Artificial intelligence-based solar panel defect classification method and system

Abstract

The present invention relates to an artificial intelligence-based method and system for classifying defects in solar panels. More specifically, the artificial intelligence-based method and system for classifying defects in solar panels according to the present invention may include the steps of: extracting pixel locations from an image including a solar panel and generating a filtered image by distorting at least a portion of the pixel locations; inputting the filtered image into a GAN model to generate a synthetic image that mimics a defect image and a normal image; and classifying classes in a Faster-RCNN model based on the synthetic image to detect defects in the solar panel from the image.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence-based method and system for classifying defects in solar panels. More specifically, the present invention relates to an artificial intelligence-based method and system for classifying defects in solar panels that can minimize energy loss of solar panels and extend the lifespan of a solar power system. Background Technology

[0002] Recently, the demand for renewable energy has been rapidly increasing worldwide due to the depletion of fossil fuels and the intensification of environmental pollution issues. Among these, solar energy plays a pivotal role in the renewable energy sector due to its abundant resources, stable supply potential, and low environmental impact. Based on advantages such as ease of installation, declining module costs, and long-term operational stability, solar power generation technology is steadily expanding its output and broadening its scope of application across various fields, including industry, households, and urban infrastructure. Consequently, interest in solar panel technology—aimed at maintaining stable output and maximizing power generation efficiency in solar power systems—is also on the rise.

[0003] However, since the power generation efficiency of solar panels is directly affected by external environmental factors and physical damage, it is difficult to ensure stable power generation performance over the long term. Specifically, airborne materials such as dust, soil, contaminants, and bird droppings that accumulate on the panel surface reduce light transmittance, leading to a decrease in power generation. In particular, fine dust deposition occurs rapidly in hot and dry regions, while localized shading caused by bird droppings easily forms in coastal and suburban areas. These contaminants cause output imbalances at the individual solar panel cell level, and in severe cases, can even trigger hotspot phenomena, leading not only to power loss but also to a decline in panel durability.

[0004] Furthermore, structural defects such as panel cracks, chipping, microcracks, electrical failures, surface scratches, and backsheet damage are known to be major factors in reducing power generation efficiency. Additionally, as extreme weather conditions like snowfall and strong winds become more frequent due to climate change, additional defect factors such as snow cover and film deformation on the panel surface are also increasing. These various forms of defects and contamination factors ultimately lead to reduced power generation efficiency, increased maintenance costs, and problems that undermine long-term operational efficiency and reliability.

[0005] Currently, various methods such as visual inspection, thermal imaging-based diagnosis, IV curve analysis, and drone-based inspection are utilized to check for defects in solar panels; however, these methods have several limitations, including a lack of real-time capability, high reliance on human labor, the burden of equipment costs, and restricted inspection areas. In particular, given that large-scale power plants must manage thousands to tens of thousands of panels, there is a growing need for automated, intelligent diagnostic technologies capable of rapid and accurate defect detection. Accordingly, deep learning-based image analysis technology has recently been widely applied to solar panel defect diagnosis; however, limitations in improving accuracy persist due to difficulties in collecting defect images and data asymmetry. Consequently, there is a need for AI-based solar panel defect detection technology that can minimize energy loss and extend the lifespan of solar power systems. The problem to be solved

[0006] The present invention is intended to provide an artificial intelligence-based solar panel defect classification method and system capable of minimizing energy loss of solar panels and extending the lifespan of solar power systems.

[0007] In addition, the present invention is intended to provide an artificial intelligence-based solar panel defect classification method and system capable of automating solar panel defect inspection.

[0008] Furthermore, the present invention aims to provide an artificial intelligence-based solar panel defect classification method and system capable of maintaining high classification performance even in situations where it is difficult to obtain images of solar panel defects. means of solving the problem

[0009] To solve the problem described above, the artificial intelligence-based solar panel defect classification method according to the present invention may include the steps of: extracting pixel locations from an image including a solar panel and generating a filtered image by distorting at least a portion of the pixel locations; inputting the filtered image into a GAN model to generate a synthetic image that mimics a defective image and a normal image; and classifying classes in a Faster-RCNN model based on the synthetic image to detect defects in the solar panel from the image.

[0010] Furthermore, the above GAN model may include a first block for generating the synthetic image and a second block for determining whether the generated synthetic image is true or false.

[0011] Furthermore, in the step of generating the above-mentioned composite image, a composite image that can be determined to be true in the above-mentioned second block can be generated in the above-mentioned first block.

[0012] Furthermore, in the step of detecting the defect, the defect of the solar panel can be detected from the image by classifying it in the Faster-RCNN model based on the synthetic image determined to be true in the second block.

[0013] Furthermore, the first block and the second block can generate the synthetic image for focusing on the crack by performing self-attention.

[0014] Furthermore, the method may further include the step of generating an enhanced image in which at least some pixels of the image are enhanced using the filtered image.

[0015] Furthermore, in the step of generating the above-mentioned emphasis image, the emphasis image can be generated by emphasizing pixels whose gradient values ​​of the filtered image satisfy a preset standard.

[0016] Furthermore, in the step of generating the above-mentioned composite image, at least one noise for generating the above-mentioned composite image can be generated using the filtered image, and the above-mentioned composite image can be generated using the generated noise.

[0017] Meanwhile, the artificial intelligence-based solar panel defect classification system according to the present invention may include a memory for storing instructions and at least one processor electrically connected to said memory, and when said instructions are executed by said at least one processor, said at least one processor extracts pixel locations from an image including a solar panel, distorts at least a portion of said pixel locations to generate a filtered image, inputs said filtered image into a GAN model to generate a synthetic image simulating a defective image and a normal image, and classifies classes in a Faster-RCNN model based on said synthetic image to detect defects of said solar panel from said image.

[0018] Meanwhile, the program according to the present invention is a program that is executed by one or more processes in an electronic device and can be stored on a computer-readable recording medium, and the program may include instructions for performing the steps of: extracting pixel locations from an image including a solar panel and generating a filtered image by distorting at least a portion of the pixel locations; inputting the filtered image into a GAN model to generate a synthetic image that mimics a defective image and a normal image; and classifying classes in a Faster-RCNN model based on the synthetic image to detect defects in the solar panel from the image. Effects of the invention

[0019] As described above, the artificial intelligence-based solar panel defect classification method and system according to the present invention can detect and classify defects in solar panels using an artificial intelligence model. Through this, even minute defects or atypical contamination elements present within solar panel images can be precisely recognized, and problems that may lead to a decrease in power generation efficiency can be detected early by systematically analyzing the shape, location, and scale of the defects.

[0020] Furthermore, the artificial intelligence-based solar panel defect classification method and system according to the present invention can stably train a defect classification model by generating synthetic training data containing various defect patterns even in environments where solar panel images are scarce. Through this, the problem of data bias caused by a lack of actual defect images is reduced, and by sufficiently learning in advance the defect characteristics that may exist in images captured under various environments, angles, and illumination levels, high accuracy and consistent performance can be maintained even in actual operating environments.

