A method and system for substation image detection with region-of-interest self-adjustment

Abstract

A method and system for substation image detection with self-adjusting region of interest (ROI) is disclosed. The method comprises the following steps: Step 1, extracting a target template image from a substation image and dividing the substation scene based on the target template image; Step 2, calibrating the image to be tested against the closest target template image and determining the scene of the image to be tested; Step 3, selecting the region of interest (ROI) in the image to be tested based on the scene, performing YOLO target detection on the ROI, and outputting an anomaly alarm when an anomaly is detected. This invention is ingeniously conceived and yields accurate results, providing targeted ROI extraction for the operating scene and abnormal state of the target, thereby achieving precise target anomaly alarms.

Description

Technical Field

[0001] This invention relates to the field of power systems, and more specifically, to a method and system for detecting images of substations with region-of-interest self-adjustment. Background Technology

[0002] Currently, power systems typically utilize unattended image and video monitoring methods for intelligent inspection and fault detection. However, for substations, which involve a large number of different substation devices, the detection process requires customized anomaly monitoring and alarms based on the different operating scenarios of each device.

[0003] However, existing image processing methods for equipment operation images collected in power systems are too simplistic. In complex substation environments, with varying equipment operation scenarios and image acquisition methods, there is still no method capable of accurately detecting targets within images of the monitored equipment. Furthermore, due to environmental factors such as light, shadows, and background vegetation in the original acquired images, even if the target can be accurately detected, existing technologies struggle to provide effective anomaly monitoring for different targets.

[0004] To address the aforementioned problems, a new method and system for substation image detection is urgently needed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a substation image detection method and system with self-adjusting region of interest. The method extracts a target template image, pre-divides the scene, calibrates and allocates the scene of the image to be tested, and automatically adjusts the region of interest according to the scene of the image to be tested, thereby effectively achieving target detection and anomaly alarm.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of this invention relates to a substation image detection method with self-adjusting region of interest (ROI). The method includes the following steps: Step 1, extracting a target template image from a substation image and dividing the substation scene based on the target template image; Step 2, calibrating the image to be tested with the closest target template image and determining the scene of the image to be tested; Step 3, selecting the region of interest (ROI) in the image to be tested based on the scene of the image to be tested, performing YOLO target detection on the ROI, and outputting an anomaly alarm when an anomaly is detected.

[0008] Preferably, the target template image includes one or more of the following: daytime image, nighttime image, visible light image, and infrared image of the scene target in the standard coordinate direction; wherein, the scene target includes instruments, transformer breathers, insulators, cabinets, boxes, substation personnel, and outdoor structures.

[0009] Preferably, the substation scene division is generated based on different scene targets in the substation.

[0010] Preferably, the method for calibrating the image to be tested and the target template image includes: Step 2.1, extracting image feature points from the image to be tested and the target template image respectively; Step 2.2, matching the image feature points of the image to be tested and the target template image using the nearest neighbor algorithm, and generating a feature point transition matrix; Step 2.3, using a random sampling algorithm to iteratively filter the feature point transition matrix for incorrect matching until a filtered transition matrix with an error probability less than a set threshold is obtained; Step 2.4, performing an affine transformation on the image to be tested based on the filtered transition matrix to achieve calibration of the image to be tested.

[0011] Preferably, when the target template image is a handle or pressure plate in a substation, the current scene of the image to be tested is identified as a cabinet scene; when the target template image is a box door in a substation, the current scene of the image to be tested is identified as a box door scene; when the target template image is personnel in a substation, the current scene of the image to be tested is identified as a personnel intrusion scene; when the target template image is an instrument, breathing apparatus, or insulator in a substation, the proportion of the area occupied by the target in the image to be tested is determined, and when the proportion is greater than 10%, the current scene of the image to be tested is identified as an instrument scene, a breathing apparatus scene, and an insulator scene, respectively; when the target template image is an insulator and the proportion of the area occupied by the target in the image to be tested is less than 1%, the current scene of the image to be tested is identified as an outdoor elevated scene.

