A photoluminescence image processing method and system

Abstract

The present application relates to the technical field of photoluminescence, and particularly relates to a photoluminescence image processing method and system, which comprises: performing fusion processing on images to obtain a fusion image; wherein the method comprises the following steps: aligning corresponding edge pixel points in at least two images respectively; calculating average values of the edge pixel points; aligning corresponding grid line pixel points in the at least two images; calculating average values of the grid line pixel points; aligning pixel points of first feature points in the at least two images; calculating average values of the first feature point pixel points; and obtaining the fusion image according to the average values of the edge pixel points, the average values of the grid line pixel points and the average values of the first feature point pixel points. The present application can reduce the strong dependence on external control facilities, reduce the operation difficulty and cost of outdoor photoluminescence detection, and can also identify and exclude dynamic abnormal areas based on limited image sample quantities in the preprocessing stage, so as to exclude the interference of some non-defect factors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photoluminescence, in particular to a photoluminescence image processing method and system. BACKGROUND

[0002] The global energy structure is accelerating towards clean transformation, and photovoltaic power generation, as one of the core renewable energies, requires higher quality and reliability for its large-scale application. Photovoltaic modules are the core components of power stations, and their performance (such as conversion efficiency and durability) directly affects the power generation efficiency and economic benefits.

[0003] Due to the long-term exposure of photovoltaic modules to harsh environments such as high temperature, humidity, salt spray, and ultraviolet light, problems such as material aging, hidden cracks, hot spots, and PID (potential-induced degradation) are easily caused, leading to a decrease in power generation efficiency and even fire risk. For example, hidden cracks can block current transmission, and hot spots can cause local energy loss.

[0004] Among common detection methods, I-V detection can cause damage to photovoltaic modules due to reverse power supply, infrared thermal imaging detection has limited detection range, and electroluminescence (EL) detection can only be operated in a dark room, all of which have certain defects.

[0005] Photoluminescence technology (PL) can well solve the above problems. Currently, some solutions for detecting defects in photovoltaic modules using photoluminescence technology are provided in the prior art.

[0006] For example, patent application CN118758946A discloses a photovoltaic module defect failure in-situ optical imaging detection method and system, which controls the photoelectric effect of local cell pieces of the photovoltaic module through a switching unit to realize switching between different working states of the photovoltaic module.

[0007] For another example, patent application CN119941689A discloses a photovoltaic panel defect detection method, device, electronic equipment, and program product, which detects significant defects on a PL image of a photovoltaic panel through an efficient gray-scale analysis-based image processing algorithm, and further detects subtle defects that are not detected by the aforementioned method through a deep learning algorithm with high detection accuracy.

[0008] For another example, patent application CN119880917A discloses a photovoltaic panel invisible defect detection device based on high dynamic range imaging, relating to the technical field of solar cell panel application, the device comprises a light-emitting component, a face array camera component, and a control component; the light-emitting component is used to act on the photovoltaic panel to make the photovoltaic panel produce photoluminescence or electroluminescence; the face array camera component is used to receive the exposure instruction issued by the control component to change the exposure amount; the photoluminescence image or the electroluminescence image of the photovoltaic panel under different exposure amounts is shot; the control component is further used to fuse the photoluminescence images of the photovoltaic panel under different exposure amounts to obtain the photoluminescence defect image of the photovoltaic panel; the electroluminescence images of the photovoltaic panel under different exposure amounts are fused to obtain the electroluminescence defect image of the photovoltaic panel.

[0009] However, the applicant notices that the technical route adopted by these methods is to rely on external signal control to obtain photovoltaic panel images under multiple states, which is difficult to meet the outdoor detection needs of large-area photovoltaic panels. SUMMARY

[0010] The purpose of the present application is to provide a photoluminescence image processing method and system, which partially solves or alleviates the above-mentioned deficiencies in the prior art, and can obtain photoluminescence images and identify photovoltaic component defects under the same natural light conditions without relying on external signal control.

[0011] In order to solve the above-mentioned technical problems, the present application specifically adopts the following technical solutions:

[0012] The first aspect of the present application is to provide a photoluminescence image processing method, which comprises:

[0013] The photovoltaic component comprises a panel, the panel has a frame, and the panel comprises a plurality of grid lines;

[0014] The images are fused to obtain a fused image; wherein, the steps include:

[0015] The corresponding edge pixel points in at least two of the images are aligned respectively;

[0016] The average values of the corresponding edge pixel points in at least two of the images are calculated;

[0017] The corresponding grid line pixel points in at least two of the images are aligned;

[0018] The average values of the corresponding grid line pixel points in at least two of the images are calculated;

[0019] The pixel points of the first feature points in at least two of the images are aligned;

[0020] Calculate the average value of corresponding first feature points in at least two images;

[0021] Obtain the fusion image according to the average value of the edge pixel, the average value of the grid line pixel, and the average value of the first feature point pixel.

[0022] In some embodiments, the method comprises selecting a target image and at least two contrast images from the images, and identifying an abnormal point through comparison of the target image and the contrast images; wherein the method comprises the steps of:

[0023] Selecting the target image and at least two contrast images from a plurality of images;

[0024] Calculating at least two pixel difference values between a target pixel of the target image and a plurality of corresponding pixels of at least two contrast images; wherein the pixels include the edge pixel, the grid line pixel, and the first feature point pixel;

[0025] When the at least two pixel difference values corresponding to the target pixel are all greater than a preset pixel threshold value, the target pixel is identified as an abnormal point.

[0026] In some embodiments, the method comprises regarding one abnormal point as an abnormal region;

[0027] Or, regarding a plurality of adjacent abnormal points as the same abnormal region, the method comprises the steps of:

[0028] Traversing the abnormal points to identify at least two abnormal points with a distance less than a preset distance threshold value as an abnormal point set;

[0029] Identifying a region corresponding to the abnormal point set as the abnormal region;

[0030] Performing abnormal processing on the abnormal region.

[0031] In some embodiments, the abnormal processing on the abnormal region comprises:

[0032] Determining whether the abnormal region is a dynamic abnormal region;

[0033] Performing dynamic abnormal processing on the dynamic abnormal region.

[0034] In some embodiments, the method comprises:

[0035] Determining whether the abnormal region is the dynamic abnormal region, which comprises:

[0036] Determining whether there are at least two target images that both have the abnormal region;

[0037] If yes, calculate shape similarity between at least two of the abnormal regions;

[0038] When the shape similarity between at least two of the abnormal regions is greater than a preset similarity threshold, it is considered that at least two of the abnormal regions are similar;

[0039] Then, the similar at least two of the abnormal regions are identified as one of the dynamic abnormal regions.