[0021] Furthermore, the artificial intelligence-based solar panel defect classification method and system according to the present invention can automate the condition diagnosis process of solar panels based on images highlighting defect features and a trained artificial intelligence model. Through this, even in environments where the number of panels to be inspected is very large, such as large-scale solar power plants, where conventional manual inspection would require excessive time and cost, it is possible to quickly identify whether the panels are damaged and make immediate maintenance decisions. Brief explanation of the drawing

[0022] FIG. 1 is a block diagram illustrating a computing system in which the present invention can be implemented. FIGS. 2 and FIGS. 3 are block diagrams illustrating an embodiment of a computing device according to the present invention. FIG. 4 is a conceptual diagram illustrating an artificial intelligence-based solar panel defect classification system according to the present invention. FIG. 5 is a flowchart illustrating an artificial intelligence-based solar panel defect classification method according to the present invention. FIG. 6 is a conceptual diagram illustrating a method for collecting solar panel images according to the present invention. FIG. 7 is a conceptual diagram illustrating defect classes and labeling of solar panel images according to the present invention. FIGS. 8 and 9 are conceptual diagrams for explaining a method for generating an enhanced image according to the present invention. FIGS. 10, FIGS. 11 and FIGS. 12 are conceptual diagrams for explaining a learning method of an artificial intelligence model according to the present invention. FIG. 13 is a conceptual diagram illustrating the results of a solar panel defect classification by an artificial intelligence model according to the present invention. FIG. 14 shows formulas related to the artificial intelligence-based solar panel defect classification method and system according to the present invention. Specific details for implementing the invention

[0023] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components are assigned the same reference number regardless of the drawing symbols, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not have distinct meanings or roles in themselves. Furthermore, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the embodiments disclosed in this specification, such detailed description will be omitted. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings, and it should be understood that they include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention.

[0024] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. These terms are used solely for the purpose of distinguishing one component from another.

[0025] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0026] Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0027] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0028] Hereinafter, the present invention will be examined in more detail with reference to the attached drawings. FIG. 1 is a block diagram illustrating a computing system in which the present invention can be implemented. FIG. 2 and FIG. 3 are block diagrams illustrating an embodiment of a computing device according to the present invention. FIG. 4 is a conceptual diagram for explaining an artificial intelligence-based solar panel defect classification system according to the present invention.

[0029] Furthermore, the solar panel defect classification system (100) according to the present invention can be implemented through a computing device described below and can perform data processing related to the solar panel defect classification method described above.

[0030] Meanwhile, FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented.

[0031] Referring to FIG. 1, a computing system (10000) for performing a solar panel defect classification method according to one embodiment of the present invention may include at least one computing device. At this time, the at least one computing device may be a single processor or a multi-processor computing device.

[0032] The components of at least one computing device of the present invention may include various hardware components such as one or more processors, memory, other hardware, and a system bus (not shown) that connects various system components so that they can transmit and receive data to and from each other (e.g., telecommutatively connected, physically connected, electrically connected), and the components of at least one computing device are not limited thereto and may be very diverse.

[0033] Meanwhile, at least one computing device included in the computing system (10000) that performs the solar panel defect classification method may be connected to communicate via a network (1070). For example, at least one computing device included in the computing system (10000) may be clustered or part of a local area network (LAN). Additionally, at least one computing device may be part of a wide area network (WAN) or connected to at least one of a client-server network and a peer-to-peer network within the cloud.

[0034] Meanwhile, when at least one computing device is used in at least one of a network environment and a cloud computing environment, the at least one computing device may be connected to at least one of a public and private network through a network interface or adapter. In one embodiment, other communication connection devices, such as a modem, may be used to establish communication through the network. The modem may be at least one of an internal modem and an external modem, and may be connected to a system bus through a network interface or a specific mechanism, etc. A wireless network component consisting of an interface and an antenna may be coupled to the network through a device such as an access point, a peer computer, etc. In the present invention, the method of connecting at least one computing device to communicate through the network (1070) is not limited, and it may be connected to communicate in a manner different from the described example.

[0035] Furthermore, other computer-type devices and / or systems not shown in FIG. 1 may also interact technically with at least one computing device or other system through one or more connections to the network (1070) via a network interface. Here, the network interface may include network interface equipment such as a physical network interface controller (NIC) or a virtual network interface (VIF).

[0036] The network (1070) of the present invention may include various forms such as the Internet, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™ RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, Wireless USB (Wireless Universal Serial Bus), etc., and in the present invention, data transmission may be performed based on standard communication protocols such as TCP / IP, HTTP, SSL, etc.

[0037] A computing system (10000) for performing a solar panel defect classification method according to the present invention may include at least one of a user computing device (1010), a training computing system (1050), and a server computing system (1030).

[0038] A user computing device (1010) according to the present invention may be understood as a computing device comprising at least one processor (1011) and a memory (1012) for performing a solar panel defect classification method. For example, the user computing device (1010) may include at least one computing device among a smartphone, a smart TV, a laptop computer, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, a slate PC, a tablet PC, an ultrabook, a wearable device (e.g., a smartwatch, a smart glass, and a head-mounted display).

[0039] At least one processor (1011) constituting the user computing device (1010) may include one or more general-purpose processors and / or one or more special-purpose processors. For example, at least one processor (1011) constituting the user computing device (1010) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), an application integrated circuit, an application semiconductor (ASIC), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0040] Furthermore, at least one processor (1011) may be configured to execute computer-readable instructions contained in memory (1012) and / or other instructions described herein.

[0041] The memory (1012) constituting the user computing system (1010) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media and / or other types of physically durable storage media.

[0042] For example, the memory (1012) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, and combinations thereof, and may include web storage of a server that performs the storage function of memory on the internet. This memory (1012) may store data and instructions necessary for the at least one processor (1011) to perform the operation of an application for classifying solar panel defects.

[0043] A user computing device (1010) may include one or more user input components (1021) that detect user input. For example, the user input component (1021) may also be referred to as a user interface module. The user input component (1021) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of user input component (1021). In this case, the user input component (1021) in the present invention does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user. Meanwhile, the user of the present invention may refer to an automated agent, script, playback software, etc., that operates on behalf of one or more people.

[0044] A user can interact with a computing system (10000) including at least one computing device through input text, touch, voice, movement, computer vision, gestures and / or other forms of input / output using a user input component (1021). For example, the user input component (1021) may include one or more of a command line interface (CLI), a graphical user interface (GUI), a natural user interface (NUI), a voice command interface and / or other user interface (UI) representations.

[0045] Between the user input component (1021) and the user computing device (1010), one or more application programming interface (API) calls may be made based on user input received from the user interface and / or network.

[0046] Here, the expression "based on" may be interpreted to include cases where it is based on the use of a specific configuration, modified from, derived from, influenced by, dependent on, or otherwise derived from a specific configuration. In some embodiments, an API call may be configured for a specific API, which may be interpreted or converted into an API call configured for another API. Here, an API may refer to a defined interface or connection between computers or between computer programs.

[0047] In one embodiment, the user computing device (1010) may store at least one machine learning model (1020). For example, the user computing device (1010) may be various machine learning models, such as a plurality of neural networks (e.g., deep neural networks) that perform solar panel defect classification using solar panel images, or other types of machine learning models including non-linear models and / or linear models, and may be composed of a combination thereof.

[0048] According to an embodiment of the present invention, a user computing device (1010) may perform a solar panel defect classification method using a local or / and external machine learning model (1020). Alternatively, the user computing device (1010) may perform a solar panel defect classification method using a machine learning model (1040) provided by a server.