[0012] Preferably, the region of interest is selected for the current scene; wherein, if the current scene is an instrument scene, a transformer breather scene, an insulator scene, a box door scene, or a personnel intrusion scene, the region of interest is the area where the target is located; if the current scene is a cabinet scene or an outdoor structure scene, the region of interest is obtained based on the area where the colored pixels in the image are located.

[0013] Preferably, the image to be tested and the target template image are subjected to difference operation and binarization; the binarized image of interest is divided into regions, and regions with an area smaller than a set threshold and regions located at the edge of the image are discarded; the remaining regions are combined and set as the abnormal alarm image to be output.

[0014] Preferably, the abnormal alarm image is determined based on the abnormal target template image; when it is determined to be abnormal, an abnormal alarm containing the abnormal alarm image is output.

[0015] Preferably, the region block with an area smaller than the set threshold is a region block whose area is less than 5% of the area of ​​the image to be tested.

[0016] A second aspect of the present invention relates to a substation image detection system with self-adjusting region of interest (ROI). The system implements the steps of the substation image detection method with self-adjusting ROI as described in the first aspect of the present invention. The system includes an extraction module, a calibration module, and a detection module. The extraction module extracts a target template image from a substation image and performs scene segmentation of the substation based on the target template image. The calibration module calibrates the image to be tested against the target template image and determines the scene of the image to be tested. The detection module selects the region of interest (ROI) in the image to be tested based on the scene of the image to be tested, performs YOLO target detection on the ROI, and outputs an anomaly alarm when an anomaly is detected.

[0017] The beneficial effects of this invention are that, compared with the prior art, the substation image detection method and system with self-adjusting region of interest (ROI) can extract target template images, pre-divide scenes, calibrate and allocate scenes for the image under test, and automatically adjust the ROI according to the scene of the image under test, thereby effectively achieving target detection and anomaly alarm. This invention is ingeniously conceived and yields accurate results, providing targeted ROI extraction for the operating scene and abnormal state of the target, thus achieving precise target anomaly alarm.

[0018] The beneficial effects of the present invention also include:

[0019] 1. The present invention first constructs a target template image, which includes various image types such as daytime, nighttime, visible light, and infrared. This enables the method to effectively obtain comprehensive feature information about the target under different environmental influencing factors, thereby making the target detection results more accurate.

[0020] 2. This invention takes into account the potential for target position shifts in images due to camera shake during image cropping in dynamic videos. Therefore, the images are pre-calibrated. This invention uses nearest neighbor and random sampling algorithms to match and filter feature points, making the image calibration results more accurate, ensuring the reliability of target detection, and preventing target detection failures caused by target shifts in the image.

[0021] 3. This invention achieves image scene segmentation based on the different targets of anomaly detection in substations. This scene segmentation allows the method to extract regions of interest (ROIs) from images using different methods depending on the scene in which the target is located. Because the ROI extraction method varies depending on the scene, the method can accurately and comprehensively acquire key areas in the image regardless of the scene, and achieve precise target detection and anomaly determination based on the ROI.

[0022] 4. In view of the problem that the edge information of the image and the edge information of the target template image are greatly different due to the calibration process, the present invention filters out these image features from the region of interest, thereby further ensuring the accurate extraction of the region of interest. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the steps of a substation image detection method with region-of-interest self-adjustment according to the present invention.

[0024] Figure 2 This is a schematic diagram of the module structure of a substation image detection system with region of interest self-adjustment according to the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments not described in this invention obtained by those skilled in the art based on the embodiments described in this invention without creative effort should fall within the protection scope of this invention.

[0026] Figure 1 This is a schematic diagram illustrating the steps of a substation image detection method with region-of-interest self-adjustment according to the present invention. Figure 1 As shown, in a first aspect, the present invention relates to a substation image detection method with region of interest self-adjustment, the method comprising steps 1 to 3.

[0027] Step 1: Extract the target template image from the substation image, and divide the substation scene based on the target template image.

[0028] The purpose of this invention is to perform general power equipment status monitoring in substation scenarios by comparing images of the same scene taken at different times, such as the opening and closing status of transformer box doors, disconnectors, instrument damage, or personnel intrusion. However, image processing is affected by environmental factors such as light, shadows, and background vegetation, making it difficult to propose a universal image processing method for image change monitoring. Therefore, this invention proposes an image processing model that first divides the substation into different scenarios, then extracts regions of interest and detects areas of image change based on the corresponding scenarios.