[0040] In some embodiments, comprising:

[0041] Before identifying at least two of the abnormal regions as one of the dynamic abnormal regions, further comprising the step of:

[0042] Obtaining the shooting time sequence between multiple target images, and obtaining the target position of each similar abnormal region in multiple target images;

[0043] According to the shooting time sequence and the target position, the moving speed of the abnormal region is calculated;

[0044] When the moving speed is greater than a preset speed threshold, the corresponding abnormal region is identified as the dynamic abnormal region.

[0045] In some embodiments, comprising:

[0046] The fusion image is input into a photovoltaic module defect AI autonomous identification model;

[0047] The defect type of the photovoltaic module is identified;

[0048] According to the defect type, a defect image of the photovoltaic module is generated;

[0049] A corresponding preset defect type label is labeled at a corresponding position in the defect image.

[0050] The second aspect of the application provides a photoluminescence image processing system, comprising:

[0051] The photovoltaic module comprises a panel, the panel has a frame, and the panel comprises a plurality of grid lines;

[0052] An image fusion module is configured to perform fusion processing on images to obtain a fusion image; wherein the system comprises:

[0053] An edge pixel point alignment unit is configured to align corresponding edge pixel points in at least two of the images, respectively;

[0054] An edge pixel point calculation unit is configured to calculate the average value of corresponding edge pixel points in at least two of the images;

[0055] The grid line pixel point alignment unit is configured to align corresponding grid line pixel points in at least two images.

[0056] The grid line pixel point calculation unit is configured to calculate average values of corresponding grid line pixel points in at least two images.

[0057] The first feature point alignment unit is configured to align pixel points of first feature points in at least two images.

[0058] The first feature point calculation unit is configured to calculate average values of corresponding first feature point pixel points in at least two images.

[0059] The image fusion unit is configured to obtain the fusion image according to the average values of the edge pixel points, the average values of the grid line pixel points, and the average values of the first feature point pixel points.

[0060] In some embodiments, the image comparison module comprises:

[0061] The image comparison module is configured to select a target image and at least two comparison images from the images, and identify an abnormal point through comparison of the target image and the comparison images.

[0062] The image selection unit is configured to select the target image and at least two comparison images from a plurality of images.

[0063] The pixel difference calculation unit is configured to calculate at least two pixel difference values of corresponding pixel points between a target pixel point of the target image and at least two comparison images, wherein the pixel points include the edge pixel points, the grid line pixel points, and the pixel points of the first feature points.

[0064] The abnormal point identification unit is configured to identify the target pixel point as an abnormal point when at least two pixel difference values corresponding to the target pixel point are greater than a preset pixel threshold.

[0065] In some embodiments, the image comparison module comprises:

[0066] The abnormal region identification unit is configured to regard one abnormal point as an abnormal region, or regard a plurality of adjacent abnormal points as the same abnormal region, and comprises the following steps:

[0067] The abnormal point set identification unit is configured to traverse the abnormal points to identify at least two abnormal points with a distance less than a preset distance threshold as an abnormal point set.

[0068] The abnormal region identification subunit is configured to identify a region corresponding to the abnormal point set as the abnormal region.

[0069] An abnormality processing subunit is configured to perform abnormality processing on the abnormal region.

[0070] Beneficial technical effects:

[0071] The application sets a step-by-step filtering method in the acquisition stage of the photoluminescence image, which can reduce the strong dependence on external control facilities, reduce the operation difficulty and cost of outdoor photoluminescence detection, and further identify and exclude dynamic abnormal regions based on limited image sample quantity in the preprocessing stage, thereby excluding the interference of part of non-defect factors.

[0072] 1. The step-by-step filtering rule is used, first, the spectrum of the photovoltaic module image is within the preset spectrum range through spectral wavelength filtering, and then the difference between the photovoltaic module images in the open / short circuit working state is calculated, thereby filtering the solar intensity. Further, direct imaging under the same natural light condition can be realized without the need of external light source, covering the module or moving to a dark room, which greatly reduces the cost and difficulty of photoluminescence image acquisition.

[0073] 2. The photoluminescence image under the same natural light condition is preprocessed, the dynamic abnormal region in the photoluminescence image is identified by comparison, the noise of non-defect factors in the defect identification process is effectively filtered, and the photovoltaic module defect identification efficiency is greatly improved.

[0074] 3. The dynamic abnormal region is filtered out by the conservative exclusion strategy, and the non-defect interference factors are excluded, which not only filters out the non-defect interference factors, but also limits the degree of filtering out the non-defect interference factors, greatly saves the algorithm resources, and effectively avoids the case that the instantaneous sudden phenomenon is misjudged as photovoltaic module failure.

[0075] 4. The photovoltaic module image is segmented and processed, and the corresponding level of non-defect factor removal rules is formulated according to the relative position and range of the dynamic abnormal region, which is beneficial to the maximization of the utilization rate of the photoluminescence image. BRIEF DESCRIPTION OF DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. In all the drawings, similar elements or parts are generally identified by similar reference signs. In the drawings, each element or part is not necessarily drawn according to the actual proportion. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without any creative labor.

[0077] Figure 1 The method flowchart in an exemplary embodiment of the present application is shown in the figure.

[0078] Figure 2An actual defect annotation diagram in an exemplary embodiment of the present application;

[0079] Figure 3 A predicted defect annotation diagram in an exemplary embodiment of the present application;

[0080] Figure 4 A representative image selection diagram in an exemplary embodiment of the present application;

[0081] Figure 5 A defect recognition model structure diagram in an exemplary embodiment of the present application;

[0082] Figure 6 Another actual defect diagram in an exemplary embodiment of the present application;

[0083] Figure 7 Another actual defect diagram in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0084] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0085] Herein, the suffix such as "module", "part" or "unit" used for representing an element is only for the convenience of description of the present application, and has no specific meaning by itself. Therefore, "module", "part" or "unit" can be mixedly used.

[0086] Herein, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of description of the present application and simplification of the description, and do not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0087] In this document, unless otherwise indicated and / or unless the context clearly dictates otherwise, the terms "mounting", "provided with", "connected" and the like are to be interpreted broadly, for example, "connected" can be fixed connection, can be detachable connection, or integrally connected; can be mechanical connection, can be direct connection, can be indirect connection through intermediate medium, can be internal connection of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0088] In this document, "and / or" includes any and all combinations of one or more of the associated items.