[0049] In addition, according to another embodiment of the present invention, a server computing system (1030) communicating with a user computing device (1010) may provide a solar panel defect classification result to the user computing device (1010) on an application or / and the web in accordance with a request from a user received through the user computing device (1010).

[0050] In addition, according to another embodiment of the present invention, at least a part of the user computing device (1010) and the server computing system (1030) are interconnected to perform a solar panel defect classification method, thereby providing the solar panel defect classification result to the user.

[0051] Additionally, according to various embodiments of the present invention, a user computing device (1010) and / or a server computing system (1030) can learn machine learning models (1020, 1040) performed in a solar panel defect classification method through interaction with a training computing system (1050) that is communicatedly connected via a network (1070). In this case, the training computing system (1050) may be a computing system separate from the server computing system (1030). Alternatively, in some embodiments, the training computing system (1050) may be part of the server computing system (1030) or part of the user computing device (1010).

[0052] Meanwhile, the server computing system (1030) may include at least one processor (1031) and memory (1032). Here, the processor (1031) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an application integrated circuit, an application semiconductor (ASIC), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions. For example, at least one processor (1031) may include a circuit and a transistor configured to execute instructions from memory (1032).

[0053] The memory (1032) constituting the server computing system (1030) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media, and / or other types of physically durable storage media. For example, the memory (1032) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, etc., and combinations thereof, and may include web storage of a server that performs the storage function of memory over the internet. Additionally, the server computing system (1030) may further include a data storage (data store). For example, the data storage may be composed of at least one of a relational database, a NoSQL database, a data warehouse, and a local file system.

[0054] In the memory (1032) constituting the server computing system (1030) according to the present invention, data and instructions necessary for the at least one processor (1031) to perform the operation of an application for classifying solar panel defects may be stored.

[0055] In one embodiment, the server computing system (1030) may be composed of a single device or a plurality of computing devices, and these may be configured to operate according to a sequential or parallel computing architecture. Additionally, a distributed processing system may be configured with a plurality of networked devices.

[0056] Meanwhile, the training computing system (1050) may include at least one processor (1051) and memory (1052). The model trainer (1060) is a logical component that executes the training of at least one machine learning model (1020, 1040) and may be implemented in the form of hardware, firmware, or software. For example, the model trainer (1060) may be executed by the processor (1051) after loading training data (1061) stored in a storage device into memory (1052). For example, the model trainer (1060) may be configured to execute one or more operations (e.g., model training, model reconstruction, model validation, model testing) on ​​at least one machine learning model.

[0057] The machine learning model of the present invention may include at least one of a statistical model, an algorithm, a neural network (NN), a convolutional neural network (CNN), a generative neural network (GNN), a Word2Vec model, a Bag of Words model, a TF-IDF (document frequency-inverse document frequency) model, a GPT (Generative Pre-trained Transformer) model (or other autoregressive models), a PPO (Proximal Policy Optimization) model, a nearest neighbor model (e.g., a k-nearest neighbor model), a linear regression model, a K-means clustering model, a Q-learning model, a TD (Temporal Difference) model, a Deep Adversarial Network model, and all other types of models further described herein.

[0058] Specifically, the model trainer (1060) may execute operations to train a machine learning model, and said operations may include at least one of adding, removing, and modifying model parameters. At this time, the training of the machine learning model may be at least one of supervised learning, semi-supervised learning, and unsupervised learning. In one embodiment, the training of the machine learning model may include the step of repeatedly inputting training data (1061) based on epochs and repeatedly performing the machine learning model training process configured in this way. Here, an epoch may refer to a unit in which the entire set of training data (1061) undergoes forward and backpropagation processing once. In some implementations, different levels of training methods (e.g., supervised learning, semi-supervised learning, unsupervised learning) may be used for different epochs.

[0059] The training data (1061) of the present invention may include input data and / or data previously output from at least one machine learning model (e.g., recursive learning feedback).

[0060] At least one parameter of a machine learning model may include at least one of a seed value, a model node, a model layer, an algorithm, a function, connections between different machine learning models, connections between parameters, machine learning model constraints, and other digital components that influence the output of the machine learning model. In this case, model connections between different machine learning models may include or represent relationships between model parameters and / or models, which may be dependent or interdependent, hierarchical, and / or static or dynamic. The combinations and configurations of model parameters described herein may be too complex to be maintained or utilized by human cognitive abilities.

[0061] In the present invention, the machine learning parameters described according to the embodiments are not limited, and a single machine learning model may further include a plurality of model parameters.

[0062] Meanwhile, FIG. 2 illustrates an example of a block diagram of a computing device (1100) that may be included in a user computing device (1010), a server computing system (1030), and a training computing system (1050), as an embodiment of a computing system (10000) in which the present invention can be implemented.

[0063] As illustrated in FIG. 2, the computing device (1100) may include at least one application (e.g., Application 1 to Application N), and each of the at least one application may include a machine learning library and a model execution environment for performing a machine learning-based solar panel defect classification method. The at least one application included in the computing device (1100) may communicate with the sensor, context manager, device state manager, or additional component(s) within the computing device (1100) via an Application Programming Interface (API). In one embodiment, the at least one application may interface with device components, such as receiving sensor data or state data or transmitting prediction results to an output device via a public or private API.

[0064] Meanwhile, FIG. 3 illustrates an example of a block diagram in another aspect of a computing device (1200), which is one of the components of a computing system (10000) that performs a solar panel defect classification method according to an embodiment of the present invention.

[0065] A computing device (1200) according to the present invention may include at least one application (e.g., Application 1 to Application N), and at least one application may communicate with a central intelligence layer (1210). Each application may interact with a shared model within the central intelligence layer (1210) through an API (e.g., a common API).

[0066] The central intelligence layer (1210) includes one or more machine learning models and may share them among multiple applications or provide them independently to each. In one embodiment, the central intelligence layer (1210) may be integrated as part of an operating system or implemented as a separate logical layer.

[0067] Additionally, the central intelligence layer (1210) can communicate with the central device data layer (1220). The central device data layer (1220) can integrate and store solar panel images, etc., stored within the computing device (1200) and provide them as input data required for solar panel defect classification. Each device component (e.g., sensor, state manager, etc.) can communicate with the central device data layer (1220) through a private API, etc.

[0068] The technology described herein may be composed of a single or multiple computing devices, and a machine learning model for performing a solar panel defect classification method may be executed sequentially or in parallel on one component or multiple distributed components. Data storage, machine learning models, and applications may be distributed and operated locally or over a network, and these configurations can be flexibly applied to various system architectures.

[0069] The present invention relates to an artificial intelligence-based solar panel defect classification method and system. More specifically, the artificial intelligence-based solar panel defect classification system according to the present invention may be a system capable of detecting defects from solar panel images and classifying the types of defects.

[0070] Here, a defect according to the present invention may refer to a factor that can reduce the power generation efficiency of a solar panel or the structural stability of the panel. For example, it may include at least one of contaminants occurring on the surface of the solar panel, such as dust and bird droppings; cracks and damage formed on the panel cells or surface; abnormal temperature distribution or discoloration caused by electrical anomalies; or obstruction of view caused by snow or external foreign substances.

[0071] The present invention enables an artificial intelligence model to be trained to classify defects in a solar panel image using an enhanced image that emphasizes the boundary information of the solar panel image. Specifically, the present invention transforms the pixel locations of a solar panel image based on a pre-set criterion, generates an enhanced image with emphasized edge information using the solar panel image with transformed pixel locations, and trains an artificial intelligence model to classify defects in the solar panel image using the generated enhanced image.

