Visual on-line inspection method and system for photovoltaic module appearance quality

Abstract

This invention relates to the field of image processing technology and discloses a method and system for online visual inspection of the appearance quality of photovoltaic modules. The method includes: acquiring a sequence of surface images of the photovoltaic module; extracting the grayscale gradient map of each frame in the sequence of surface images; superimposing the gradient magnitude maxima points in the grayscale gradient maps to generate a first intermediate image of the sequence of surface images; performing de-reflection processing on the sequence of surface images to generate a second intermediate image of the sequence of surface images; fusing the first and second intermediate images to obtain a high-frequency enhanced reconstructed image; inputting the high-frequency enhanced reconstructed image into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module; marking pixels with probability values ​​greater than an adaptive judgment threshold in the defect probability distribution map as defect pixels; and performing connectivity analysis on the defect pixels to output the defect detection result of the photovoltaic module. This invention can improve the accuracy of visual inspection of appearance quality defects in photovoltaic modules.

Description

Technical Field

[0001] This invention relates to a method and system for online visual inspection of the appearance quality of photovoltaic modules, belonging to the field of image processing technology. Background Technology

[0002] Online visual inspection of photovoltaic module appearance quality refers to the process of real-time, non-contact, automated identification and classification of surface defects in photovoltaic modules on the production line using high-resolution optical imaging and intelligent algorithm technology. By replacing traditional manual visual inspection, it achieves high-speed and accurate capture of minute defects such as microcracks, broken pieces, and appearance flaws. This not only significantly improves the quality consistency and yield of products leaving the factory, but also effectively reduces the power generation degradation and safety risks of modules during long-term outdoor operation. Furthermore, it provides key technical support for photovoltaic enterprises to reduce costs and increase efficiency, and promote the transformation of manufacturing towards intelligence and digitalization.

[0003] Traditional methods for visual online inspection of photovoltaic module appearance quality typically rely on fixed-angle area array light source illumination combined with ordinary industrial cameras for image acquisition, and mainly use edge detection based on image processing for feature recognition. This method cannot effectively eliminate the strong specular reflection and glare interference on the surface of photovoltaic tempered glass, resulting in low contrast between defect areas and background, thus leading to a high misjudgment rate in visual inspection of photovoltaic module appearance quality defects. Summary of the Invention

[0004] This invention provides a method and system for online visual inspection of the appearance quality of photovoltaic modules, the main purpose of which is to improve the accuracy of visual inspection of appearance quality defects of photovoltaic modules.

[0005] To achieve the above objectives, the present invention provides a method for online visual inspection of the appearance quality of photovoltaic modules, comprising: When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset time sequence, and a high-resolution line array camera is used to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. Extract the grayscale gradient map of each frame in the sequence surface image, and superimpose the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. The sequential surface images are subjected to de-reflection processing to generate a second intermediate image of the sequential surface images; The first intermediate image and the second intermediate image are fused to obtain a high-frequency enhanced reconstructed image. The high-frequency enhanced reconstructed image is then input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module. The adaptive judgment threshold of the high-frequency enhanced reconstructed image is analyzed, and pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map are marked as defect pixels. Connectivity analysis is then performed on the defect pixels to output the defect detection results of the photovoltaic module.

[0006] Optionally, the high-frequency enhanced reconstructed image is input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module, including: The defect segmentation model is used to construct a multi-scale feature extraction path in the backbone network to simultaneously extract large-size defect features and small-size defect features in the high-frequency enhanced reconstructed image. The fusion module of the defect segmentation model is used to fuse the extracted large-size defect features and small-size defect features to generate a fused feature map. Based on the fused feature map, the decision head of the defect segmentation model outputs the probability that each pixel belongs to a defect, thereby obtaining the defect probability distribution map of the photovoltaic module.

[0007] Optionally, when the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence. The multi-angle striped structured light is projected through at least three light source modules arranged in sequence perpendicular to the direction of the conveyor belt. On a cross section perpendicular to the direction of the conveyor belt movement, the angles between the lighting center axes of the first light source module, the second light source module, and the third light source module and the normal direction of the photovoltaic module surface are -45°, 0°, and +45°, respectively. The preset timing control triggers the first light source module, the second light source module, and the third light source module alternately according to a fixed exposure time interval, so that adjacent rows of images acquired by the high-resolution line scan camera correspond to striped structured light at different angles.

