Photovoltaic panel hot spot detection optimization method based on double-model dynamic architecture

Abstract

The invention relates to the technical field of hot spot detection, in particular to a photovoltaic panel hot spot detection optimization method based on a double-model dynamic architecture. The method comprises the following steps: screening out a target visible light resolution according to hot spot confidence coefficients of hot spot matching regions in different super-resolution images corresponding to all hot spot regions in a visible light image, the number of the hot spot matching regions and the number of the hot spot regions in each super-resolution image; taking the target visible light resolution as the initial infrared resolution, and judging whether the initial infrared resolution needs to be adjusted or not according to gradient distribution, temperature distribution and corresponding hot spot confidence of the photovoltaic panel at different positions in the infrared image between the initial infrared resolution and the next super-resolution; if the adjustment is needed, taking the next super-resolution as a new initial infrared resolution until the adjustment is not needed, and obtaining a target infrared resolution; and the hot spot detection of the photovoltaic panel is realized. According to the invention, the hot spot detection efficiency is improved by obtaining the accurate detection resolution.

Description

Technical Field

[0001] This invention relates to the field of hot spot detection technology, and specifically to an optimized method for hot spot detection of photovoltaic panels based on a dual-model dynamic architecture. Background Technology

[0002] Hot spots occur when a photovoltaic (PV) panel experiences localized shading or performance differences, causing the affected area to be reverse-charged by other normally functioning solar cells instead of acting as a power generator. This generates high temperatures. Prolonged hot spots can burn out the affected solar cells, ultimately rendering the entire PV panel unusable. Therefore, timely management of hot spots in PV panels is crucial to ensure the power generation efficiency of the PV system and minimize economic losses.

[0003] In existing technologies, photovoltaic panel hot spots are detected by regularly flying drones and using target detection models and infrared hot spot detection for direct upsampling. However, considering that upsampling is prone to loss of detail, and that the resolution of the image during drone flight directly affects the drone's flight power consumption and the model's processing time, a higher resolution requires a model with strong computing power and requires more power consumption, while a lower resolution may result in poor image quality and reduced detection accuracy. Therefore, an unsuitable image resolution is less effective for detecting photovoltaic panel hot spots. Summary of the Invention

[0004] To address the technical problem of poor performance in photovoltaic panel hot spot detection due to unsuitable image resolution, the present invention aims to provide an optimized method for photovoltaic panel hot spot detection based on a dual-model dynamic architecture. The specific technical solution adopted is as follows: This invention proposes an optimization method for photovoltaic panel hot spot detection based on a dual-model dynamic architecture, the method comprising: Acquire visible light and infrared images of a photovoltaic panel containing a hot spot region during dynamic operation, wherein the hot spot region contains a hot spot confidence label; Multiple super-resolution images of the visible light image in the initial dynamic architecture are obtained. Based on the positional distribution of hot spot regions between the visible light image and different super-resolution images, the hot spot matching region of each hot spot region in the visible light image is obtained in different super-resolution images. The target visible light resolution is selected based on the hot spot confidence, the number of hot spot matching regions, and the number of hot spot regions in each super-resolution image corresponding to all hot spot regions in the visible light image. Using the target visible light resolution as the initial infrared resolution, the gradient distribution, temperature distribution, and corresponding hot spot confidence of the photovoltaic panel at different locations in the infrared image between the initial infrared resolution and the next super-resolution determine whether the initial infrared resolution needs to be adjusted. If adjustment is required, the next super-resolution is used as the new initial infrared resolution until no adjustment is needed, thus obtaining the target infrared resolution. Based on the target infrared resolution and the target visible light resolution, the hot spot detection of the photovoltaic panel is achieved.

[0005] Furthermore, the method for obtaining the hot spot matching region includes: The center positions of hot spot regions in different adjacent super-resolution images are matched, and the corresponding matched hot spot regions are used as hot spot matching regions to obtain the hot spot matching regions of each hot spot region in the visible light image in different super-resolution images.

[0006] Furthermore, the method for obtaining the target visible light resolution includes: Based on the hot spot confidence of each hot spot region in the visible light image corresponding to the hot spot matching region in different super-resolution images, and the number of hot spot matching regions, the local information loss degree of each hot spot region in each super-resolution image is obtained. The overall information loss of each super-resolution image is obtained based on the local information loss of all hot spot regions in each super-resolution image and the difference in the number of hot spot regions in each super-resolution image and the visible light image. The super-resolution image with the smallest overall information loss is selected, and the resolution of the corresponding super-resolution image is taken as the target visible light resolution.