[0072] Meanwhile, as illustrated in FIG. 4, the solar panel defect classification system (100) may include at least one of an input unit (110), a storage unit (120), a communication unit (130), an output unit (140), and a control unit (150). However, the components of the solar panel defect classification system (100) according to the present invention are not limited thereto and may further include various hardware components that perform the same or similar roles as described in the description of the present specification.

[0073] Although not illustrated, the solar panel defect classification system (100) may include one or more processors, and such processors may include one or more general-purpose processors and / or one or more special-purpose processors (e.g., digital signal processors, tensor processing units (TPUs), graphics processing units (GPUs), neural network processing units (NPUs), application integrated circuits, application semiconductors (ASICs), etc.). One or more processors may be configured to execute instructions, computer-readable instructions, and / or other instructions described herein that are stored (or included) in the storage unit (120). The solar panel defect classification system (100) may perform data processing described below in cooperation with memory and at least one processor. The processor may perform a series of operations and data processing using data and information stored in memory. In this case, memory may be a component of the storage unit (120).

[0074] Meanwhile, the solar panel defect classification system (100) according to the present invention may exist inside a server (hereinafter referred to as the server) established to perform a specific purpose (e.g., generating solar panel defect classification results), or it may exist as a separate system from said server. When the solar panel defect classification system (100) exists inside the server, various services related to the present invention (e.g., generating solar panel defect classification results, etc.) can be provided through at least one component located inside the server, or through configuration modules that perform functions similar to said components. In this case, an application can provide various services related to the present invention on an electronic device on which the application is installed through communication with the server.

[0075] The input unit (110) can be configured in various ways as a means of data input. For example, the input unit (110) can be configured to receive user input. The input unit (110) can be configured to receive user input from a user terminal. Here, “receiving input” may mean receiving an input signal (or selection signal) corresponding to the user’s input based on input being made by the user through the configuration of the input unit provided in the user terminal.

[0076] The input unit (110) may also be referred to as a user interface module. The input unit may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of input unit (110). Furthermore, in the present invention, the input unit does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user.

[0077] Here, user input may include documents, text, images (or videos), voice, etc. In this case, the solar panel defect classification system (100) may further include a module that converts voice into text.

[0078] The storage unit (120) may include computer-readable instructions and additional data. The storage unit (120) may include a storage necessary to perform at least some of the methods, scenarios, and techniques described herein and / or at least some of the functions of the device and network.

[0079] Furthermore, at least a portion of the storage unit (120) may be a cloud storage or a cloud server. At least a portion of the data corresponding to user input received from the input unit (110) and the training data may be stored in the storage unit (120).

[0080] That is, the storage unit (120) is sufficient as a space where information necessary for the operation of the solar panel defect classification system (100) is stored, and it can be understood that there are no restrictions on the physical space.

[0081] Meanwhile, the communication unit (130) may be connected via a wireless or wired network to a user terminal, a server (e.g., a central server, an external server, etc.), a device, and at least one network, and may be configured to receive or transmit overall data and information necessary for the operation of the solar panel defect classification system (100).

[0082] Here, the user terminal may include at least one of a mobile phone, a smartphone, a notebook computer, a laptop computer, a slate PC, a tablet PC, an ultrabook, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, and a wearable device (e.g., a smartwatch, a smart glass, a head-mounted display).

[0083] Furthermore, the communication unit (130) can support various communication methods according to the communication standards of the communicating device.

[0084] For example, the communication unit (130) may be configured to communicate with a communication target using at least one of the following technologies: WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™ RFID (Radio Frequency Identification), Infrared Communication (Infrared Data Association; IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus).

[0085] Next, the output unit (140) can output information through an output unit configuration (e.g., a display unit, a touch screen, a speaker, etc.) provided in a user terminal linked to the solar panel defect classification system (100). For example, the output unit (140) can output a page or service page linked to the solar panel defect classification system (100) to the display unit of the user terminal. Additionally, the output unit (140) does not necessarily mean a hardware means, but can be understood as a channel for outputting results to the user.

[0086] More specifically, the output unit (140) can provide the user with the solar panel defect classification results.

[0087] Next, the control unit (150) can perform the role of controlling the overall operation of the solar panel defect classification system (100) related to the present invention. Specifically, the control unit (150) may include at least one of an image preprocessing unit (151) and an artificial intelligence model (152). At this time, although the present invention describes the control unit (150) as including at least one of the image preprocessing unit (151) and the artificial intelligence model (152), it is not limited thereto. If at least one of the image preprocessing unit (151) and the artificial intelligence model (152) exists on an external server, the control unit (150) may call or control at least one of the image preprocessing unit (151) and the artificial intelligence model (152) by linking with the external server.

[0088] The control unit (150) can generate an enhanced image using the image preprocessing unit (151).

[0089] The control unit (150) can generate a composite image using an artificial intelligence model (152) and classify cracks in a solar panel image using at least one of the enhanced image and the composite image.

[0090] The artificial intelligence model (152) according to the present invention may be composed of a single model, but is not limited thereto, and may be configured to operate independently or in conjunction with each other. Furthermore, the artificial intelligence models may be implemented based on the same structure or algorithm, but may include different models designed to perform different structures, different learning methods, or different purposes as needed. For example, a GAN model may be distinguished to perform a function for generating synthetic images, while another Faster-RCNN model may perform a discrimination or classification function, or multiple models performing the same function may operate in an auxiliary manner. As such, the present invention is not limited to a specific number of models or models of a specific structure, and various forms of artificial intelligence model configurations may be applied. For convenience of explanation, the models used in the present invention, including the GAN model and the Faster-RCNN model, are not distinguished below and are described as artificial intelligence models (152); however, this is not intended to limit the invention to a specific model structure or functional configuration, and the present invention can be equally applied to various forms of artificial intelligence model configurations.

[0091] The control unit (150) can process signals, data, information, etc. that are input or output through the components of the solar panel defect classification system (100) described above, or perform a series of data processing to provide or process appropriate information and functions to the user. The control unit (150) can be physically implemented by the processor described above.

[0092] The solar panel defect classification system (100) according to the present invention may aim to generate a defect classification result (20) of a solar panel by taking a solar panel image (10) as input.

[0093] In the foregoing, the solar panel defect classification system (100) of the present invention has been described, and it can operate to implement the solar panel defect classification method described below. FIG. 5 is a flowchart for explaining the artificial intelligence-based solar panel defect classification method according to the present invention. FIG. 6 is a conceptual diagram for explaining the solar panel image collection method according to the present invention. FIG. 7 is a conceptual diagram for explaining the defect class and labeling of the solar panel image according to the present invention. FIG. 8 and FIG. 9 are conceptual diagrams for explaining the enhanced image generation method according to the present invention. FIG. 10, FIG. 11 and FIG. 12 are conceptual diagrams for explaining the learning method of the artificial intelligence model according to the present invention. FIG. 13 is a conceptual diagram for explaining the solar panel defect classification result of the artificial intelligence model according to the present invention. FIG. 14 is a set of formulas related to the artificial intelligence-based solar panel defect classification method and system according to the present invention.

[0094] Meanwhile, in the present invention, a process may be performed to extract pixel locations from an image including a solar panel and to generate a filtered image by distorting at least a portion of the pixel locations (S510, see FIG. 5).