[0089] In this document, "a plurality of" means two or more, that is, it includes two, three, four, five, etc.

[0090] In this specification, the term "about" typically means + / - 5% of the stated value, more typically + / - 4% of the stated value, more typically + / - 3% of the stated value, more typically + / - 2% of the stated value, even more typically + / - 1% of the stated value, even more typically + / - 0.5% of the stated value.

[0091] In this specification, certain embodiments can be disclosed in one format in terms of a range. It should be understood that such "in terms of a range" description is merely for the convenience and brevity and should not be construed as a rigid limitation on the disclosed range. Therefore, the description of the range should be considered to have specifically disclosed all possible sub-ranges and individual numerical values within the range. For example, the description of the range 1-6 should be considered to have specifically disclosed sub-ranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., as well as individual numbers within the range, such as 1, 2, 3, 4, 5 and 6. The above rule applies regardless of the breadth of the range.

[0092] Glossary:

[0093] Photoluminescence refers to a physical process in which a substance absorbs light energy, electrons jump to an excited state, and then release energy through radiative transition to emit light of a specific wavelength.

[0094] Photoluminescence image is an image formed by fluorescence or phosphorescence emitted by a photovoltaic module after being excited by light. For example, in the present application, the data picture obtained by subtracting the picture taken in the open circuit state from the picture taken in the short circuit state of the photovoltaic module can be directly used as the photoluminescence image.

[0095] Photovoltaic module (solar panel) is a device that encapsulates a plurality of solar cells together for converting sunlight directly into electrical energy.

[0096] The grid lines in a photovoltaic module are thin metal lines (usually silver) on the surface of the solar cell that collect current, they collect the current generated inside the solar cell and transmit it to the main grid line, while minimizing the shading of incident sunlight.

[0097] A bandpass filter is an optical element that allows only a specific range of wavelengths of light to pass through, while blocking other wavelengths.

[0098] InGaAs sensors are photodetectors based on indium gallium arsenide (InGaAs) semiconductor materials, specifically designed to detect light in the short-wave infrared band (usually 900-1700 nanometers), with high sensitivity, commonly used in night vision, spectral analysis, and industrial detection.

[0099] Embodiment one:

[0100] See Figure 1 The present application provides a photoluminescence-based photovoltaic module defect detection method, which can greatly reduce the cost and difficulty of obtaining photoluminescence images, including the following steps:

[0101] When the photovoltaic module is in an open circuit state, filter out light outside the first wavelength range to reduce the collected wavelength range, and then take a first image;

[0102] When the photovoltaic module is in a short circuit state, filter out light outside the first wavelength range to reduce the collected wavelength range, and then take a second image;

[0103] Subtract the second image from the first image to reduce the intensity of sunlight, and obtain a third image.

[0104] That is, the present application provides a method for obtaining photoluminescence images based on wavelength and light intensity step-by-step filtering means, it is worth noting that the step-by-step filtering means provided by the present application can reduce the strong dependence on external facilities (such as lamp control equipment or current control equipment), and reduce the operation difficulty and cost of outdoor photoluminescence detection.

[0105] In some embodiments, the working state (open circuit state or short circuit state) of the photovoltaic module can be controlled according to the inverter or other current control equipment.

[0106] In some embodiments, the specific steps of controlling the working state of the photovoltaic module according to the inverter can be: S1, mounting the photoluminescence camera on a carrier device; S2, using the photoluminescence camera to take a photovoltaic module image when the inverter is in a short circuit state; S3, changing the inverter from a short circuit state to an open circuit state; S4, using the photoluminescence camera to take a photovoltaic module image when the inverter is in an open circuit state.

[0107] It is worth noting that the step-by-step filtering rule proposed by the present application at least contains the following two meanings:

[0108] 1) Filter out the spectral range of the image during shooting, so that only the spectrum within the preset spectral range is displayed in the photovoltaic component image, thereby filtering out stray light for subsequent direct acquisition of photoluminescence images; 2) Subtract the photovoltaic component images under open-circuit and short-circuit working conditions, thereby canceling out the sunlight (in some embodiments, natural light) for subsequent accurate acquisition of photoluminescence images. Under the synergistic action of the step-by-step filtering rule, the acquisition process of the photoluminescence image is greatly simplified, without the need for external signal control, and the accuracy of photoluminescence defect identification can be further improved from the data source level (i.e., the acquisition of the photoluminescence image).

[0109] In some embodiments, the present application presets the spectral range where the photovoltaic cell emits light and the sunlight is relatively weak as the first wavelength range. Filtering out the first wavelength range can effectively avoid the PL signal being completely overwhelmed by sunlight during imaging (the intensity of sunlight (>1000 W / m²) is usually much higher than the light intensity of the photovoltaic component (usually <1 mW / m²)), which is conducive to filtering out the photoluminescence intensity of the photovoltaic component more obviously, and greatly avoids the interference of sunlight on the photoluminescence image acquisition process.

[0110] In some embodiments, the step-by-step filtering can also remove stray light outside the first wavelength range, avoiding the interference of the reflected light on the photoluminescence image. For images obtained based on the step-by-step filtering steps, please refer to Figure 6 .

[0111] Unlike the patent application CN118758946A, which uses narrowband correlation operation to filter out the light intensity of the component defect part, the present application shoots the images of the photovoltaic component under open-circuit and short-circuit working conditions, respectively obtains the image under open-circuit (the image illumination intensity is the sunlight intensity) and the image under short-circuit (the image illumination intensity is the sum of the sunlight intensity and the photoluminescence intensity under short-circuit), and then performs difference operation on the two shot images, thereby realizing the "cancellation" of sunlight and obtaining the photoluminescence image of the photovoltaic component under open-circuit, which greatly simplifies the photoluminescence image acquisition process while ensuring the accuracy of the calculation.

[0112] Preferably, the photoluminescence image of the photovoltaic component can be exported in TIFF format to support lossless compression or even no compression, which can effectively reduce the jitter between the two images (the jitter between the images is usually manifested as loss of high-frequency details), thereby maximizing the retention of complete pixel data for the subsequent photovoltaic component defect identification process and ensuring the image quality.