[0008] Optionally, a high-resolution linear array camera is used to simultaneously acquire a sequence of surface images of the photovoltaic module under multiple different illumination angles, including: Acquire the real-time pulse signal emitted by the high-resolution encoder installed on the conveyor belt shaft corresponding to the photovoltaic module; The real-time motion speed of the photovoltaic module is calculated based on the real-time pulse signal, and the line frequency of the high-resolution linear array camera is dynamically adjusted based on the real-time motion speed to acquire a sequence of surface images of the photovoltaic module under multiple different illumination angles.

[0009] Optionally, extracting the grayscale gradient map of each frame in the sequence of surface images includes: For each frame of the sequence surface image, convolution operations are performed using the Sobel horizontal operator and the Sobel vertical operator respectively to obtain the horizontal gradient component and the vertical gradient component. The gradient magnitudes of the horizontal and vertical gradient components are calculated to obtain the grayscale gradient image.

[0010] Optionally, the gradient magnitude maxima points in the grayscale gradient image are superimposed to generate a first intermediate image of the sequence surface images, including: Non-maximum suppression processing is performed on the grayscale gradient map of each frame to retain local maxima points in the gradient direction and remove non-edge pixels to obtain the refined gradient feature map corresponding to each frame. The refined gradient feature maps of all frames are subjected to a pixel value up operation in the pixel coordinate dimension to merge the edge features detected at different angles and generate the first intermediate image.

[0011] Optionally, the sequential surface images are subjected to de-reflection processing to generate a second intermediate image of the sequential surface images, including: Calculate the dark channel component map of each frame of the sequence surface image in the RGB color space, and the value of each pixel in the dark channel component map is the minimum value of the pixel in the R, G and B channels; The dark channel component map is smoothed using a guided filtering algorithm to obtain an initial transmittance analysis map; Based on the surface grid line structure characteristics of the photovoltaic module corresponding to the sequence surface images, the transmittance value located in the grid line region in the initial transmittance analysis image is selected as the reference transmittance. Pixels in the initial transmittance analysis diagram that are smaller than the preset ratio of the reference transmittance are identified as reflective high-brightness areas. The reflective highlights in the sequential surface images are color-corrected using an atmospheric scattering model to generate the second intermediate image.

[0012] Optionally, the high-frequency enhanced reconstructed image is input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module, including: The defect segmentation model is used to construct a multi-scale feature extraction path in the backbone network to simultaneously extract large-size defect features and small-size defect features in the high-frequency enhanced reconstructed image. The fusion module of the defect segmentation model is used to fuse the extracted large-size defect features and small-size defect features to generate a fused feature map. Based on the fused feature map, the decision head of the defect segmentation model outputs the probability that each pixel belongs to a defect, thereby obtaining the defect probability distribution map of the photovoltaic module.

[0013] Optionally, the adaptive decision threshold for analyzing the high-frequency enhanced reconstructed image includes: Calculate the global mean brightness and global contrast of the high-frequency enhanced reconstructed image; Based on the global average brightness and global contrast, an adaptive judgment threshold for the high-frequency enhanced reconstructed image is generated through a preset mapping function.

[0014] To address the above problems, the present invention also provides a photovoltaic module appearance quality visual online inspection system, the system comprising: The surface pattern acquisition module is used to control the strobe light source to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence when the photovoltaic module moves into the inspection station with the conveyor belt, and to use a high-resolution line array camera to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. The image noise removal module is used to extract the grayscale gradient map of each frame in the sequence surface image, and to perform superposition operation on the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. An image de-reflection module is used to perform de-reflection processing on the sequence surface images to generate a second intermediate image of the sequence surface images; The image defect probability analysis module is used to fuse the first intermediate image and the second intermediate image to obtain a high-frequency enhanced reconstructed image, and input the high-frequency enhanced reconstructed image into a preset defect segmentation model to output the defect probability distribution map of the photovoltaic module. The image target defect analysis module is used to analyze the adaptive judgment threshold of the high-frequency enhanced reconstructed image, mark the pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map as defect pixels, and perform connectivity analysis on the defect pixels to output the defect detection results of the photovoltaic module.

