-readable storage medium can be: RAM (Radom Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), any type of storage disk (such as CD, DVD, etc.), or similar storage medium, or a combination thereof.

[0132] The non-volatile medium may be a non-volatile memory, a flash memory, a storage drive (such as a hard drive), any type of storage disk (such as a CD, DVD, etc.), or a similar non-volatile storage medium, or a combination thereof.

[0133] It can be understood that the specific operation methods of each functional module in this embodiment can refer to the detailed description of the corresponding steps in the above method embodiment, and will not be repeated here.

[0134] The computer-readable storage medium provided in the embodiment of the present invention stores a computer program. When the computer program code is executed, the method described in any of the above embodiments can be implemented. For specific implementation, please refer to the method embodiment, which will not be described in detail here.

[0135] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0136] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0137] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0138] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the aforementioned embodiments, those of ordinary skill in the art should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed by the present invention, or can easily conceive of changes, or make equivalent replacements for some of the technical features therein. Such modifications, changes or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the protection scope of the present invention.

Claims

1. A method for detecting the surface quality of solar panels based on holographic imaging, characterized in that: include: The reference light emitted by the coherent laser light source and the object light emitted by the coherent laser light source and passed through the sample to be tested are combined to obtain an interference hologram of the sample to be tested; wherein the sample to be tested includes a photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested; Performing noise reduction and numerical reconstruction on the interference hologram to obtain multiple reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested; The multiple reproduced images are input into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested; wherein the self-focusing holographic defect recognition network includes a plurality of U-shaped self-focusing attention units, each of which is used to extract and fuse the depth features and spatial features of the multiple reproduced images.

2. The method according to claim 1, characterized in that The step of inputting the multiple reconstructed images into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of the invisible defects on the surface of the photovoltaic panel to be tested comprises: Input the multiple reconstructed images into a downsampling branch of a self-focusing holographic defect recognition network of a U-shaped structure, and obtain a first processing result after being processed in sequence by each self-focusing attention unit in the downsampling branch; The first processing result is transmitted to the bottleneck layer of the U-shaped structure, and processed by the self-focusing attention unit of the bottleneck layer to obtain a second processing result; The second processing result is transmitted to the up-sampling branch of the self-focusing holographic defect recognition network of the U-shaped structure, and the third processing result is obtained by sequentially processing each self-focusing attention unit in the up-sampling branch; Based on the second processing result, detecting the defect type of invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type; Based on the third processing result, the spatial distribution positions of invisible defects on the surface of the photovoltaic panel to be tested are detected.

3. The method according to claim 2, characterized in that Based on the second processing result, the step of detecting the defect type of the invisible defects on the surface of the photovoltaic panel to be tested, as well as the probability value and depth information of each defect type includes: The second processing result is input into the linear layer, and the output first output result is input into the activation function to calculate the defect probability value and depth information of each invisible defect type on the surface of the photovoltaic panel to be tested.

4. The method according to claim 2, characterized in that: Based on the third processing result, the step of detecting the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested comprises: The third processing result is input into the convolution layer, the output second output result is input into the activation function, and the output third output result is input into the binarization layer to obtain a position mask for characterizing the spatial distribution position of the invisible defects on the surface of the photovoltaic panel to be tested.

5. The method according to claim 2, characterized in that: The steps for each self-focusing attention unit to process input information include: The input information is subjected to a de-meaning process by the upstream branch of the self-focusing attention unit, and then subjected to a first depthwise separable convolutional layer and an activation function process to generate an attention matrix for characterizing the spatial distribution of invisible defects; The input information is processed by a second depth-separable convolution layer and a third depth-separable convolution layer through the middle branch of the self-focusing attention unit to extract spatial features; The input information is processed by Laplace autofocus algorithm and point convolution layer through the lower branch of the autofocus attention unit to extract deep features; The deep feature is first multiplied by the preset weight of the lower branch, then multiplied by the dot product result of the attention matrix and the spatial feature, and then processed by the point convolution layer and the fourth deep convolution layer respectively to output the fusion information of the deep feature and the spatial feature in the input information.

6. The method according to claim 1, characterized in that The step of performing noise reduction and numerical reconstruction on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested comprises: Performing difference calculation on the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel to obtain a denoised hologram; Performing Wiener filtering on the denoised hologram to obtain a denoised hologram; The noise reduction hologram is numerically reconstructed according to the angular spectrum method to obtain a plurality of grayscale images of invisible defects at different distances from the surface of the photovoltaic panel to be tested.

7. The method according to claim 1, characterized in that The step of combining the reference light emitted by a coherent laser light source and the object light emitted by the coherent laser light source and passing through the sample to be tested to obtain an interference hologram of the sample to be tested comprises: The light beam emitted by the coherent laser light source is divided into reference light and object light by a beam splitting lens; The reference light that has been phase-delayed by the phase shift device is combined and interfered with the object light that has passed through the sample to be tested, so as to obtain and collect the interference hologram of the photovoltaic panel to be tested and the interference hologram of the background panel.

8. A solar panel surface quality detection device based on holographic imaging, characterized in that: include: A holographic imaging module combines the reference light emitted by a coherent laser light source and the object light emitted by the coherent laser light source and passed through the sample to be tested, so as to obtain an interference hologram of the sample to be tested; wherein the sample to be tested includes a photovoltaic cell panel to be tested and a background plate for placing the photovoltaic cell panel to be tested; An image reconstruction module performs noise reduction and numerical reconstruction on the interference hologram to obtain a plurality of reconstructed images at different reconstructed distances from the surface invisible defects of the photovoltaic panel to be tested; A defect recognition module inputs the multiple reproduced images into a self-focusing holographic defect recognition network to detect the spatial distribution position, defect type and depth information of invisible defects on the surface of the photovoltaic panel to be tested; wherein the self-focusing holographic defect recognition network includes a plurality of U-shaped distributed self-focusing attention units, each of which is used to extract and fuse the depth features and spatial features of the multiple reproduced images.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a program stored in the memory and capable of being run on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images