[0113] In some embodiments, it also includes:

[0114] taking at least two of the first image and the second image;

[0115] The taking environment / taking condition between the first image or the second image is the same or the approximation degree is higher than a preset environment / condition approximation degree.

[0116] In some embodiments, the taking environment / taking condition between the images in the two working states is ensured to be the same or the approximation degree is higher than a preset environment / condition approximation degree, which can be that the light intensity difference is less than a first difference threshold value, so that the light intensity of the sunlight can be completely or approximately completely offset, and the purpose of the step-by-step filtering rule proposed in the application can be better achieved.

[0117] In some embodiments, the time interval for taking the photovoltaic module in the same group and the same working state can be set to be less than a first time interval threshold value (such as 0.1 seconds). The sunlight has little change in the extremely short time interval, and the offset of the sunlight intensity can be achieved or approximately achieved after subtraction.

[0118] In some embodiments, the method further comprises:

[0119] The first image and the second image are taken by a camera with a sensor, and the responsivity of the sensor to light in the first wavelength range is greater than a preset responsivity threshold value.

[0120] Preferably, the bandpass filter and the InGaAs sensor can be used in combination for spectral wavelength filtering, so that the photoluminescence image can be directly obtained under natural light conditions without the need for an additional light source, covering the module, or moving to a dark room, which greatly improves the detection efficiency and effectively reduces the cost and difficulty of obtaining the photoluminescence image.

[0121] In some embodiments, the bandpass filter with a wavelength range of 1120nm-1150nm can be preferably used.

[0122] In some embodiments, the InGaAs sensor is selected by taking advantage of the performance characteristic that the responsivity of the InGaAs sensor to light in the spectral range where the photovoltaic cell emits light and the sunlight is weak is higher than that of a common silicon sensor. The InGaAs sensor can efficiently respond to light in the purified spectral range, enhance the defect identification capability under outdoor sunlight conditions, and improve the photovoltaic module defect identification efficiency.

[0123] In some embodiments, before step S102, the method further comprises:

[0124] determining a taking parameter according to an actual working condition;

[0125] The taking parameter at least includes exposure, definition, and light exposure time.

[0126] Preferably, the exposure, and / or the definition, and / or the exposure time of the shooting camera can be set according to the actual working condition to further balance the cost of computing power and the requirement of accuracy, so as to improve the efficiency of photoluminescence image acquisition.

[0127] In some embodiments, a feature point detection algorithm can also be used to pre-filter the acquired images, thereby reducing the computing pressure of the photovoltaic module defect recognition algorithm. Specifically, the regular cell grid lines, frame corner points and other rich features on the surface of the photovoltaic module are used to calculate a homography matrix, and a projection transformation alignment is realized, which can to some extent solve the problem of poor photovoltaic module image acquisition effect caused by shooting angle deviation, and further pre-filter the noise generated by non-defect factors in the defect recognition process, avoid the interference of non-defect factors on the defect recognition process, and greatly improve the photovoltaic module defect recognition efficiency.

[0128] In summary, the present application can meet the outdoor detection needs of large-area photovoltaic modules without relying on external signal control, and the detection effect is shown in Figure 7 .

[0129] Embodiment two:

[0130] In some embodiments, based on the images obtained by the step-by-step filtering method proposed in the present application, the present application proposes an image pre-filtering method, which includes:

[0131] The photovoltaic module includes a panel, the panel has a frame, and the panel includes a plurality of grid lines;

[0132] Fusion processing is performed on the images to obtain a fusion image; the images are the third images; wherein the steps include:

[0133] The corresponding edge pixel points in at least two images are aligned respectively;

[0134] The average values of the corresponding edge pixel points in at least two images are calculated;

[0135] The corresponding grid line pixel points in at least two images are aligned;

[0136] The average values of the corresponding grid line pixel points in at least two images are calculated;

[0137] The pixel points of the first feature points in at least two images are aligned;

[0138] The average values of the corresponding first feature point pixel points in at least two images are calculated;

[0139] The fusion image is obtained according to the average values of the edge pixel points, the average values of the grid line pixel points, and the average values of the first feature point pixel points.

[0140] Wherein, at least two third images are needed to be acquired for image fusion processing.

[0141] Wherein, the average value of the pixel points can be calculated according to the gray value or the brightness value of each pixel point in the set of third images.

[0142] In the embodiment, the images can be fused by aligning the pixel points, and then the average gray value or brightness value of each corresponding pixel point in the fused image can be calculated, which can solve the calculation error of the photoluminescence image caused by the offset of the shooting angle and improve the photoluminescence image processing efficiency.

[0143] In some embodiments, the present application can also optimize the feature points. In the case that the photovoltaic module defect detection amount is large and the calculation amount increases, such as when the first feature points are few or the non-defect factors are obvious, the first feature points can be directly aligned in the frame, and the grid lines are not aligned, which adaptively balances the conflict between the computing power and the calculation amount, so as to realize a substantial improvement in detection efficiency at the expense of part of the accuracy.

[0144] In some embodiments, the first feature points can be other feature points in addition to the edge feature points and the grid line feature points. The first feature points can be foreign matter shielding the image, or defects of the photovoltaic module itself. The defect feature points and the non-defect feature points in the first feature points are distinguished and identified in order to exclude the non-defect interference factors in the image pre-filtering step, so as to perform more efficient defect identification.

[0145] In some embodiments, the present application further comprises:

[0146] Selecting a target image and at least two contrast images from the images, and identifying an abnormal point through the contrast between the target image and the contrast images; wherein the steps include:

[0147] Selecting the target image and at least two contrast images from a plurality of images;

[0148] Calculating at least two pixel point difference values between a target pixel point of the target image and a plurality of corresponding pixel points of at least two contrast images; wherein the pixel points include the edge pixel points, the grid line pixel points, and the pixel points of the first feature points;

[0149] When the at least two pixel point difference values of the target pixel point in the contrast images are all greater than a preset pixel threshold value, the target pixel point is identified as an abnormal point.

[0150] It should be noted that in the embodiment, after a set of third images is obtained according to the first embodiment of the present application, an abnormality recognition step can be further performed on the third images. When a large difference is found between the target image and the comparison image, the abnormal point is preferably recognized to avoid interference with the photovoltaic module defect recognition process.

[0151] In some embodiments, the present application can preferably take at least three images of the photovoltaic module in two working states (open circuit state / short circuit state) respectively, take one of them as a target image of one working state, and take the remaining two as comparison images. The pixel point average value closest to the real situation can be calculated, and a reference for comparison and recognition can be determined, which does not require excessive algorithm resource consumption and can control the error within an acceptable range.