[0007] Furthermore, the method for obtaining the local information loss degree includes: For any hot spot region in a visible light image, the hot spot confidence difference of each hot spot region in the hot spot matching region between each super-resolution image and the next super-resolution image is obtained as the first confidence difference. Obtain the number of hotspot matching regions for each hotspot region in different super-resolution images and the comparison difference between the order of each super-resolution image; calculate the sum of the positive integer 1 and the comparison difference as the first sum; calculate the ratio of the first sum to the number of hotspot matching regions as the information loss weight. The product of the first confidence difference and the information loss weight is obtained and normalized to form the local information loss of each hotspot region in each super-resolution image.

[0008] Furthermore, the method for obtaining the overall information loss degree includes: The sum of the differences between the positive integer 1 and the number of hot spot regions is obtained, and the sum is divided by the number of hot spot regions as the loss coefficient. The cumulative value of the local information loss of all hot spot regions in each super-resolution image is obtained as the cumulative loss value; The product of the accumulated loss and the loss coefficients is calculated as the overall information loss for each super-resolution image.

[0009] Furthermore, the determination of whether the initial infrared resolution needs to be adjusted includes: Based on the gradient distribution, temperature distribution, and corresponding hot spot confidence at different locations in an infrared image at any resolution, the feature saliency at the corresponding resolution is obtained. The reference value of the initial infrared resolution is obtained based on the saliency of features between the initial infrared resolution and the next resolution, the gradient distribution of the infrared image, and the corresponding hotspot confidence. Based on the reference value of the initial reference resolution, determine whether the initial infrared resolution needs to be adjusted.

[0010] Furthermore, the method for obtaining the saliency of the features includes: For hot spot or non-hot spot regions in an infrared image at any resolution, the mean value of the gradient magnitude of all pixels in each region is obtained as the sharpness of each region. The difference in the average temperature of all pixels between hot spot and non-hot spot regions in an infrared image is obtained as the temperature difference. The ratio of the sharpness difference between the hot spot region and the non-hot spot region to the temperature difference is obtained. The product of the ratio result and the sharpness of the hot spot region is calculated as the adjustment necessity at the corresponding resolution. The ratio of the hot spot confidence and the adjustment necessity of the corresponding infrared image is obtained and normalized and mapped as the feature saliency at the corresponding frequency.

[0011] Furthermore, the method for obtaining the reference value includes: Obtain the sharpness ratio of the non-defect region between the initial infrared resolution and the next super-resolution, as an adjustment factor; obtain the product of the adjustment factor and the feature significance of the super-resolution, and calculate the difference between the feature significance of the initial infrared resolution and the product result, as the significant change between the corresponding resolutions; The confidence level of the infrared image at the initial infrared resolution is obtained as a percentage of the significant change between the corresponding resolutions, and then normalized and mapped to serve as a reference value for the initial infrared resolution.

[0012] Furthermore, the method for determining whether the initial infrared resolution needs adjustment based on the reference value of the initial reference resolution includes: If the reference value of the initial reference resolution is less than or equal to the preset reference threshold, it is determined that the corresponding initial reference resolution needs to be adjusted.

[0013] Furthermore, the method for obtaining the target infrared resolution includes: Once it is determined that adjustment is needed, the next super-resolution is used as the new initial infrared resolution. The reference value of the super-resolution is used for further judgment until no adjustment is needed, at which point the new initial infrared resolution is used as the target infrared resolution.