[0015] This invention effectively overcomes the specular reflection limitations of photovoltaic tempered glass surfaces by projecting multi-angle striped structured light and combining it with synchronous acquisition, obtaining a sequence of images with high dynamic range and motion blur elimination. A first intermediate image is generated by superimposing grayscale gradients, significantly enhancing the edge features of micro-cracks and solving the problem of easily missed detection of small defects. A second intermediate image is generated by combining de-reflection processing of grating features, eliminating strong light interference and restoring realistic texture. Based on the fused high-frequency enhanced reconstructed image, a preset defect segmentation model and adaptive judgment threshold are used to achieve accurate segmentation of defects of different sizes and suppress false background alarms, significantly improving the accuracy and robustness of photovoltaic module defect detection under complex lighting conditions. Therefore, this invention can improve the accuracy of visual detection of appearance quality defects in photovoltaic modules. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an online visual inspection method for the appearance quality of photovoltaic modules according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a module for implementing the online visual inspection method for the appearance quality of photovoltaic modules according to an embodiment of the present invention; Figure 3 A schematic diagram of a computer device for an online visual inspection method for the appearance quality of photovoltaic modules provided in an embodiment of the present invention; The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] This application provides a method for online visual inspection of the appearance quality of photovoltaic modules. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the online visual inspection method for the appearance quality of photovoltaic modules can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0019] Reference Figure 1 The diagram shown is a schematic flowchart of a method for online visual inspection of the appearance quality of photovoltaic modules according to an embodiment of the present invention. In this embodiment, the method for online visual inspection of the appearance quality of photovoltaic modules includes: S1. When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset time sequence, and a high-resolution line array camera is used to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles.

[0020] When a photovoltaic module moves into the testing station with a conveyor belt, the present invention controls a strobe light source to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence. Through the combination of multi-angle striped structured light and strobe light, the limitation of specular reflection caused by a single angle on the surface of photovoltaic tempered glass can be effectively overcome.

[0021] It should be noted that when the photovoltaic module moves into the detection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence. In this process, the multi-angle striped structured light is projected through at least three light source modules arranged in sequence along the direction perpendicular to the movement of the conveyor belt. On a cross section perpendicular to the direction of the conveyor belt movement, the angles between the lighting center axes of the first light source module, the second light source module, and the third light source module and the normal direction of the photovoltaic module surface are -45°, 0°, and +45°, respectively. The preset timing control triggers the first light source module, the second light source module, and the third light source module alternately according to a fixed exposure time interval, so that adjacent rows of images acquired by the high-resolution line scan camera correspond to striped structured light at different angles.

[0022] Wherein, the photovoltaic module refers to the object being inspected, namely, the finished or semi-finished solar panel; the conveyor belt refers to the transmission device in an industrial automated production line used to carry the photovoltaic module and continuously or intermittently transport it to the inspection station; the inspection station refers to the area for image acquisition of the photovoltaic module; the stroboscopic light source refers to an illumination device that can emit high-intensity light in a very short time, usually composed of a high-brightness LED array; the light source module refers to an independent illumination unit containing a light emitter, a focusing lens group, and a grating structured light generator; the illumination central axis refers to the center line of the brightness distribution of the beam emitted by the light source module; the normal direction refers to the imaginary straight line direction perpendicular to the light-receiving surface of the photovoltaic module; and the alternating triggering of the exposure time interval means that the controller activates the first light source, the second light source, and the third light source in sequence according to a fixed period on the time axis, and only one light source is lit at any given time.

[0023] This invention utilizes a high-resolution line scan camera to simultaneously acquire a sequence of surface images of the photovoltaic module under multiple different illumination angles. By utilizing the different angles of illumination of the photovoltaic module, surface features of different directions and depths are revealed. Combined with the synchronous acquisition by the high-resolution line scan camera, motion blur is eliminated while ensuring high-speed production, and a sequence of images containing rich surface details is obtained, laying a high-quality data foundation for the detection of minute defects.

[0024] Specifically, the method of simultaneously acquiring a sequence of surface images of the photovoltaic module under multiple different illumination angles using a high-resolution linear array camera includes: Acquire the real-time pulse signal emitted by the high-resolution encoder installed on the conveyor belt shaft corresponding to the photovoltaic module; The real-time motion speed of the photovoltaic module is calculated based on the real-time pulse signal, and the line frequency of the high-resolution linear array camera is dynamically adjusted based on the real-time motion speed to acquire a sequence of surface images of the photovoltaic module under multiple different illumination angles.