[0152] Among them, the pixel point difference value can be calculated according to the gray value or brightness value of each corresponding position pixel point in the target image and the comparison image.

[0153] For example, in some embodiments, at least three photos (i.e. images) can be taken as target photos in turn, and the target photos are compared with the remaining at least two comparison photos to identify abnormal points of the multiple target photos.

[0154] In some embodiments, further comprising:

[0155] Regarding one of the abnormal points as an abnormal region;

[0156] In some embodiments, preferably, multiple adjacent abnormal points are regarded as the same abnormal region, comprising the steps of:

[0157] Traversing the abnormal points to identify at least two abnormal points with a distance less than a preset distance threshold as an abnormal point set;

[0158] Regarding the region corresponding to the abnormal point set as the abnormal region;

[0159] Performing abnormality processing on the abnormal region.

[0160] In the embodiment, the regional abnormal point (i.e. abnormal region) is preferably locally and entirely eliminated. In other words, the present embodiment provides a restrictive abnormal region elimination method, which focuses on the abnormal processing of instantaneous sudden phenomena (such as birds and insects flying across the screen at the moment of shooting), to avoid over-elimination of abnormal points (such as causing real defects to be filtered in image processing).

[0161] To this end, the restrictive exception elimination method proposed in this embodiment can eliminate irrelevant interference factors to a certain extent in advance to reduce the recognition pressure of the subsequent AI model, and can also avoid the elimination of real defects due to the pre-elimination of interference factors.

[0162] Specifically, the present application identifies abnormal points with a distance less than a preset distance threshold as an abnormal point set, and then identifies the abnormal point set as an abnormal region. By using the progressive abnormal region identification method based on points and surfaces, discrete abnormal pixel points are upgraded to continuous abnormal regions, which can avoid the calculation pressure caused by isolated processing of abnormal points, and enhance the shape perception ability of transient targets such as flying birds and insect shadows.

[0163] In some embodiments, the abnormal region is subjected to abnormal processing, including:

[0164] determining whether the abnormal region is a dynamic abnormal region;

[0165] performing dynamic abnormal processing on the dynamic abnormal region.

[0166] In some embodiments, preferably, the present application identifies part of the abnormal region, especially the dynamic abnormal region, which can avoid misremoval of static abnormal points or static abnormal regions, and reduce the calculation pressure of the subsequent defect identification process. The present embodiment proposes an identification method for dynamic abnormal regions with significant motion trend, which can reduce the pressure of subsequent AI model identification of photovoltaic module defects through image preprocessing, and can avoid excessive pre-filtering through progressive restriction of the progressive dynamic abnormal region identification rule. It is not only a further verification of the accuracy of photoluminescence image acquisition, but also saves the computing power cost of photovoltaic module defect identification and improves the efficiency of the defect identification process.

[0167] Specifically, it is determined whether the abnormal region is a dynamic abnormal region, and if so, the dynamic abnormal region is subjected to dynamic abnormal processing,

[0168] In some embodiments, it includes:

[0169] determining whether the abnormal region is the dynamic abnormal region, which includes:

[0170] determining whether there are at least two target images that both have the abnormal region;

[0171] if so, calculating the shape similarity between at least two abnormal regions;

[0172] when the shape similarity between at least two abnormal regions is greater than a preset similarity threshold, it is considered that at least two abnormal regions are similar;

[0173] The at least two similar abnormal regions are identified as one dynamic abnormal region.

[0174] In some embodiments, the shape similarity can be characterized by area similarity or morphological similarity. The morphological similarity can be calculated by a contour matching algorithm. The calculation result of the shape similarity can be obtained by the prior art, which is not limited here.

[0175] It should be noted that the application further identifies the dynamic abnormal region, that is, further takes the dynamic abnormal region as a non-defect interference factor, so as to ensure that the image preprocessing process excludes interference factors caused by external reasons, rather than defects (aging, hidden cracks, hot spots, PID, etc.) of the photovoltaic module itself (i.e. static defects), thereby avoiding filtering out the defects that should be identified, which is beneficial to saving algorithm resources while constraining the abnormal identification process, balancing the contradiction between identification efficiency and accuracy.

[0176] In some embodiments, an abnormal feature region with a shape similarity greater than a preset similarity threshold but with a relative position change can be determined as a dynamic abnormal region and filtered out. In some embodiments, the relative position change needs to be greater than a first displacement threshold, so as to avoid the photovoltaic module shaking caused by wind being mistaken for an abnormal feature region with motion.

[0177] In some embodiments, the method further comprises:

[0178] Before identifying the at least two abnormal regions as one dynamic abnormal region, the method further comprises the step of:

[0179] obtaining a shooting time sequence between the multiple target images and obtaining a target position of each similar abnormal region in the multiple target images;

[0180] calculating a moving speed of the abnormal region according to the shooting time sequence and the target position;

[0181] When the moving speed is greater than a preset speed threshold, the corresponding abnormal region is identified as the dynamic abnormal region.

[0182] For example, the preset speed threshold can be adaptively set in combination with the average flight speed of flying objects such as birds.

[0183] It should be understood that in the present embodiment, before identifying the at least two abnormal regions as one dynamic abnormal region, the moving speed of the dynamic abnormal region can be calculated according to the shooting time sequence and the target position. Whether the abnormal region belongs to the dynamic abnormal region can be determined by whether the moving speed of the abnormal region is greater than the preset speed threshold.

[0184] The application can guarantee the accuracy of dynamic abnormal area identification by verifying the abnormal area based on shape similarity and movement speed (or position change speed). Specifically, the dual-dimension verification based on static and dynamic can avoid excluding static abnormal points caused by shooting angle and other problems, and can also avoid affecting the accuracy of defect identification.

[0185] In summary, for static abnormal areas and dynamic abnormal areas, the application provides a dual-dimension verification mechanism based on the shape and position of the abnormal area to further achieve accurate filtering of dynamic abnormal areas.

[0186] Further, the method for identifying dynamic abnormal areas in the embodiment can significantly reduce the operation difficulty and cost of outdoor photovoltaic module defect detection in cooperation with the secondary filtering rule based on wavelength and light intensity. Specifically, the step-by-step filtering step in the image acquisition process cooperates with the dynamic abnormal area identification step in the image preprocessing process, so that the photoluminescence image can be acquired by shooting a small number of photovoltaic module images under the condition of limited sample size, and the dynamic abnormal area in the photoluminescence image can be identified based on the limited sample size to exclude the interference of dynamic non-defect factors.