[0014] The present invention has the following beneficial effects: This invention obtains the hot spot matching region for each hot spot region in the visible light image across different super-resolution images based on the positional distribution of hot spot regions between visible light images and different super-resolution images, ensuring that the same physical hot spot is analyzed at different resolutions. Based on the hot spot confidence score, the number of hot spot matching regions corresponding to all hot spot regions in the visible light image in different super-resolution images, and the number of hot spot regions in each super-resolution image, a target visible light resolution is selected. This selection of resolutions beneficial for hot spot detection and analysis helps to more accurately analyze photovoltaic panel hot spots. Using the target visible light resolution as the initial infrared resolution, the gradient distribution, temperature distribution, and corresponding hot spot confidence score of the photovoltaic panel at different positions in the infrared images between the initial infrared resolution and the next super-resolution image determine whether the initial infrared resolution needs adjustment. Based on the unique characteristics of infrared images that are essential to hot spots, a balance between accuracy and efficiency is efficiently found. If adjustment is needed, the next super-resolution image is used as the new initial infrared resolution until no further adjustment is required, thus obtaining the target infrared resolution. This enables the detection of photovoltaic panel hot spots. This invention improves the efficiency of hot spot detection by obtaining accurate detection resolution. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an optimization method for photovoltaic panel hot spot detection based on a dual-model dynamic architecture, provided in one embodiment of the present invention; Figure 2 A flowchart illustrating a method for obtaining local information loss degree according to an embodiment of the present invention; Figure 3 This is a flowchart of a method for determining whether initial infrared resolution needs to be adjusted, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of an optimization method for photovoltaic panel hot spot detection based on a dual-model dynamic architecture, according to an embodiment of the present invention. The specific method includes: Step S1: Obtain visible light and infrared images of the photovoltaic panel containing hot spot regions during dynamic operation. The hot spot regions contain hot spot confidence labels.

[0021] In embodiments of the present invention, to reduce energy consumption during UAV flight and improve the handling of reflection and hot spot problems, a dual-model dynamic architecture is adopted. The dual models are Binomial-PV1, whose basic framework is YOLOX-Nano, and which belongs to the photovoltaic feature enhancement layer PV-LEA in the model to solve the problem of false detection of reflection in photovoltaic images; and Binomial-ThermoScannerV1 model, whose basic framework is NanoDet, and which belongs to the temperature gradient attention TGAM in the model to solve the problem of low recognition rate of micro hot spots during detection. The acquired corresponding images are input into the model to obtain information such as the location of hot spots and hot spot confidence in the images. Visible light images and infrared images of the photovoltaic panel containing hot spot regions during dynamic operation are acquired, and the hot spot regions contain hot spot confidence labels.

[0022] Hot spot confidence can measure the probability of hot spot defects in an image. The higher the confidence, the more likely there are hot spots affecting the operation of the photovoltaic panel. Visible light images are input into PV-LEA, and infrared images are input into TGAM to obtain the corresponding hot spot confidence. The hot spot confidence of each hot spot region is marked to obtain a hot spot confidence label. The specific methods are well known to those skilled in the art and will not be described in detail here.

[0023] Step S2: Obtain multiple super-resolution images of the visible light image in the initial dynamic architecture. Based on the positional distribution of hot spot regions between the visible light image and different super-resolution images, obtain the hot spot matching region of each hot spot region in the visible light image in different super-resolution images.

[0024] The choice of image resolution affects the performance of the dual model. High resolution leads to information redundancy and increased computational power consumption, while low resolution may result in poor image quality and decreased detection accuracy. Therefore, it is necessary to adjust the image resolution to better identify photovoltaic hot spots while improving recognition efficiency. Multiple super-resolution images of the visible light image in the initial dynamic architecture are obtained.

[0025] It should be noted that, in one embodiment of the present invention, a super-resolution algorithm is used to obtain multiple super-resolution images of the visible light image. That is, the visible light image in the initial dynamic architecture is processed using a super-resolution scaling factor to obtain multiple super-resolution images. The scaling factor can be set according to specific circumstances. The specific means are well known to those skilled in the art and will not be described in detail here.

[0026] The center position of the hot spot region can determine the precise location of the hot spot region. By analyzing the distribution of the center positions of the hot spot regions, the performance of the same hot spot at different resolutions can be correlated. Based on the positional distribution of the hot spot regions between the visible light image and different super-resolution images, the hot spot matching region of each hot spot region in the visible light image in different super-resolution images can be obtained.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining the hot spot matching region includes: The center positions of hot spot regions in different adjacent super-resolution images are matched, and the corresponding matched hot spot regions are used as hot spot matching regions to obtain the hot spot matching regions of each hot spot region in the visible light image in different super-resolution images.

[0028] It should be noted that the SIFT algorithm is used to match the center positions of hot spot regions in different super-resolution images. If the center positions of hot spot regions match in the previous super-resolution image and the next super-resolution image, the corresponding hot spot region is taken as the hot spot matching region. Therefore, the hot spot matching region of each hot spot region in the visible light image in different super-resolution images can be obtained. The specific means are well known to those skilled in the art and will not be described in detail here.