[0025] The high-resolution encoder refers to a high-precision rotary sensor that is directly mounted on the drive roller or shaft of the conveyor belt. The real-time pulse signal refers to the digital square wave signal continuously output by the high-resolution encoder as the conveyor belt rotates. The real-time motion speed refers to the instantaneous linear velocity of the photovoltaic module on the conveyor belt. The high-resolution line scan camera refers to an industrial camera that performs imaging through a pixel sensor. The line frequency refers to the number of times the high-resolution line scan camera acquires pixel rows per second. The sequence surface image refers to the image set of multiple sub-channels acquired by the high-resolution line scan camera.

[0026] S2. Extract the grayscale gradient map of each frame in the sequence surface image, and perform superposition operation on the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image.

[0027] This invention extracts the grayscale gradient map of each frame in the sequence of surface images, and performs a superposition operation on the maximum gradient magnitude points in the grayscale gradient map to generate a first intermediate image of the sequence of surface images. This first intermediate image can filter out the most significant feature edges from the sequence of images, remove stable background noise, and significantly enhance the contour features of micro-cracks in the first intermediate image, thus solving the problem of weak defects being easily missed. The grayscale gradient map... Specifically, extracting the grayscale gradient map of each frame in the sequence surface image includes: For each frame of the sequence surface image, convolution operations are performed using the Sobel horizontal operator and the Sobel vertical operator respectively to obtain the horizontal gradient component and the vertical gradient component. The gradient magnitudes of the horizontal and vertical gradient components are calculated to obtain the grayscale gradient image.

[0028] The Sobel horizontal operator is used to detect edge changes in the vertical direction of an image, and the Sobel vertical operator is used to detect edge changes in the horizontal direction of an image. The horizontal gradient component is the numerical result obtained after each pixel in the image matrix is ​​convolved by the Sobel horizontal operator, and the vertical gradient component is the numerical result obtained after each pixel in the image matrix is ​​convolved by the Sobel vertical operator. The gradient magnitude is the scalar value obtained by vector synthesis of the horizontal and vertical gradient components. The grayscale gradient map is a new image with the same size as the original sequence surface image, but the grayscale value of the pixel represents the gradient magnitude at that position.

[0029] Specifically, the step of superimposing the gradient magnitude maxima points in the grayscale gradient image to generate the first intermediate image of the sequence surface image includes: Non-maximum suppression processing is performed on the grayscale gradient map of each frame to retain local maxima points in the gradient direction and remove non-edge pixels to obtain the refined gradient feature map corresponding to each frame. The refined gradient feature maps of all frames are subjected to a pixel value up operation in the pixel coordinate dimension to merge the edge features detected at different angles and generate the first intermediate image.

[0030] The non-edge pixel refers to a pixel in the neighborhood of an image pixel whose gray-level gradient magnitude has not reached a local maximum value. The thinned gradient feature map refers to the gray-level gradient map after non-maximum suppression processing. The first intermediate image refers to the final gradient feature map generated by fusing the thinned gradient feature maps under all different lighting angles by taking the largest pixel value.

[0031] S3. Perform anti-reflection processing on the sequence surface image to generate a second intermediate image of the sequence surface image.

[0032] The present invention performs de-reflection processing on the sequence surface images to generate a second intermediate image of the sequence surface images, which removes strong light spots and glare interference on the tempered glass surface of the photovoltaic module and restores the true surface texture that was blocked by reflection.

[0033] Specifically, the step of performing de-reflection processing on the sequence surface images to generate a second intermediate image of the sequence surface images includes: Calculate the dark channel component map of each frame of the sequence surface image in the RGB color space, and the value of each pixel in the dark channel component map is the minimum value of the pixel in the R, G and B channels; The dark channel component map is smoothed using a guided filtering algorithm to obtain an initial transmittance analysis map; Based on the surface grid line structure characteristics of the photovoltaic module corresponding to the sequence surface images, the transmittance value located in the grid line region in the initial transmittance analysis image is selected as the reference transmittance. Pixels in the initial transmittance analysis diagram that are smaller than the preset ratio of the reference transmittance are identified as reflective high-brightness areas. The reflective highlights in the sequential surface images are color-corrected using an atmospheric scattering model to generate the second intermediate image.