[0187] In some embodiments, the application also provides a grid-based segmented dynamic abnormal area removal rule for photoluminescence images, which can effectively limit the process of excluding dynamic abnormal areas and avoid over-exclusion, including the steps of:

[0188] In some embodiments, the application also provides a grid-based segmented dynamic abnormal area removal rule for photoluminescence images, which can effectively limit the process of excluding dynamic abnormal areas and avoid over-exclusion, including the steps of:

[0189] Dividing the image into at least two grids according to the grid line pixels;

[0190] Calculating the number of grids occupied by the dynamic abnormal area;

[0191] If the number exceeds the first preset number, and the overall area ratio of the dynamic abnormal area in the image is greater than the first area ratio, the image containing the dynamic abnormal area is considered as an invalid image.

[0192] In some embodiments, according to the characteristic that the grid lines of the photovoltaic module are regularly distributed, the photoluminescence image of the photovoltaic module can be divided into several grids with equal or approximate size. By gridding the photoluminescence image, non-defect factors can be located and processed more carefully, so that the segmented removal of dynamic abnormal areas proposed in the application can be better achieved, which can not only improve the accuracy of non-defect factor removal, but also improve the utilization rate of photoluminescence images.

[0193] In some embodiments, if the number of grid occupied by the dynamic abnormal area exceeds the first preset number, and the overall area proportion of the dynamic abnormal area in the photoluminescence image is greater than the first area proportion, it indicates that the dynamic abnormal area occupies most of the grid of the photoluminescence image of the photovoltaic module (for example, the photovoltaic module has a total of nine grids, the first preset number is six, and the dynamic abnormal area occupies seven grids), and the area occupied by the dynamic abnormal area is also greater than the first area proportion (for example, 90%). Through the double constraints of the number of grid occupied by the dynamic abnormal area and the overall area proportion, it can be approximately determined that the dynamic abnormal area has seriously affected the presentation effect of the photoluminescence image. If such a photoluminescence image is input into the subsequent defect recognition process, it will lead to a decrease in the accuracy of defect recognition. The present application regards such a photoluminescence image as an invalid image and deletes the whole image, which can avoid the situation that the average value calculation result deviates too much from the actual situation caused by excessive abnormal values, and also avoid the distortion of the photovoltaic module defect analysis result caused by the fact that the non-defect factors shield the defect factors in the image.

[0194] In some embodiments, comprising:

[0195] If the number does not exceed the first preset number,

[0196] the grid occupied by the dynamic abnormal area is identified as an abnormal grid;

[0197] the abnormal grid is replaced with the corresponding grid in the same group of images that is not occupied by the dynamic abnormal area.

[0198] In some embodiments, if the number does not exceed the first preset number (for example, the photovoltaic module has a total of nine grids, the first preset number is six, and the dynamic abnormal area only occupies four grids), it indicates that the dynamic abnormal area occupies a small part of the photovoltaic module image. The abnormal grid is replaced with the corresponding grid in the same group of photovoltaic module images that does not contain the dynamic abnormal area. Through local replacement of the grid instead of overall replacement of the image, it is beneficial to maximize the utilization rate of the photoluminescence image.

[0199] In some embodiments, comprising:

[0200] If the number does not exceed the second preset number; wherein the second preset number is less than the first preset number;

[0201] and the grid area proportion of the dynamic abnormal area in the grid is less than the second area proportion;

[0202] the grid area occupied by the dynamic abnormal area is segmented to obtain an abnormal grid sub-area;

[0203] Correspondingly, the corresponding grid occupied by the dynamic abnormal area in the same group of photoluminescence images is segmented to obtain a replacement grid sub-area;

[0204] The replacement grid sub-area replaces the abnormal grid sub-area.

[0205] In some embodiments, if the number does not exceed a second preset number (for example, the photovoltaic module has a total of nine grids, and the second preset number is three, and the dynamic abnormal area occupies one), and the grid area ratio of the dynamic abnormal area in the grid is less than a second area ratio (for example, 50%), at this time, an abnormal grid sub-area is segmented in the grid, and a replacement grid sub-area corresponding in position and size is segmented in the same group of photoluminescence images, and the replacement grid sub-area replaces the abnormal grid sub-area, which can minimize the impact of the preprocessing step of removing non-defect factors on the photoluminescence image, thereby preventing the over-exclusion of weak defects (such as some initial defects with very fine cracks, which may produce less obvious lines on the image) and reducing the accuracy of defect recognition.

[0206] In some embodiments, after the photoluminescence image is pre-filtered, the method further comprises the steps of: a) transmitting the pre-filtered photoluminescence image to a cloud server through a wireless network; b) starting an AI autonomous recognition algorithm for photovoltaic module defect detection on the cloud server; c) identifying the defect type of the photovoltaic module and adding a label at the corresponding position of the defect; d) outputting the risk level of the module according to the defect type and number of the photovoltaic module to the terminal display.

[0207] In some embodiments, after step S103, the method further comprises:

[0208] Inputting the fused image into an AI autonomous recognition model for photovoltaic module defects;

[0209] Identifying the defect type of the photovoltaic module.

[0210] In some embodiments, the method further comprises:

[0211] Generating a defect image of the photovoltaic module according to the defect type.

[0212] In some embodiments, the method further comprises:

[0213] Labeling a corresponding preset defect type label at the corresponding position in the defect image.

[0214] Deep learning has made outstanding achievements in the field of computer vision image recognition, especially convolutional neural networks, which have a special network structure that is very robust to complex backgrounds, uneven lighting, angle changes, and other two-dimensional images. In recent years, it has been widely used in classification, detection, segmentation, and other tasks in computer vision.

[0215] This embodiment uses a CLIP-embedded YOLO algorithm for photovoltaic module defect detection, thereby enhancing the contextual understanding of visual input, enabling the model to achieve better performance even with limited data sample size, and improving the adaptability and accuracy of recognition.