[0029] Step S3: Based on the hot spot confidence, the number of hot spot matching regions, and the number of hot spot regions in each super-resolution image corresponding to all hot spot regions in the visible light image, the overall information loss of each super-resolution image is obtained, and the target visible light resolution is selected.

[0030] Hotspot confidence can assess the probability of hotspots existing in a hotspot region. Selecting a super-resolution image, if the difference in hotspot confidence between that image and the next super-resolution image is small, it indicates that changing the resolution does not significantly affect hotspot identification on the photovoltaic panel, resulting in less information loss. A larger number of hotspot matching regions indicates that hotspot defects can be identified at different resolutions. The closer the current image resolution is to a larger image resolution, the more information the image itself contains, and the less information loss. A smaller number of hotspot regions indicates that the image is less likely to contain hotspots, resulting in relatively less information loss. Based on the hotspot confidence, the number of hotspot matching regions corresponding to all hotspot regions in the visible light image in different super-resolution images, and the number of hotspot regions in each super-resolution image, the target visible light resolution is selected.

[0031] Preferably, in one embodiment of the present invention, the method for obtaining the target visible light resolution includes: Based on the hot spot confidence of each hot spot region in the visible light image corresponding to the hot spot matching region in different super-resolution images, and the number of hot spot matching regions, the local information loss degree of each hot spot region in each super-resolution image is obtained. Preferably, in one embodiment of the present invention, the method for obtaining the local information loss degree is described in [reference needed]. Figure 2 It illustrates a flowchart of a method for obtaining local information loss, including: Step S201: For any hot spot region in the visible light image, obtain the hot spot confidence difference of each hot spot region in the hot spot matching region between each super-resolution image and the next super-resolution image, and use it as the first confidence difference.

[0032] Hotspot confidence can assess the hotspot probability in a hotspot region. The closer the hotspot confidence between each super-resolution image and the next super-resolution image, the more similar the information features they can represent.

[0033] Step S202: Obtain the number of hot spot matching regions for each hot spot region in different super-resolution images and the comparison difference of the order of each super-resolution image; calculate the sum of the positive integer 1 and the comparison difference as the first sum value.

[0034] The ratio of the first sum to the number of hotspot matching regions is calculated and used as the information loss weight.

[0035] It should be noted that the higher the image resolution, the more information it contains, and the closer it is to the position of the subsequent super-resolution image, the greater its impact on the information. In order to analyze the impact of the relative position of the image on the information and avoid the comparison difference being 0, the sum of the positive integer 1 and the comparison difference is calculated as the first sum value. The larger the comparison difference, the smaller the super-resolution image order is relative to the number of hot spot matching regions, the less information the super-resolution image itself contains, and the greater the information loss weight.

[0036] Step S203: Obtain the product of the first confidence difference and the information loss weight, and perform normalization mapping as the local information loss degree of each hotspot region in each super-resolution image.

[0037] It should be noted that in some embodiments of the present invention, normalization can be performed by linear normalization or a normalization function. The specific means are well known to those skilled in the art and will not be described in detail here.

[0038] The overall information loss of each super-resolution image is obtained based on the local information loss of all hot spot regions in each super-resolution image and the difference between the number of hot spot regions in the visible light image and each super-resolution image. Preferably, in one embodiment of the present invention, the method for obtaining the overall information loss includes: The sum of the differences between the positive integer 1 and the number of hotspot regions is obtained, and the sum is divided by the number of hotspot regions as the loss coefficient; the cumulative value of the local information loss of all hotspot regions in each super-resolution image is obtained as the cumulative loss value. The product of the accumulated loss and the loss coefficients is calculated as the overall information loss for each super-resolution image.

[0039] It should be noted that the greater the local information loss, the greater the impact on the overall information loss. The smaller the number of hotspot regions when the resolution increases compared to the number of hotspot regions in the initial visible light image, the greater the difference in the number of hotspot regions, the more hotspot region details are lost, and the greater the overall information loss at that resolution. Therefore, based on the above basic mathematical operations, a correlation is constructed between the local information loss, the difference in the number of hotspot regions, and the overall information loss. That is, the greater the local information loss and the greater the difference in the number of hotspot regions, the greater the overall information loss.