[0095] The control unit (150) can receive an image of a solar panel (i.e., a solar panel image, 10). At this time, the received solar panel image (10) is not limited to a specific device or shooting method and can be provided from any external device or external server capable of generating or transmitting image data.

[0096] For example, as illustrated in FIG. 6, the solar panel image (10) may be an image captured from at least one of a surveillance camera, a satellite, a drone, and a mobile device. However, the present invention is not limited thereto, and the solar panel image may be an image generated or captured by any type of device, such as a fixed-mount device, a mobile device, a remote-controlled device, or a network-based image collection device. Additionally, the solar panel image (10) may be various types of image data, such as a real-time captured image, as well as a stored image, a streaming image, and a pre-processed image.

[0097] Additionally, the solar panel image (10) may include defects (30) that may cause performance degradation of the solar panel, and the defects (30) may be included as information that visually indicates the surface or internal condition of the solar panel regardless of the type of shooting device, shooting environment, shooting distance, etc. The defects (30) may include any pattern on the image indicating contaminants, snow accumulation, cracks, damage, discoloration, shadows, or other abnormal conditions, and may appear without limitation in shape, size, or location.

[0098] The control unit (150) can classify the received solar panel image (10) into at least one of a plurality of classes to build training data for training an artificial intelligence model.

[0099] For example, as illustrated in FIG. 7, the control unit (150) can classify the received solar panel image (10) into at least one of the classes: clean, bird drop, snow covered, electric damage, dusty, and breaking. Specifically, as illustrated in FIG. 7 (a), the control unit (150) can classify a solar panel image that is free of defects as training data corresponding to the clean class. The clean class may include images in which defect elements such as contamination, snow cover, discoloration, or cracks do not exist on the panel surface, and can be used to enable an artificial intelligence model to learn the characteristics of a normal state.

[0100] Additionally, as shown in FIG. 7(b), the control unit (150) can classify an image in which bird droppings are attached to the surface of a solar panel as a bird drop class. Bird droppings have irregular shapes and sizes and can be considered as defects that may affect power generation efficiency by obscuring part of the panel surface.

[0101] The snow covered class shown in Fig. 7 (c) may refer to a state where the surface of a solar panel is partially or completely covered by snow or ice. When such an image is received, the control unit (150) classifies it as a snow covered class and can reflect the risk of reduced output or panel damage due to snow in the learning data.

[0102] The electric damage class illustrated in FIG. 7 (d) may include defects caused by electrical abnormalities such as abnormal heating, discoloration, and spot damage of the internal cells of the panel. The control unit (150) can identify an image having the corresponding features and classify it into an electric damage class.

[0103] The dusty class shown in Fig. 7 (e) indicates a state in which dust or soil is accumulated on the panel surface in an area larger than a certain range, and the control unit (150) can classify an image containing such a contamination pattern as a dusty class.

[0104] Finally, the breaking class shown in (f) of FIG. 7 indicates a state in which the panel surface components are damaged due to impact, cracks, fragments, etc., and the control unit (150) can classify an image containing such structural damage patterns as a breaking class.

[0105] That is, the control unit (150) can build training data for training an artificial intelligence model by performing labeling on the received solar panel image (10), thereby classifying the solar panel image (10) into at least one of a plurality of classes.

[0106] The control unit (150) can convert the pixel location of at least one solar panel image (10) included in the constructed training data. Specifically, the control unit (150) can process the solar panel image (10) as an input to the image preprocessing unit (151) in order to convert the pixel location of the solar panel image (10).

[0107] As illustrated in FIG. 8, the image preprocessing unit (151) can extract pixel locations from the input solar panel image (10) and generate a filtered image by distorting at least some of the pixel locations.

[0108] In this regard, referring to FIGS. 8 and FIGS. 9 together, the image preprocessing unit (151) can receive a solar panel image (10) such as FIGS. 9 (a). The image preprocessing unit (151) can perform a grayscale conversion on the solar panel image (10) to generate a grayscale converted image such as FIGS. 9 (b).

[0109] Here, grayscale conversion may refer to an image conversion process that converts multi-channel color information, such as RGB (Red, Green, Blue), included in a solar panel image (10) into a single intensity component to minimize the influence of color information and to express the contrast and structural features of the panel surface more clearly.

[0110] Furthermore, the image preprocessing unit (151) can perform pixel position distortion (or transformation) of the solar panel image (10) based on a pre-set standard (801).

[0111] Here, distorting pixel positions may mean a process of non-linearly changing the spatial structure of the original image by rearranging or transforming the coordinates of each pixel constituting the solar panel image (10) according to a preset operation rule.

[0112] For example, the image preprocessing unit (151) can perform pixel position distortion of the solar panel image (10) based on a chaos-based processing technique to generate a Henon map such as (c) of FIG. 9.

[0113] More specifically, the image preprocessing unit (151) can set initial pixel coordinates of the solar panel image (10) (see FIG. 14 (a)). Furthermore, the image preprocessing unit (151) can perform iterative operations on at least one pixel of the solar panel image (10) using parameters of a Henon map. At this time, the parameters of the Henon map can be expressed as shown in FIG. 14 (b).

[0114] Furthermore, the image preprocessing unit (151) can generate a Henon map based on the formula of (c) of FIG. 14 and perform normalization based on the formula of (d) of FIG. 14 to generate a Henon map image such as (d) of FIG. 9.

[0115] The pixel position distortion process resulting from the generation of a Henon map based on the chaos-based processing technique described above is merely an example, and the present invention may include various forms of preprocessing processes to emphasize texture or boundary features of a solar panel image.

[0116] Furthermore, the image preprocessing unit (151) can generate an enhanced image with boundary information emphasized using a Henon map image (see 802 in FIG. 8). Specifically, the image preprocessing unit (151) can apply a filter kernel to the Henon map image. At this time, the filter kernel can be represented as shown in (e) of FIG. 14.

[0117] Furthermore, the image preprocessing unit (151) can calculate the x-direction and y-direction gradient values ​​of the Henon map image based on the formulas of (f) and (g) of FIG. 14.

[0118] Furthermore, the image preprocessing unit (151) can generate an enhanced image (810) as shown in (f) of FIG. 9 by calculating the gradient magnitude based on the formula (h) of FIG. 14 and marking the corresponding pixel as a boundary as the gradient exceeds a preset threshold (see (i) of FIG. 14 and (j) of FIG. 4).

[0119] Meanwhile, in the present invention, a process may be performed in which a filtered image is input into a GAN model to generate a synthetic image that mimics a defective image and a normal image (S520, see FIG. 5).

[0120] The control unit (150) can use the generated highlighting image (810) to enable an artificial intelligence model to detect defects in the solar panel image.

[0121] As illustrated in FIGS. 10 and 11, the control unit (150) can process a highlight image (or filtered image, 810) as input to an artificial intelligence model (152). The artificial intelligence model (152) can detect defects in a solar panel image using the input highlight image (810).

[0122] To this end, the artificial intelligence model (152) can input the input enhancement image (810) into the first block (820). Here, the first block (820) may be a generator block that generates a synthetic image using the input enhancement image (810).

[0123] The composite image according to the present invention may refer to image data generated to mimic the statistical characteristics or visual structure of an actual solar panel image by using feature information extracted from or generated based on an emphasis image (810).