[0216] In some embodiments, the implementation steps can be:

[0217] 1.1 Dataset Preprocessing

[0218] Selecting representative images requires the use of K-medoids clustering to optimize diversity within each category. The structure is as shown in Figure 4 . First, unique features are extracted from each image, generating a feature vector for each image. Then, K-medoids clustering is applied using Euclidean distance to divide the dataset into K clusters. The representative image of each cluster is selected as the medoid, i.e., the image with the minimum total distance to all other images in the cluster. This technique ensures that the selected images cover a wide range of variations within each category, thereby improving the generalization ability of the deep learning model and enabling it to more accurately identify photovoltaic defects.

[0219] 1.2 Combining CLIP with YOLOv11 to enhance photovoltaic defect detection

[0220] The method of integrating CLIP embedding with YOLOv11 leverages the multi-modal capabilities of CLIP to improve the initialization and training process of YOLO. The embeddings generated by CLIP for the representative dataset provided are used to initialize the layers of YOLOv11, enabling the model to have comprehensive contextual understanding capabilities and faster convergence from the start. The model implements target detection in two stages:

[0221] Stage 1: Extracting embedding vectors using CLIP

[0222] The goal of this stage is to generate a knowledge vector for each defect category that contains rich semantic information. The input is pre-processed defect images and corresponding labels (such as cracks). As shown in Figure 5 , the input image is processed by the image encoder of CLIP, converting it into a picture vector representing the visual features of the picture. The label is processed by the text encoder of CLIP, converting the label text into another text vector representing the semantic features of the label. Next, the picture vector and the text vector are fused, resulting in a higher-dimensional embedding vector that contains both picture and semantic information.

[0223] Stage 2: Initialize the convolutional neural network with the embedding vectors, and train the model on the dataset.

[0224] The embedding vector e obtained in stage 1 is converted and used as the initial weights of the first layer of the convolutional kernel of the neural network. Then the model is trained on the photovoltaic defect dataset, and the performance is evaluated using standard target detection indicators such as precision, recall, mAP and robustness.

[0225] The photovoltaic panel defect recognition algorithm based on deep neural network described above is used in this embodiment, which can effectively operate in the case of limited training samples through the integrated multi-modal understanding ability, showing higher robustness, and improving the reliability and effectiveness of the model in identifying photovoltaic component defects in practical applications.

[0226] To solve the problem of limited data samples, this embodiment introduces a multi-modal model, combines CLIP embedding and YOLO model, and provides a photovoltaic panel defect detection algorithm with robustness and effectiveness. Verification is carried out on the PVEL-AD dataset, which is a dataset for benchmark testing of photovoltaic component anomaly defect detection methods. It contains 36,543 near-infrared images with various internal defects and heterogeneous backgrounds, including 1 type of non-anomalous image and 12 types of anomalous defect images, such as cracks (linear and star-shaped), broken grids, black cores, misalignment, thick lines, scratches, fragments, broken corners, and material defects. In addition, the dataset provides more than 40,000 real annotation boxes for 12 types of anomalous defects for defect detection. In the experimental verification, the model uses 30 images of each category for training. The integration of CLIP embedding provides a significant advantage in learning from a small number of samples, enhancing the generalization ability and detection accuracy of the model. As shown in Figure 2 、 Figure 3 , Figure 2 、 Figure 3 The comparison shows the actual defect image and the predicted image, which shows that the accuracy of photoluminescence image defect recognition can be effectively improved by the present application.

[0227] In some embodiments, the photovoltaic component defect AI autonomous recognition model identifies the defect type of the photovoltaic component, which can include the following steps:

[0228] (1) Extract image features and generate feature vectors for each image;

[0229] (2) Perform K-medoids clustering to group similar defect images into the same cluster;

[0230] (3) Select representative images in the same cluster, including:

[0231] (3a) Set the number of clusters K according to the defect complexity;

[0232] (3b) Calculate the feature vector similarity of the image, and select a representative image based on the feature vector similarity to minimize the sum of the Euclidean distances from the representative image to all images in the cluster;

[0233] (4) Extract the embedding vector using CLIP, including:

[0234] (4a) Convert the image into an image visual feature vector. ;

[0235] (4b) Convert the label text into another text vector. ;

[0236] (4c) Based on visual feature vectors and text vector Obtain the embedding vector of the dataset ;

[0237] (5) Using embedding vectors Perform model training;

[0238] (6) Validate the dataset.

[0239] Example 3:

[0240] The present invention also provides a photoluminescence image processing system, comprising:

[0241] A photovoltaic module includes a panel with a frame and a plurality of grid lines within the panel; an image fusion module is used to fuse images to obtain a fused image; wherein the system includes: an edge pixel alignment unit for aligning corresponding edge pixels in at least two images respectively; an edge pixel calculation unit for calculating the average value of corresponding edge pixels in at least two images; a grid line pixel alignment unit for aligning corresponding grid line pixels in at least two images; a grid line pixel calculation unit for calculating the average value of corresponding grid line pixels in at least two images; a first feature point alignment unit for aligning pixels of a first feature point in at least two images; a first feature point calculation unit for calculating the average value of corresponding first feature point pixels in at least two images; and an image fusion unit for obtaining the fused image based on the average value of edge pixels, the average value of grid line pixels, and the average value of the first feature point pixels.

[0242] In some embodiments, the image contrast module is configured to select a target image and at least two contrast images from the images, and identify an abnormal point through contrast between the target image and the contrast images. The image selection unit is configured to select the target image and at least two contrast images from a plurality of images. The pixel difference calculation unit is configured to calculate at least two pixel point differences between a target pixel point of the target image and corresponding pixel points of at least two contrast images. The pixel points include the edge pixel point, the grid line pixel point, and the pixel point of the first feature point. The abnormal point identification unit is configured to identify the target pixel point as an abnormal point when at least two pixel point differences corresponding to the target pixel point are all greater than a preset pixel threshold.

[0243] In some embodiments, the abnormal region identification unit is configured to regard one abnormal point as an abnormal region, or regard a plurality of adjacent abnormal points as the same abnormal region. The abnormal point set identification unit is configured to traverse the abnormal points to identify at least two abnormal points with a distance less than a preset distance threshold as an abnormal point set. The abnormal region identification subunit is configured to identify a region corresponding to the abnormal point set as the abnormal region. The abnormal processing subunit is configured to perform abnormal processing on the abnormal region.

[0244] It should be noted that, in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that processes, methods, articles, or devices that include a series of elements not only include those elements, but also include other elements not explicitly listed, or inherent to such processes, methods, articles, or devices. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article, or device that includes the element.