[0040] The overall information loss reflects the accuracy of image analysis at the given resolution. The smaller the overall information loss, the more details are displayed, and the more accurate the judgment of the defect location. The target visible light resolution is selected by choosing the super-resolution image with the smallest overall information loss value and using the resolution of the corresponding super-resolution image as the target visible light resolution.

[0041] Step S4: Using the target visible light resolution as the initial infrared resolution, determine whether the initial infrared resolution needs to be adjusted based on the gradient distribution, temperature distribution, and corresponding hot spot confidence of the photovoltaic panel at different positions in the infrared image between the initial infrared resolution and the next super-resolution. If adjustment is required, use the next super-resolution as the new initial infrared resolution until no adjustment is needed to obtain the target infrared resolution.

[0042] To improve the accuracy of hot spot identification on photovoltaic panels, infrared images need to be analyzed to provide temperature information for more precise judgment. Gradient distribution reflects the texture complexity and edge strength of the image; therefore, the greater the gradient between the hot spot area and the non-hot spot area, the greater the texture complexity, edge strength, and sharpness, and the more obvious the hot spot features. Due to the presence of hot spots, there is a significant temperature difference between the hot spot area and the non-hot spot area of ​​the photovoltaic panel; the greater the temperature difference, the greater the probability of hot spots. Therefore, the target visible light resolution is used as the initial infrared resolution. Based on the gradient distribution, temperature distribution, and corresponding hot spot confidence of different positions in the infrared image of the photovoltaic panel between the initial infrared resolution and the next super-resolution, it is determined whether the initial infrared resolution needs to be adjusted.

[0043] Preferably, in one embodiment of the present invention, the determination of whether the initial infrared resolution needs to be adjusted is described in detail below. Figure 3 It illustrates a flowchart of a method for determining whether an initial infrared resolution adjustment is needed, including: Step S301: Based on the gradient distribution, temperature distribution, and corresponding hot spot confidence at different locations in the infrared image at any resolution, obtain the feature saliency at the corresponding resolution.

[0044] Preferably, in one embodiment of the present invention, the method for obtaining feature saliency includes: For any hot spot or non-hot spot region in an infrared image at any resolution, the average gradient magnitude of all pixels in each region is obtained as the sharpness of each region; the difference in the average temperature of all pixels between the hot spot region and the non-hot spot region in the infrared image is obtained as the temperature difference. The ratio of the sharpness difference between the hot spot region and the non-hot spot region to the temperature difference is obtained. The product of the ratio result and the sharpness of the hot spot region is calculated as the adjustment necessity at the corresponding resolution. The ratio of hotspot confidence to adjustment necessity in the corresponding infrared image is obtained and normalized to serve as the feature saliency at the corresponding frequency.

[0045] It should be noted that the gradient magnitude can be obtained using existing Sobel or Prewitt operators. The specific methods are well known to those skilled in the art and will not be elaborated here.

[0046] Based on this, the greater the image sharpness, the smaller the sharpness difference between the next super-resolution and the next higher resolution, the greater and more consistent the gradient performance, the clearer the image at the current resolution, and the less need for adjustment. In infrared images, temperature represents the thermal state of the object's surface. The greater the temperature difference, the more obvious the difference in thermal characteristics between hot spot areas and non-hot spot areas, and the less need for adjustment. The higher the confidence level of the hot spot, the greater the probability of assessing the hot spot, the greater the feature significance, and the more accurately the hot spot defects of the photovoltaic panel can be identified at the corresponding resolution.

[0047] Step S302: Based on the saliency of features between the initial infrared resolution and the next super-resolution, the gradient distribution of the infrared image, and the corresponding hotspot confidence, obtain the reference value of the initial infrared resolution.

[0048] Preferably, as the image resolution increases, the gradient becomes larger, making the contours more distinct and the details richer, i.e., the sharper the image, the more reliable the hotspot detection becomes, and the greater the reference value. If there is a significant difference in the change in resolution before and after the increase, it indicates that there is a significant change in the identification of hotspots after the resolution adjustment, the infrared image resolution is too low, the more processing is needed, and the lower the reference value. The higher the confidence level of the hotspot, the better the infrared image at the specified resolution can represent the characteristics of the hotspot region, and the greater the reference value. In one embodiment of the present invention, the method for obtaining the reference value includes: The sharpness ratio of the non-defect region between the initial infrared resolution and the next super-resolution is obtained as an adjustment factor; Obtain the product of the adjustment factor and the feature significance of the super-resolution, calculate the difference between the feature significance of the initial infrared resolution and the product result, and use it as the amount of significance change between corresponding resolutions; The confidence level of the infrared image at the initial infrared resolution is obtained as a percentage of the significant change between the corresponding resolutions, and then normalized and mapped to serve as a reference value for the initial infrared resolution.