[0034] Wherein, the dark channel component map refers to the distribution map of opacity in the image calculated based on the dark channel prior theory; the guided filtering algorithm refers to a filtering technique that can smooth image noise while preserving image edge details; the initial transmittance analysis map refers to the image obtained after guiding filtering and smoothing the dark channel component map; the surface grid structure feature refers to the regular geometric arrangement of the silver main grid lines and fine grid lines used to collect current on the surface of the photovoltaic module in the image; the reference transmittance refers to the pixel transmittance value selected in the grid line region as a reference standard in the initial transmittance analysis map; the preset ratio refers to the threshold parameter used to determine the degree of reflectivity, and the preset ratio in this invention can be 0.8; the reflective bright area refers to the image area covered by specular reflection light that is filtered by the above threshold determination logic; the atmospheric scattering model refers to a mathematical and physical model that describes the propagation law of light in a medium containing scattering particles; and the second intermediate image refers to the image generated after the reflective bright area in the sequence surface image has been corrected by the atmospheric scattering model and color restored, and the surface specular reflection has been removed.

[0035] Optionally, the atmospheric scattering model is constructed using the following formula: ; in, Represents a sequence of surface images. This represents the second intermediate image. This represents the atmospheric light value analyzed through the reflective bright areas.

[0036] The atmospheric light value refers to a global scalar value representing the color and intensity of the reflected light source, obtained by analyzing the bright reflective areas in the image.

[0037] S4. The first intermediate image and the second intermediate image are fused to obtain a high-frequency enhanced reconstructed image. The high-frequency enhanced reconstructed image is then input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module.

[0038] This invention fuses the first intermediate image and the second intermediate image to obtain a high-frequency enhanced reconstructed image that possesses both clear defect edges and realistic background texture. Specifically, the high-frequency enhanced reconstructed image refers to the final image obtained by fusing the first and second intermediate images, significantly enhancing the high-frequency details of the second intermediate image while preserving its clear base texture. More specifically, the high-frequency enhanced reconstructed image is fused using a Laplacian pyramid technique.

[0039] The present invention inputs the high-frequency enhanced reconstructed image into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module, which can greatly improve the feature extraction capability of small defects in complex backgrounds, and the output defect probability distribution map is more accurate, effectively reducing the false alarm rate caused by background textures (such as grid lines).

[0040] Specifically, the step of inputting the high-frequency enhanced reconstructed image into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module includes: The defect segmentation model is used to construct a multi-scale feature extraction path in the backbone network to simultaneously extract large-size defect features and small-size defect features in the high-frequency enhanced reconstructed image. The fusion module of the defect segmentation model is used to fuse the extracted large-size defect features and small-size defect features to generate a fused feature map. Based on the fused feature map, the decision head of the defect segmentation model outputs the probability that each pixel belongs to a defect, thereby obtaining the defect probability distribution map of the photovoltaic module.

[0041] The defect segmentation model refers to a deep learning model that can segment a high-frequency enhanced reconstructed image into foreground (defect) and background. The large-size defect feature refers to the visual features of defects occupying a large pixel area in the high-frequency enhanced reconstructed image, such as large scratches, stains, and fragments. The small-size defect feature refers to the visual features of defects occupying a small pixel area in the high-frequency enhanced reconstructed image, such as minute hidden cracks, pinholes, and tiny scratches. The fusion module is the core component that integrates feature maps of different levels. The feature map refers to the representation of the high-frequency enhanced reconstructed image at a specific level of abstraction. The decision head is the output layer that converts the output value of each pixel in the high-frequency enhanced reconstructed image into a probability between 0 and 1. The defect probability distribution map is a grayscale image of the same size as the high-frequency enhanced reconstructed image, where the grayscale value (0-255) of each pixel represents the probability that the pixel belongs to the defect. For example, a high grayscale value (e.g., 255) indicates that the pixel is highly certain by the model to be a defect, and a low grayscale value (e.g., 0) indicates that the pixel is highly certain by the model to be the background and not a defect.

[0042] S5. Analyze the adaptive judgment threshold of the high-frequency enhanced reconstructed image, mark the pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map as defect pixels, and perform connectivity analysis on the defect pixels to output the defect detection results of the photovoltaic module.