[0029] To achieve image detection in different scenarios, this application first selects some standard images from substation images as target template images. It should be noted that substation images can be screenshots of original substation images or videos acquired using existing methods. Since the monitoring equipment in substations is diverse, and the types of images acquired vary, this invention performs extensive screening and analysis on these images, extracting a subset of standard images.

[0030] These standard images meet certain basic characteristics, such as the target being very clear. During image acquisition, the sensor or camera did not shift or shake. Furthermore, considering various image acquisition environments, this invention can simultaneously include images of the target under different environments within the target template image, thus generating a target template image set. This image set includes not only targets under different environments but also target images of the target device in normal operation and various typical abnormal operating states.

[0031] Preferably, the target template image includes one or more of the following: daytime image, nighttime image, visible light image, and infrared image of the scene target in the standard coordinate direction; wherein, the scene target includes instruments, transformer breathers, insulators, cabinets, boxes, substation personnel, and outdoor structures.

[0032] Considering that the purpose of image detection in this invention is to target key equipment in substations that are prone to failure and require close monitoring, or not to target a single device but to common substation scenarios that are prone to danger, this invention divides the scene targets into the seven types mentioned above.

[0033] Preferably, the substation scene division is generated based on different scene targets in the substation.

[0034] Considering that scene segmentation of the images to be tested is required in this application, all target template images extracted in this invention should have corresponding tags. For example, the tags can indicate what device or environment the image was acquired with, namely, daytime, nighttime, visible light, and infrared images as mentioned above. Furthermore, the tags can also indicate which specific scene the target template image is for. In this way, tags can achieve scene segmentation of the target template image. Using image matching, the corresponding scene can be further assigned to the image to be tested.

[0035] Step 2: Calibrate the image to be tested against the target template image and determine the scene of the image to be tested.

[0036] It is understood that, in order to perform real-time substation operation anomaly detection in this invention, the substation images acquired in real time are usually input into the image processing model in this invention as the images to be tested, and abnormal targets are detected and alarms are issued.

[0037] Therefore, the image to be tested in this invention can be a raw substation image acquired in real time, while the target template image can be pre-acquired according to certain rules before image detection. In this invention, such standard images can be obtained manually or through automatic computer screening. This invention does not impose strict limitations on the acquisition process of the target template image; various acquisition methods in the prior art are acceptable.

[0038] This invention compares the features of the image to be tested with various target template images in the image set, and selects the closest target template image as the standard for calibration of the image to be tested. In this invention, an improved image calibration method can also be used to achieve calibration between the image to be tested and the closest target template image.

[0039] Preferably, the method for calibrating the image to be tested and the target template image includes: Step 2.1, extracting image feature points from the image to be tested and the target template image respectively; Step 2.2, matching the image feature points of the image to be tested and the target template image using the nearest neighbor algorithm, and generating a feature point transition matrix; Step 2.3, using a random sampling algorithm to iteratively filter the feature point transition matrix for incorrect matching until a filtered transition matrix with an error probability less than a set threshold is obtained; Step 2.4, performing an affine transformation on the image to be tested based on the filtered transition matrix to achieve calibration of the image to be tested.

[0040] It is understood that the method in this invention supports the extraction and matching of image feature points to calibrate the image under test. Specifically, this invention can first extract feature point information from two images. In one embodiment of this invention, the SIFT (Scale Invariant Feature Transform) method can be used to perform image processing. Of course, this invention does not exclude the use of other methods to obtain image feature points. After obtaining the feature points of the two images, the nearest neighbor algorithm can be used to match the feature points between the two images. This method can restrict the feature points of the second nearest neighbors, greatly improving the accuracy of feature point matching.

[0041] After obtaining the matching relationships between all feature points using the nearest neighbor algorithm, this invention can employ a random sampling algorithm to filter out incorrectly matched feature points. The random sampling algorithm iteratively and repeatedly filters out incorrectly matched points, and after each iteration, the error probability is calculated. If the error probability is less than a set threshold, the iteration is considered complete. The matching results at this point generate a filtering transition matrix, and based on this matrix, the coordinates of the image under test are sequentially converted to coordinates consistent with the target template image, thereby completing the calibration of the image under test.