[0049] Step S303: Determine whether the initial infrared resolution needs to be adjusted based on the reference value of the initial reference resolution.

[0050] Preferably, in one embodiment of the present invention, if the reference value of the initial reference resolution is less than or equal to a preset reference threshold, it is determined that the corresponding initial reference resolution needs to be adjusted.

[0051] It should be noted that, in one embodiment of the present invention, the preset value threshold is 0.8; in other embodiments of the present invention, the preset value threshold may be set according to specific circumstances, and will not be limited or elaborated here.

[0052] If adjustments are needed, the next super-resolution will be used as the new initial infrared resolution until no further adjustments are required, thus obtaining the target infrared resolution.

[0053] It should be noted that, in one embodiment of the present invention, the method for obtaining the target infrared resolution includes: after determining that adjustment is needed, taking the next super-resolution as the new initial infrared resolution, obtaining the reference value of the super-resolution for judgment, until no adjustment is needed, and taking the new initial infrared resolution as the target infrared resolution.

[0054] Step S5: Detect hot spots on the photovoltaic panel based on the target infrared resolution and the target visible light resolution.

[0055] Based on this, the target infrared resolution provides the best recognition effect for hot spots on infrared images, which helps to accurately analyze the authenticity of hot spots; the target visible light resolution can effectively obtain images with minimal information loss, which helps to obtain comprehensive hot spot areas for analysis; therefore, after obtaining appropriate target infrared and target visible light resolutions, the visible light image under the target visible light resolution and the infrared image under the target infrared resolution are registered to perform accurate hot spot localization, reduce the computational power requirements of the dual model, and improve the efficiency and accuracy of detection.

[0056] In summary, this invention obtains the hot spot matching region for each hot spot region in a visible light image across different super-resolution images based on the positional distribution of hot spot regions between visible light images and different super-resolution images. Based on the hot spot confidence score, the number of hot spot matching regions corresponding to all hot spot regions in the visible light image in different super-resolution images, and the number of hot spot regions in each super-resolution image, a target visible light resolution is selected. Using the target visible light resolution as the initial infrared resolution, the gradient distribution, temperature distribution, and corresponding hot spot confidence score of the photovoltaic panel at different positions in the infrared images between the initial infrared resolution and the next super-resolution image determine whether the initial infrared resolution needs adjustment. If adjustment is required, the next super-resolution image is used as the new initial infrared resolution until no further adjustment is needed, thus obtaining the target infrared resolution. This achieves the detection of hot spots on photovoltaic panels. This invention improves the efficiency of hot spot detection by obtaining accurate detection resolution.

[0057] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0058] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An optimization method for photovoltaic panel hot spot detection based on a dual-model dynamic architecture, characterized in that, The method includes: Acquire visible light and infrared images of a photovoltaic panel containing a hot spot region during dynamic operation, wherein the hot spot region contains a hot spot confidence label; Multiple super-resolution images of the visible light image in the initial dynamic architecture are obtained. Based on the positional distribution of hot spot regions between the visible light image and different super-resolution images, the hot spot matching region of each hot spot region in the visible light image is obtained in different super-resolution images. The target visible light resolution is selected based on the hot spot confidence, the number of hot spot matching regions, and the number of hot spot regions in each super-resolution image corresponding to all hot spot regions in the visible light image. Using the target visible light resolution as the initial infrared resolution, the gradient distribution, temperature distribution, and corresponding hot spot confidence of the photovoltaic panel at different locations in the infrared image between the initial infrared resolution and the next super-resolution determine whether the initial infrared resolution needs to be adjusted. If adjustment is required, the next super-resolution is used as the new initial infrared resolution until no adjustment is needed, thus obtaining the target infrared resolution. Based on the target infrared resolution and the target visible light resolution, the hot spot detection of the photovoltaic panel is achieved.

2. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 1, characterized in that, The method for obtaining the hot spot matching region includes: The center positions of hot spot regions in different adjacent super-resolution images are matched, and the corresponding matched hot spot regions are used as hot spot matching regions to obtain the hot spot matching regions of each hot spot region in the visible light image in different super-resolution images.

3. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 1, characterized in that, The method for obtaining the target visible light resolution includes: Based on the hot spot confidence of each hot spot region in the visible light image corresponding to the hot spot matching region in different super-resolution images, and the number of hot spot matching regions, the local information loss degree of each hot spot region in each super-resolution image is obtained. The overall information loss of each super-resolution image is obtained based on the local information loss of all hot spot regions in each super-resolution image and the difference in the number of hot spot regions in each super-resolution image and the visible light image. The super-resolution image with the smallest overall information loss is selected, and the resolution of the corresponding super-resolution image is taken as the target visible light resolution.

4. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 3, characterized in that, The method for obtaining the local information loss includes: For any hot spot region in a visible light image, the hot spot confidence difference of each hot spot region in the hot spot matching region between each super-resolution image and the next super-resolution image is obtained as the first confidence difference. Obtain the number of hotspot matching regions for each hotspot region in different super-resolution images and the comparison difference between the order of each super-resolution image; calculate the sum of the positive integer 1 and the comparison difference as the first sum; calculate the ratio of the first sum to the number of hotspot matching regions as the information loss weight. The product of the first confidence difference and the information loss weight is obtained and normalized to form the local information loss of each hotspot region in each super-resolution image.

5. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 4, characterized in that, The method for obtaining the overall information loss includes: The sum of the differences between the positive integer 1 and the number of hot spot regions is obtained, and the sum is divided by the number of hot spot regions as the loss coefficient. The cumulative value of the local information loss of all hot spot regions in each super-resolution image is obtained as the cumulative loss value; The product of the accumulated loss and the loss coefficients is calculated as the overall information loss for each super-resolution image.

6. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 1, characterized in that, The determination of whether the initial infrared resolution needs to be adjusted includes: Based on the gradient distribution, temperature distribution, and corresponding hot spot confidence at different locations in an infrared image at any resolution, the feature saliency at the corresponding resolution is obtained. The reference value of the initial infrared resolution is obtained based on the saliency of features between the initial infrared resolution and the next resolution, the gradient distribution of the infrared image, and the corresponding hot spot confidence. Based on the reference value of the initial reference resolution, determine whether the initial infrared resolution needs to be adjusted.

7. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 6, characterized in that, The method for obtaining the saliency of the features includes: For hot spot or non-hot spot regions in an infrared image at any resolution, the mean value of the gradient magnitude of all pixels in each region is obtained as the sharpness of each region. The difference in the average temperature of all pixels between hot spot and non-hot spot regions in an infrared image is obtained as the temperature difference. The ratio of the sharpness difference between the hot spot region and the non-hot spot region to the temperature difference is obtained. The product of the ratio result and the sharpness of the hot spot region is calculated as the adjustment necessity at the corresponding resolution. The ratio of the hot spot confidence and the adjustment necessity of the corresponding infrared image is obtained and normalized and mapped as the feature saliency at the corresponding frequency.

8. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 7, characterized in that, The methods for obtaining the reference value include: The sharpness ratio of the non-defect region between the initial infrared resolution and the next super-resolution is obtained as an adjustment factor; Obtain the product of the adjustment factor and the feature significance of the super-resolution, calculate the difference between the feature significance of the initial infrared resolution and the product result, and use it as the amount of significance change between corresponding resolutions; The confidence level of the infrared image at the initial infrared resolution is obtained as a percentage of the significant change between the corresponding resolutions, and then normalized and mapped to serve as a reference value for the initial infrared resolution.

9. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 6, characterized in that, The method for determining whether the initial infrared resolution needs adjustment based on its reference value includes: If the reference value of the initial reference resolution is less than or equal to the preset reference threshold, it is determined that the corresponding initial reference resolution needs to be adjusted.

10. The photovoltaic panel hot spot detection optimization method based on a dual-model dynamic architecture according to claim 1, characterized in that, The method for obtaining the target infrared resolution includes: Once it is determined that adjustment is needed, the next super-resolution is used as the new initial infrared resolution. The reference value of the super-resolution is used for further judgment until no adjustment is needed, at which point the new initial infrared resolution is used as the target infrared resolution.

Images