[0043] The adaptive judgment threshold of the high-frequency enhanced reconstructed image analyzed in this invention can dynamically adjust the detection standard according to the real-time illumination and noise conditions of the current high-frequency enhanced reconstructed image, thus avoiding the limitations of a fixed threshold under different illumination conditions.

[0044] In detail, the adaptive decision threshold for analyzing the high-frequency enhanced reconstructed image includes: Calculate the global mean brightness and global contrast of the high-frequency enhanced reconstructed image; Based on the global average brightness and global contrast, an adaptive judgment threshold for the high-frequency enhanced reconstructed image is generated through a preset mapping function.

[0045] Wherein, the global average brightness value refers to the arithmetic mean of the brightness values ​​of all pixels in the high-frequency enhanced reconstructed image, the global contrast ratio refers to the dispersion of the pixel brightness values ​​in the high-frequency enhanced reconstructed image, and the adaptive judgment threshold refers to the final standard for distinguishing between "defects" and "background".

[0046] It should be noted that the mapping function is obtained by collecting a large number of photovoltaic module images with different illumination and contrast, manually annotating them, and fitting the "defect probability distribution map" and the corresponding "real defect label".

[0047] Finally, this invention marks pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map as defect pixels, and performs connectivity analysis on these defect pixels to output the defect detection results of the photovoltaic module, thereby achieving accurate analysis of photovoltaic module defects. Here, the defect pixel refers to a specific pixel in the defect probability distribution map corresponding to the high-frequency enhanced reconstructed image whose probability value exceeds the adaptive judgment threshold. The connectivity analysis refers to classifying all spatially adjacent defect pixels in the image that satisfy specific connectivity rules into the same independent region. These specific connectivity rules can be 4-connectivity (considering only vertical, horizontal, and lateral adjacency) or 8-connectivity (considering vertical, horizontal, left-right, and diagonal adjacency). The defect detection result refers to the final output data set that quantitatively and qualitatively describes the surface defect state of the photovoltaic module, including defect location, defect category, defect size, and defect severity.

[0048] This invention effectively overcomes the specular reflection limitations of photovoltaic tempered glass surfaces by projecting multi-angle striped structured light and combining it with synchronous acquisition, obtaining a sequence of images with high dynamic range and motion blur elimination. A first intermediate image is generated by superimposing grayscale gradients, significantly enhancing the edge features of micro-cracks and solving the problem of easily missed detection of small defects. A second intermediate image is generated by combining de-reflection processing of grating features, eliminating strong light interference and restoring realistic texture. Based on the fused high-frequency enhanced reconstructed image, a preset defect segmentation model and adaptive judgment threshold are used to achieve accurate segmentation of defects of different sizes and suppress false background alarms, significantly improving the accuracy and robustness of photovoltaic module defect detection under complex lighting conditions. Therefore, this invention can improve the accuracy of visual detection of appearance quality defects in photovoltaic modules.

[0049] like Figure 2 The diagram shown is a functional block diagram of the photovoltaic module appearance quality visual online inspection system of the present invention.

[0050] Surface pattern acquisition module, image noise removal module, image de-reflection module Image Defect Probability Analysis Module Image Target Defect Analysis Module The photovoltaic module appearance quality visual online inspection system 200 of this invention can be installed in an electronic device. Depending on the functions implemented, the photovoltaic module appearance quality visual online inspection system may include a surface pattern acquisition module 201, an image noise removal module 202, an image de-reflection module 203, an image defect probability analysis module 204, and an image target defect analysis module 205. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0051] In this embodiment of the invention, the functions of each module / unit are as follows: The surface pattern acquisition module 201 is used to control the strobe light source to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence when the photovoltaic module moves into the inspection station with the conveyor belt, and to use a high-resolution line array camera to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. The image noise removal module 202 is used to extract the grayscale gradient map of each frame in the sequence surface image, and to perform superposition operation on the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. The image de-reflection module 203 is used to perform de-reflection processing on the sequence surface image to generate a second intermediate image of the sequence surface image; The image defect probability analysis module 204 is used to fuse the first intermediate image and the second intermediate image to obtain a high-frequency enhanced reconstructed image, and input the high-frequency enhanced reconstructed image into a preset defect segmentation model to output the defect probability distribution map of the photovoltaic module. The image target defect analysis module 205 is used to analyze the adaptive judgment threshold of the high-frequency enhanced reconstructed image, mark the pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map as defect pixels, and perform connectivity analysis on the defect pixels to output the defect detection results of the photovoltaic module.