[0124] The synthetic image has a form similar to an actual captured solar panel image but is an image artificially generated for the training of an artificial intelligence model, and can express defect patterns, boundary characteristics, or texture information in various forms.

[0125] More specifically, the first block (820) can generate at least one noise vector using the input enhancement image (810). The noise vector according to the present invention is a vector generated by reflecting at least some of the characteristics of the enhancement image (810), and may refer to a value used as an input variable for generating a composite image.

[0126] That is, generating a composite image may mean that a first block (or generator block, 820) that receives a noise vector input mimics or modifies the features of an enhanced image (810) to produce new image data.

[0127] The first block (820) can perform a series of processes to generate a synthetic image. For example, the first block (820) can generate a high-dimensional representation by sequentially passing a noise vector through a plurality of transposed convolution layers to generate a synthetic image. At this time, batch normalization for learning stabilization is applied after each transposed convolution layer, and a ReLU activation function for non-linearity may be used. The final layer of the first block (820) may use a Tanh activation function to normalize the pixel values ​​of the output image to the range [-1, 1].

[0128] Furthermore, the artificial intelligence model (152) can process the synthetic image generated in the first block (820) as input to the second block (830). At this time, the second block (830) may be a discriminator block for evaluating whether the synthetic image is true or false.

[0129] Here, determining whether it is true or false may mean determining whether the second block (830) inputs an image based on an actual captured solar panel image or a composite image generated from the first block (820).

[0130] The second block (830) can determine whether the input synthetic image is true or false. The first block (820) and the second block (830) can generate a synthetic image repeatedly based on the determination result so that the determination of the synthetic image becomes true.

[0131] Here, repeatedly generating a synthetic image so that the determination of whether the synthetic image is true may mean a learning process in which the first block (820) continuously adjusts or updates the method of generating the synthetic image so that the determination result produced in the second block (830) can be evaluated as corresponding to an actual captured solar panel image.

[0132] More specifically, the first block (820) can iteratively optimize the generation parameters so that the synthetic image has characteristics similar to an actual captured image by progressively modifying the texture, boundary information, brightness distribution, local pattern, or spatial correlation of the synthetic image to meet the discrimination criteria of the second block (830).

[0133] At this time, the first block (820) and the second block (830) may each include a first attention block (821) and a second attention block (831) for improving feature extraction and positional accuracy. Specifically, the attention block (e.g., the first attention block and the second attention block, 850) may generate at least one attention map (851, 852, 853) for the input highlighting image (810). For example, the attention block (850) may generate a first attention map (851), a second attention map (852), and a third attention map (853).

[0134] Furthermore, the attention block (850) can generate at least one attention feature map (854) using at least one of the first attention map (851), the second attention map (852), and the third attention map (853).

[0135] Here, an attention block may refer to a computational module for analyzing correlations between multiple locations within an input image or feature map, or for selectively highlighting spatial regions or channel information of high importance. By assigning weights to specific regions within the entire area of ​​the input data, the attention block can perform the function of enabling the AI ​​model to focus on meaningful regions during the feature extraction process, and various attention mechanisms, including spatial attention, channel attention, self-attention, or combinations thereof, may be applied.

[0136] Additionally, an attention map may refer to a two-dimensional or multi-dimensional matrix representing the weight distribution or attention score calculated from the attention block. An attention map is a data structure that numerically represents the importance or contribution of each location in an input image or feature map, and can be generated by reflecting whether the location includes features of a defect region, the amount of texture change, boundary intensity, or semantic importance.

[0137] In addition, an attention feature map can refer to a new type of feature map with reflected weights, obtained by applying the generated attention map to an input feature map or a highlight image.

[0138] Meanwhile, in the present invention, a process of detecting defects in a solar panel from an image can be performed by classifying classes in a Faster-RCNN model based on a synthetic image (S530, see FIG. 5).

[0139] The artificial intelligence model (152) can be trained to classify defects in the solar panel image (10) using at least one of the synthetic image and the enhanced image (810) that the second model (830) has determined to be true. Specifically, the artificial intelligence model (152) can process at least one of the synthetic image and the enhanced image (810) as input to the defect classification block (840).

[0140] Here, the defect classification block (840) may be a block that performs defect classification of the solar panel image (10).

[0141] The artificial intelligence model (152) can perform at least one of generating a bounding box (841) and class classification (842) for at least one of a synthetic image and an enhanced image (810) using a defect classification block (840).

[0142] Here, generating a bounding box may mean detecting areas in an input image where defects are likely to exist, and calculating coordinate information indicating the location and size of said areas using a regression method.

[0143] In addition, performing class classification may mean predicting a class value to determine whether a defect candidate region is normal (clean) or corresponds to a specific defect type (e.g., snow accumulation, contamination, damage, etc.).

[0144] For example, the defect classification block (840) may generate an object candidate region for at least one of the input synthetic image and the enhancement image (810), and normalize the generated object candidate region to generate a bounding box indicating the location of a defect in the solar panel image (10), or perform class classification to classify the type of defect (e.g., snow accumulation). At this time, the algorithm of the artificial intelligence model (152) may be as shown in FIG. 12.

[0145] As illustrated in FIG. 13, the artificial intelligence model (152) can perform defect class classification of the solar panel image (10). For example, the artificial intelligence model (152) can classify the defect class of the solar panel image (10) into at least one of the classes clean, bird drop, snow covered, electric damage, dusty, and breaking.

[0146] Furthermore, the solar panel defect classification system (100) may provide defect classification results. For example, the provided defect classification results may include information related to the classified class (e.g., bird droppings, 30) and at least one of a bounding box indicating the defect area.

[0147] In addition to the examples described above, the present invention may also be used in various devices, systems, and services.

[0148] In one embodiment, the present invention can be used in the operation and maintenance environment of a large-scale ground-mounted solar power plant. In large-scale power complexes, thousands to hundreds of thousands of solar panel modules are placed over an extensive site, and the task of periodically inspecting all modules due to the large number of panels may require significant time and manpower.

[0149] Furthermore, defects such as contamination, cracks, snow accumulation, and damage occurring on the panel surface can only be clearly identified under certain illumination levels or specific weather conditions, making rapid and accurate diagnosis difficult through visual inspection alone. In such operating environments, the present invention can be effectively utilized to automatically analyze the condition of solar panels and detect defects.

[0150] For example, an image acquisition module included in a power plant operation server or field inspection device can periodically receive images of solar panels using fixed CCTV cameras, long-range cameras, tracking monitoring cameras, or drone cameras. For instance, a drone can fly over the installed panels along a planned route, capture high-resolution images, and transmit them to the operation server. The transmitted images can be automatically processed during subsequent analysis, and this capture and transmission structure can contribute to efficiently monitoring the condition of panels throughout the power plant.

[0151] The solar panel image received by the operation server can be processed by the image preprocessing unit (151). The image preprocessing unit can non-linearly transform the pixel coordinates of the received image to emphasize the difference between the normal pattern and the defect pattern, and then perform a gradient-based boundary enhancement operation to generate an enhanced image (810) in which the contours of the defect and texture changes are clearly expressed.

[0152] The artificial intelligence model (152) can generate a synthetic image using the generated enhancement image (810). Specifically, the artificial intelligence model (152) can generate a synthetic image using the first block (820). The first block can generate a synthetic image by receiving a noise vector extracted from the enhancement image as input. The generated synthetic image may include statistical and visual features similar to actual solar panel images, and may be used as a training image that includes rare defect patterns that are not frequently collected in actual power plant environments.