[0245] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for making a computer terminal (which can be a mobile phone, computer, server, or network device, etc.) execute the methods described in various embodiments of the present application.

[0246] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, but not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, and these all belong to the protection of the present application.

Claims

1. A photoluminescent image processing method, characterized by, Comprising: The photovoltaic assembly comprises a panel, the panel has a frame, and the panel comprises a plurality of grid lines; The image fusion processing is performed to obtain a fusion image, wherein the processing includes the following steps: aligning corresponding edge pixel points in at least two images respectively; calculating average values of the corresponding edge pixel points in the at least two images; aligning corresponding grid line pixel points in the at least two images; calculating average values of the corresponding grid line pixel points in the at least two images; aligning pixel points of first feature points in the at least two images; calculating average values of the corresponding first feature point pixel points in the at least two images; and obtaining the fusion image according to the average values of the edge pixel points, the average values of the grid line pixel points, and the average values of the first feature point pixel points. The method further comprises: Selecting a target image and at least two comparison images from the images, and identifying an abnormal point through comparison between the target image and the comparison images, wherein the method comprises the following steps: selecting the target image and at least two comparison images from a plurality of images; calculating at least two pixel difference values of a plurality of corresponding pixel points between a target pixel point of the target image and the at least two comparison images; wherein the target pixel point includes the edge pixel point, the grid line pixel point, and the pixel point of the first feature point; and when the at least two pixel difference values corresponding to the target pixel point are all greater than a preset pixel threshold, identifying the target pixel point as an abnormal point. In the method, one of the images is taken as a target image of a working state, and the other images are taken as comparison images. One abnormal point is taken as an abnormal region, or a plurality of adjacent abnormal points are taken as the same abnormal region. The abnormal region is subjected to abnormal processing, which comprises: Judging whether the abnormal region is a dynamic abnormal region, wherein the judgment comprises: Judging whether the abnormal region exists in at least two target images; If yes, calculating shape similarity between at least two abnormal regions; When the shape similarity between the at least two abnormal regions is greater than a preset similarity threshold, considering that the at least two abnormal regions are similar; Identifying the similar at least two abnormal regions as one dynamic abnormal region. Filtering out the dynamic abnormal region.

2. A photoluminescence image processing method according to claim 1, characterized in that, The step of taking a plurality of adjacent abnormal points as the same abnormal region comprises: Traversing the abnormal points to identify at least two abnormal points with a distance less than a preset distance threshold as an abnormal point set; and Identifying a region corresponding to the abnormal point set as the abnormal region.

3. A photoluminescence image processing method according to claim 1, wherein, The method further comprises: Before identifying at least two abnormal regions as one dynamic abnormal region, the method further comprises the following steps: Obtaining a shooting time sequence between a plurality of target images, and obtaining target positions of similar abnormal regions in the plurality of target images; Calculating a moving speed of the abnormal region according to the shooting time sequence and the target positions; When the moving speed is greater than a preset speed threshold, identifying the corresponding abnormal region as the dynamic abnormal region.

4. The photoluminescence image processing method of claim 1, wherein, The method further comprises: inputting the fusion image into a photovoltaic module defect AI autonomous identification model; identifying a defect type of the photovoltaic module; generating a defect image of the photovoltaic module according to the defect type; labeling a corresponding preset defect type label in a corresponding position in the defect image.

5. A photoluminescent image processing system, characterized by comprise: The photovoltaic module comprises a panel, the panel has a frame, and the panel comprises a plurality of grid lines; An image fusion module is used for fusion processing of images to obtain a fusion image; wherein the system comprises: An edge pixel point alignment unit is used for aligning corresponding edge pixel points in at least two images respectively; An edge pixel point calculation unit is used for calculating average values of corresponding edge pixel points in at least two images; A grid line pixel point alignment unit is used for aligning corresponding grid line pixel points in at least two images; A grid line pixel point calculation unit is used for calculating average values of corresponding grid line pixel points in at least two images; A first feature point alignment unit is used for aligning pixel points of first feature points in at least two images; A first feature point calculation unit is used for calculating average values of corresponding first feature point pixel points in at least two images; An image fusion unit is used for obtaining the fusion image according to the average values of the edge pixel points, the average values of the grid line pixel points, and the average values of the first feature point pixel points; An image comparison module is used for selecting a target image and at least two comparison images from the images, and identifying an abnormal point through comparison of the target image and the comparison images; wherein the image comparison module is further used for: selecting the target image and at least two comparison images from a plurality of images; calculating at least two pixel point difference values of a plurality of corresponding pixel points between a target pixel point of the target image and at least two comparison images; wherein the target pixel point comprises: the edge pixel point, the grid line pixel point, and the pixel point of the first feature point; when at least two pixel point difference values corresponding to the target pixel point are all greater than a preset pixel threshold value, the target pixel point is identified as an abnormal point; wherein one of the images is taken as a target image of one working state, and the other images are taken as comparison images; An abnormal area identification unit is used for regarding one abnormal point as one abnormal area, or regarding a plurality of adjacent abnormal points as one abnormal area; The system is further used for: performing abnormal processing on the abnormal area, which comprises: judging whether the abnormal area is a dynamic abnormal area; wherein it comprises: judging whether there are at least two target images that all have the abnormal area; if yes, calculating shape similarity between at least two abnormal areas; when the shape similarity between at least two abnormal areas is greater than a preset similarity threshold value, it is considered that at least two abnormal areas are similar; then, identifying at least two similar abnormal areas as one dynamic abnormal area; filtering out the dynamic abnormal area.

6. A photoluminescence image processing system according to claim 5, wherein, comprise: An image selection unit is used for selecting the target image and at least two comparison images from a plurality of images; A pixel difference calculation unit is configured to calculate at least two pixel difference values between a target pixel point of a target image and a plurality of corresponding pixel points of at least two comparison images, wherein the pixel points include the edge pixel points, the grid line pixel points, and pixel points of the first feature points. An abnormal point identification unit is configured to identify the target pixel point as an abnormal point when the at least two pixel difference values corresponding to the target pixel point are all greater than a preset pixel threshold.

7. A photoluminescence image processing system according to claim 5, wherein, Further comprising: An abnormal point set identification unit is configured to traverse the abnormal points to identify at least two abnormal points as an abnormal point set when a distance between the at least two abnormal points is less than a preset distance threshold. An abnormal region identification subunit is configured to identify a region corresponding to the abnormal point set as the abnormal region.

Images