[0052] In detail, the modules in the photovoltaic module appearance quality visual online inspection system 200 described in this embodiment of the invention adopt the same methods as described above during use. Figure 1 The method used is the same as the online visual inspection method for the appearance quality of photovoltaic modules described above, and it can produce the same technical effect, so it will not be repeated here.

[0053] In one embodiment, a computer device is provided, which may be a server or a client, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a visual online inspection method for the appearance quality of photovoltaic modules on the server or client side.

[0054] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset time sequence, and a high-resolution line array camera is used to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. Extract the grayscale gradient map of each frame in the sequence surface image, and superimpose the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. The sequential surface images are subjected to de-reflection processing to generate a second intermediate image of the sequential surface images; The first intermediate image and the second intermediate image are fused to obtain a high-frequency enhanced reconstructed image. The high-frequency enhanced reconstructed image is then input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module. The adaptive judgment threshold of the high-frequency enhanced reconstructed image is analyzed, and pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map are marked as defect pixels. Connectivity analysis is then performed on the defect pixels to output the defect detection results of the photovoltaic module.

[0055] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset time sequence, and a high-resolution line array camera is used to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. Extract the grayscale gradient map of each frame in the sequence surface image, and superimpose the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. The sequential surface images are subjected to de-reflection processing to generate a second intermediate image of the sequential surface images; The first intermediate image and the second intermediate image are fused to obtain a high-frequency enhanced reconstructed image. The high-frequency enhanced reconstructed image is then input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module. The adaptive judgment threshold of the high-frequency enhanced reconstructed image is analyzed, and pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map are marked as defect pixels. Connectivity analysis is then performed on the defect pixels to output the defect detection results of the photovoltaic module.

[0056] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0057] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0058] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0059] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0060] Finally, it should be noted that in the above embodiments, each embodiment can be combined with each other or independent. Deleting any one of them will not affect the technical implementation of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for online visual inspection of the appearance quality of photovoltaic modules, characterized in that, The method includes: When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset time sequence, and a high-resolution line array camera is used to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. Extract the grayscale gradient map of each frame in the sequence surface image, and superimpose the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. The sequential surface images are subjected to de-reflection processing to generate a second intermediate image of the sequential surface images; The first intermediate image and the second intermediate image are fused to obtain a high-frequency enhanced reconstructed image. The high-frequency enhanced reconstructed image is then input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module. The adaptive judgment threshold of the high-frequency enhanced reconstructed image is analyzed, and pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map are marked as defect pixels. Connectivity analysis is then performed on the defect pixels to output the defect detection results of the photovoltaic module.

2. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, The high-frequency enhanced reconstructed image is input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module, including: The defect segmentation model is used to construct a multi-scale feature extraction path in the backbone network to simultaneously extract large-size defect features and small-size defect features in the high-frequency enhanced reconstructed image. The fusion module of the defect segmentation model is used to fuse the extracted large-size defect features and small-size defect features to generate a fused feature map. Based on the fused feature map, the decision head of the defect segmentation model outputs the probability that each pixel belongs to a defect, thereby obtaining the defect probability distribution map of the photovoltaic module.

3. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, When the photovoltaic module moves into the inspection station with the conveyor belt, the strobe light source is controlled to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence. The multi-angle striped structured light is projected through at least three light source modules arranged in sequence along the direction perpendicular to the movement of the conveyor belt. On a cross section perpendicular to the direction of the conveyor belt movement, the angles between the lighting center axes of the first light source module, the second light source module, and the third light source module and the normal direction of the photovoltaic module surface are -45°, 0°, and +45°, respectively. The preset timing control triggers the first light source module, the second light source module, and the third light source module alternately according to a fixed exposure time interval, so that adjacent rows of images acquired by the high-resolution line scan camera correspond to striped structured light at different angles.

4. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, The photovoltaic module was simultaneously acquired using a high-resolution linear array camera under multiple different illumination angles, including: Acquire the real-time pulse signal emitted by the high-resolution encoder installed on the conveyor belt shaft corresponding to the photovoltaic module; The real-time motion speed of the photovoltaic module is calculated based on the real-time pulse signal, and the line frequency of the high-resolution linear array camera is dynamically adjusted based on the real-time motion speed to acquire a sequence of surface images of the photovoltaic module under multiple different illumination angles.

5. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, Extracting the grayscale gradient map of each frame in the sequence of surface images includes: For each frame of the sequence surface image, convolution operations are performed using the Sobel horizontal operator and the Sobel vertical operator respectively to obtain the horizontal gradient component and the vertical gradient component. The gradient magnitudes of the horizontal and vertical gradient components are calculated to obtain the grayscale gradient image.

6. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, The gradient magnitude maxima points in the grayscale gradient image are superimposed to generate the first intermediate image of the sequence surface image, including: Non-maximum suppression processing is performed on the grayscale gradient map of each frame to retain local maxima points in the gradient direction and remove non-edge pixels to obtain the refined gradient feature map corresponding to each frame. The refined gradient feature maps of all frames are subjected to a pixel value up operation in the pixel coordinate dimension to merge the edge features detected at different angles and generate the first intermediate image.

7. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, Performing de-reflection processing on the sequence of surface images to generate a second intermediate image of the sequence of surface images includes: Calculate the dark channel component map of each frame of the sequence surface image in the RGB color space, and the value of each pixel in the dark channel component map is the minimum value of the pixel in the R, G and B channels; The dark channel component map is smoothed using a guided filtering algorithm to obtain an initial transmittance analysis map; Based on the surface grid line structure characteristics of the photovoltaic module corresponding to the sequence surface images, the transmittance value located in the grid line region in the initial transmittance analysis image is selected as the reference transmittance. Pixels in the initial transmittance analysis diagram that are smaller than the preset ratio of the reference transmittance are identified as reflective high-brightness areas. The reflective highlights in the sequence of surface images are color-corrected using an atmospheric scattering model to generate the second intermediate image.

8. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, The high-frequency enhanced reconstructed image is input into a preset defect segmentation model to output a defect probability distribution map of the photovoltaic module, including: The defect segmentation model is used to construct a multi-scale feature extraction path in the backbone network to simultaneously extract large-size defect features and small-size defect features in the high-frequency enhanced reconstructed image. The fusion module of the defect segmentation model is used to fuse the extracted large-size defect features and small-size defect features to generate a fused feature map. Based on the fused feature map, the decision head of the defect segmentation model outputs the probability that each pixel belongs to a defect, thereby obtaining the defect probability distribution map of the photovoltaic module.

9. The online visual inspection method for the appearance quality of photovoltaic modules as described in claim 1, characterized in that, The analysis of the adaptive decision threshold of the high-frequency enhanced reconstructed image includes: Calculate the global mean brightness and global contrast of the high-frequency enhanced reconstructed image; Based on the global average brightness and global contrast, an adaptive judgment threshold for the high-frequency enhanced reconstructed image is generated through a preset mapping function.

10. A photovoltaic module appearance quality visual online inspection system, characterized in that, The system includes: The surface pattern acquisition module is used to control the strobe light source to project multi-angle striped structured light onto the photovoltaic module according to a preset timing sequence when the photovoltaic module moves into the inspection station with the conveyor belt, and to use a high-resolution line array camera to simultaneously acquire sequential surface images of the photovoltaic module under multiple different illumination angles. The image noise removal module is used to extract the grayscale gradient map of each frame in the sequence surface image, and to perform superposition operation on the gradient magnitude maxima points in the grayscale gradient map to generate the first intermediate image of the sequence surface image. An image de-reflection module is used to perform de-reflection processing on the sequence surface images to generate a second intermediate image of the sequence surface images; The image defect probability analysis module is used to fuse the first intermediate image and the second intermediate image to obtain a high-frequency enhanced reconstructed image, and input the high-frequency enhanced reconstructed image into a preset defect segmentation model to output the defect probability distribution map of the photovoltaic module. The image target defect analysis module is used to analyze the adaptive judgment threshold of the high-frequency enhanced reconstructed image, mark the pixels with probability values ​​greater than the adaptive judgment threshold in the defect probability distribution map as defect pixels, and perform connectivity analysis on the defect pixels to output the defect detection results of the photovoltaic module.

Images