[0153] Furthermore, the second block can receive at least one of the enhanced image and the synthetic image generated by the first block as input and generate a discrimination signal to determine whether it is true or false. The first block and the second block can learn from each other iteratively to continuously adjust the synthetic image generation method so that the synthetic image generated by the first block is determined to be true by the second block. Through such an iterative structure, a high-quality synthetic image that mimics an actual defect pattern can be generated.

[0154] In the learning phase, the artificial intelligence model (152) can receive at least one of the synthetic image generated in the first block and the enhancement image (810) generated by the image preprocessing unit as input and perform learning for defect detection. By including the synthetic image in the learning data, the present invention can significantly alleviate the problem of deviation in actual captured images or the lack of defect data.

[0155] Meanwhile, after training is completed, the enhanced image (810) of the actual solar panel image can be processed as input to the artificial intelligence model to classify whether there is a defect and the type of defect. The defect classification block (840) can generate an object candidate region for at least one of the enhanced images, and calculate the location and size of the defect by normalizing the candidate region and performing bounding box regression.

[0156] In addition, by performing a defect class classification operation on candidate defect regions, the defect can be classified into at least one of normal (clean), contaminated (dusty), snow covered, bird drop, or crack. Through this, the present invention can prevent a decrease in power generation efficiency and efficiently allocate maintenance resources by detecting defects early, which can contribute to reducing overall operating costs of the power plant and improving long-term profitability.

[0157] In one embodiment, the present invention can be used in the operating environment of a drone-based solar panel inspection system. As the number of cases in which solar power plants are installed on sites with poor accessibility, such as mountainous, waterfront, and island regions, has recently increased, video inspection using drones may be essential in areas where ground inspection is difficult.

[0158] While drones can capture large areas in a short period, image quality can be unstable due to environmental factors such as shooting angle, altitude changes, and variations in atmospheric pressure and wind speed. Furthermore, since contaminants, microcracks, snow accumulation, and bird droppings on the panel surface may not be clearly visible depending on shooting conditions, it can be difficult to accurately detect defects through simple image analysis alone. In such environments, the present invention can be effectively utilized to automatically analyze defect characteristics included in drone-captured images and classify defects.

[0159] For example, a drone can fly over a solar power plant to capture high-resolution video or still images and transmit them to a ground control system or a cloud server. Although images captured by the drone may contain distortions due to illumination, shadows, and shooting angles, such images can also be reliably analyzed through the processing structure of the present invention. The image acquisition module can contribute to real-time or near-real-time monitoring of the solar panel status by continuously receiving video data from the drone and transmitting it to a subsequent analysis stage.

[0160] The drone-captured image transmitted to the server can be processed by an image preprocessing unit (151). The image preprocessing unit can emphasize the difference between the pattern of the defect area and the normal area by non-linearly transforming the pixel coordinates of the input image to reduce the effect of geometric distortion or changes in illumination caused by variations in shooting conditions.

[0161] Next, by performing a gradient-based boundary enhancement operation, an enhanced image (810) can be generated in which defect features such as cracks, traces of contamination, and snow boundaries are highlighted. This enhanced image can more clearly represent defects despite the distortion and shaking characteristic of drone photography.

[0162] The artificial intelligence model (152) can generate a synthetic image using the generated enhancement image (810). Specifically, the artificial intelligence model (152) can generate a synthetic image using the first block (820).

[0163] The first block receives a noise vector extracted from a highlight image as input and can generate a synthetic image having visual and statistical patterns similar to the characteristics of an actual drone-captured image. Since drone-captured images have the characteristic that it is difficult to capture defects multiple times from the same location and angle, the synthetic image generated by the first block can be usefully utilized as defect learning data that reflects various shooting conditions.

[0164] Furthermore, the second block (830) can receive at least one of the enhanced image and the synthetic image generated by the first block and generate a determination signal to determine whether it is true or false. The first block and the second block can be repeatedly trained to continuously improve the synthesis generation method so that the synthetic image generated by the first block is determined to be true by the second block. Through this iterative structure, a high-quality synthetic image similar to the defect characteristics that appear in actual drone photography can be generated.

[0165] In the learning phase, the artificial intelligence model (152) can receive at least one of the synthetic image generated in the first block and the enhancement image (810) generated by the image preprocessing unit as input to perform learning for defect detection. Since the shooting conditions are very diverse due to the characteristics of drone-captured images, the use of synthetic images can contribute to improving the generalization performance of the defect detection model. In particular, since it is difficult to obtain defect data in actual drone shooting environments, the synthetic image generation structure of the present invention can effectively compensate for the problem of insufficient defect data.

[0166] Meanwhile, after training is completed, the enhanced image (810) of the actual drone-captured image can be processed as input to the artificial intelligence model to classify whether there is a defect and the type of defect. The defect classification block (840) can generate an object candidate region for at least one of the enhanced images, and calculate the location and size of the defect by normalizing the candidate region and performing bounding box regression.

[0167] In addition, by performing defect class classification on candidate defect areas, the defect can be classified into at least one of normal (clean), contaminated (dusty), snow covered, bird drop, or crack. Through this, the present invention can detect defects early even in a drone-based inspection environment to optimize maintenance plans, and can contribute to improved maintenance efficiency and reduced operating costs.

[0168] In one embodiment, the present invention can be used in the operating environment of an inspection system for automating quality inspection in a solar panel manufacturing process. Solar panels are manufactured through processes such as cell assembly, wiring attachment, encapsulation, lamination, and frame installation, and at each stage, defects such as microcracks, electrode defects, silver wire breakage, foreign substance contamination, and lamination defects may occur. These manufacturing defects are difficult to identify quickly through visual inspection or random sample inspection methods, and if defects are missed, they can cause long-term performance degradation or safety issues. Therefore, the present invention, which can detect defects by automatically analyzing real-time video data of a manufacturing line, can be usefully utilized to improve manufacturing quality.

[0169] For example, a high-resolution industrial camera, a line-scan camera, or a near-infrared (NIR) camera may be installed above the manufacturing process line. The camera continuously captures the panels being produced, and the captured images may be transmitted to a manufacturing inspection server or a quality control system. Although the manufacturing process images may include elements specific to the shooting environment, such as lighting reflection, light scattering from the lamination film, and regularity of the cell pattern, they can be reliably processed through the analysis structure of the present invention.

[0170] The manufacturing process image transmitted to the server can be processed by an image preprocessing unit (151). The image preprocessing unit can emphasize the pattern difference between the defective area and the normal area by non-linearly transforming the pixel coordinates of the input image to reduce the influence of surface reflection or non-uniformity of illumination occurring during the panel manufacturing stage. Subsequently, by performing a gradient-based boundary enhancement operation, an enhanced image (810) can be generated that highlights the structural features of manufacturing defects such as micro-cracks, wire breakage, foreign substance boundaries, and lamination lifting. This enhanced image can more clearly show defects that are difficult to distinguish with the naked eye due to the characteristics of the manufacturing process.

[0171] The artificial intelligence model (152) can generate a synthetic image using the generated enhancement image (810). Specifically, the artificial intelligence model (152) can generate a synthetic image using the first block (820). The first block receives a noise vector extracted from the enhancement image and can generate a synthetic image that reflects the characteristics of defects occurring in the manufacturing process. Since certain defects occur very rarely in the manufacturing process, the synthetic image generated by the first block can be used as training data that includes various types of defects that are difficult to collect from an actual production line.

[0172] Furthermore, the second block (830) can receive at least one of the enhanced image and the synthetic image generated by the first block and generate a discrimination signal to determine whether it is true or false. The first block and the second block can be repeatedly trained to continuously optimize the generation method of the first block in a direction in which the second block determines it to be true. Through this iterative structure, a high-quality synthetic image that precisely simulates defect characteristics appearing in an actual manufacturing process can be generated.

[0173] In the learning phase, the artificial intelligence model (152) can receive at least one of the synthetic image generated in the first block and the enhancement image (810) generated by the image preprocessing unit as input and perform learning for defect detection. In manufacturing processes, it is difficult to obtain sufficient defect data because real-time production volume is high and the frequency of defect occurrence is low. The present invention can support the model in broadly learning various types of defects that may occur in manufacturing processes by including synthetic images in the learning data.

[0174] Meanwhile, after training is completed, a highlight image (810) of a solar panel image taken during the actual manufacturing process can be processed as input to an artificial intelligence model to classify whether there is a defect and the type of defect. The defect classification block (840) can generate an object candidate region for at least one of the highlight images, and calculate the location and size of a defect that occurred during the manufacturing process by normalizing the candidate region and performing bounding box regression. In addition, a class classification operation can be performed on the defect candidate region to classify the defect into at least one of normal (clean), microcrack, electrode defect, dust / foreign object, or lamination delamination. Through this, the present invention can contribute to reducing the defect rate and improving quality stability by detecting defects early in real time during the manufacturing process.

[0175] As described above, the artificial intelligence-based solar panel defect classification method and system according to the present invention can detect and classify defects in solar panels using an artificial intelligence model. Through this, even minute defects or atypical contamination elements present within solar panel images can be precisely recognized, and problems that may lead to a decrease in power generation efficiency can be detected early by systematically analyzing the shape, location, and scale of the defects.

[0176] Furthermore, the artificial intelligence-based solar panel defect classification method and system according to the present invention can stably train a defect classification model by generating synthetic training data containing various defect patterns even in environments where solar panel images are scarce. Through this, the problem of data bias caused by a lack of actual defect images is reduced, and by sufficiently learning in advance the defect characteristics that may exist in images captured under various environments, angles, and illumination levels, high accuracy and consistent performance can be maintained even in actual operating environments.

[0177] Furthermore, the artificial intelligence-based solar panel defect classification method and system according to the present invention can automate the condition diagnosis process of solar panels based on images highlighting defect features and a trained artificial intelligence model. Through this, even in environments where the number of panels to be inspected is very large, such as large-scale solar power plants, where conventional manual inspection would require excessive time and cost, it is possible to quickly identify whether the panels are damaged and make immediate maintenance decisions.

[0178] Meanwhile, the present invention described above can be implemented as a program that is executed by one or more processes on a computer and can be stored on a computer-readable medium (or recording medium).

[0179] Furthermore, the present invention described above can be implemented as computer-readable code or instructions on a medium on which a program is recorded. That is, the present invention can be provided in the form of a program.

[0180] Meanwhile, computer-readable media include all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0181] Furthermore, the computer-readable medium may be a server or cloud storage that includes a storage and is accessible to an electronic device via communication. In this case, the computer may download the program according to the present invention from the server or cloud storage via wired or wireless communication.

[0182] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, namely a CPU (Central Processing Unit), and no special limitations are placed on its type.

[0183] Meanwhile, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.

Claims

Claim 1 A method for classifying defects in an artificial intelligence-based solar panel, comprising: a step of extracting pixel locations from an image including a solar panel and generating a filtered image by distorting at least a portion of the pixel locations; a step of generating an enhanced image in which boundary information is emphasized by calculating the gradient magnitude of the filtered image and marking the corresponding pixel as a boundary as the gradient magnitude exceeds a preset threshold; a step of generating a synthetic image that mimics a defect image and a normal image by inputting the enhanced image into a GAN model; and a step of classifying classes in a Faster-RCNN model based on the synthetic image to detect defects in the solar panel from the image, wherein in the step of generating the filtered image, at least a portion of the pixel locations are distorted to non-linearly change the spatial structure of the image by rearranging or transforming the coordinates of each pixel constituting the image according to preset operation rules, and in the step of generating the synthetic image, at least one noise vector is generated by reflecting at least a portion of the characteristics of the enhanced image and the synthetic image is generated using the generated noise vector. Claim 2 An artificial intelligence-based solar panel defect classification method according to claim 1, wherein the GAN model comprises a first block for generating the synthetic image and a second block for determining whether the generated synthetic image is true or false. Claim 3 An artificial intelligence-based solar panel defect classification method according to claim 2, wherein in the step of generating the synthetic image, a synthetic image that can be judged as true in the second block is generated in the first block. Claim 4 An artificial intelligence-based solar panel defect classification method according to claim 3, wherein in the step of detecting the defect, the defect of the solar panel is detected from the image by classifying the Faster-RCNN model based on the synthetic image determined to be true in the second block. Claim 5 An artificial intelligence-based solar panel defect classification method according to claim 3, wherein the first block and the second block generate the synthetic image for focusing on cracks by performing self-attention. Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 In an electronic device, a memory for storing instructions; An electronic device comprising at least one processor electrically connected to the memory, wherein when the instructions are executed by the at least one processor, the at least one processor: extracts pixel locations from an image including a solar panel, generates a filtered image by distorting at least a portion of the pixel locations, calculates the gradient magnitude of the filtered image, and generates an enhanced image in which boundary information is emphasized by marking the corresponding pixel as a boundary as the gradient magnitude exceeds a preset threshold, inputs the enhanced image into a GAN model to generate a synthetic image that mimics a defective image and a normal image, and classifies classes in a Faster-RCNN model based on the synthetic image to detect defects in the solar panel from the image, wherein generating the filtered image involves distorting at least a portion of the pixel locations to non-linearly change the spatial structure of the image by rearranging or transforming the coordinates of each pixel constituting the image according to preset operation rules, and generating the synthetic image involves generating at least one noise vector by reflecting at least a portion of the characteristics of the enhanced image and generating the synthetic image using the generated noise vector. Claim 10 A program that is executed by one or more processes in an electronic device and stored on a computer-readable recording medium, wherein the program comprises the steps of: extracting pixel locations from an image including a solar panel and generating a filtered image by distorting at least a portion of the pixel locations; calculating the gradient magnitude of the filtered image and generating an enhanced image in which boundary information is emphasized by marking the corresponding pixel as a boundary as the gradient magnitude exceeds a preset threshold; and inputting the enhanced image into a GAN model to generate a synthetic image that mimics a defective image and a normal image. A program stored on a computer-readable recording medium, comprising instructions for performing a step of classifying a class in a Faster-RCNN model based on the composite image and detecting defects in the solar panel from the image, wherein in the step of generating the filtered image, at least some of the pixel locations are distorted to non-linearly change the spatial structure of the image by rearranging or transforming the coordinates of each pixel constituting the image according to a preset operation rule, and in the step of generating the composite image, at least one noise vector is generated by reflecting at least some of the characteristics of the enhanced image, and the composite image is generated using the generated noise vector.

Images