[0042] Preferably, when the target template image is a handle or pressure plate in a substation, the current scene of the image to be tested is identified as a cabinet scene; when the target template image is a box door in a substation, the current scene of the image to be tested is identified as a box door scene; when the target template image is personnel in a substation, the current scene of the image to be tested is identified as a personnel intrusion scene; when the target template image is an instrument, breathing apparatus, or insulator in a substation, the proportion of the area occupied by the target in the image to be tested is determined, and when the proportion is greater than 10%, the current scene of the image to be tested is identified as an instrument scene, a breathing apparatus scene, and an insulator scene, respectively; when the target template image is an insulator and the proportion of the area occupied by the target in the image to be tested is less than 1%, the current scene of the image to be tested is identified as an outdoor elevated scene.

[0043] It is understandable that after the target template detection and the test image calibration are completed, scene allocation can be performed based on the feature properties in the calibrated image. It is also understandable that during the calibration process, the closest target template image obtained for the test image has a pre-assigned label. Based on the information recorded in the label, this invention can determine the specific scene target present in the test image, as well as the environment and image type acquired by the test image.

[0044] Therefore, this invention can simultaneously classify the scene of the image under test during the calibration process. As described above, if the image under test contains handles, pressure plates, doors, and people, it can be identified as the corresponding scene image based on the targets in the image.

[0045] If the target in the image to be tested is an instrument or a respirator, the method of this invention needs to further determine the area proportion occupied by the target in the image. Since instruments and respirators may be captured by substation monitoring devices from different angles and distances, this determination of the area proportion can quickly filter out a large number of invalid images and allow for the analysis of the equipment status or substation safety in the current scenario for the useful images to be tested.

[0046] Furthermore, this invention not only collects data on individual substation equipment but also on the overall substation scene, such as outdoor elevated structures. In this invention, the specific shapes of insulators in outdoor elevated structures can be used to distinguish whether a scene is a useful outdoor image for anomaly identification and alarm purposes. Therefore, in this invention, if an insulator is detected in an image, the proportion of the insulator in the image is used to determine the scene type. If the proportion is greater than 10%, it can be identified as an insulator scene; if the proportion is less than 1%, it can be identified as an outdoor elevated structure scene.

[0047] Step 3: Based on the scene of the image to be tested, select the region of interest in the image to be tested, and perform YOLO target detection on the region of interest. When an abnormality is detected, output an abnormality alarm.

[0048] The reason for classifying and identifying different scenarios in the images under test in this invention is to select regions of interest (ROIs) in the images using different methods based on the different attributes of the images under test. These selected ROIs can not only be used to achieve specific target detection, such as whether the target equipment is operating normally or in an abnormal state, or whether there is a danger to personnel in the target environment; on the other hand, the ROIs can also be used as extracted abnormal images to participate in further fault type assessment, or as part of abnormal alarm information directly output to the substation operation and maintenance personnel.

[0049] Preferably, the region of interest is selected for the current scene; where the current scene is an instrument scene, a transformer breather scene, an insulator scene, a box door scene, or a personnel intrusion scene, the region of interest is the area where the target is located; if the current scene is a cabinet scene or an outdoor structure scene, the region of interest is obtained based on the area where the colored pixels in the image are located.

[0050] Understandably, the method for selecting the region of interest (ROI) varies depending on the current scenario. When the current scenario involves instruments, transformer breathers, insulators, enclosure doors, or personnel intrusion, the range of the target, i.e., the ROI, can be directly determined using existing target detection methods. In one embodiment of this invention, the YOLO (You Only Look Once) model can be directly used to achieve real-time, fast target detection. However, when the current scenario involves a cabinet or outdoor structure, the image is more susceptible to interference from external factors such as light and weather, making it difficult to accurately locate the actual target detection area. Therefore, this invention adds an additional step before using the YOLO model: extracting the ROI based on the area where the colored pixels in the image are located.

[0051] Preferably, the image to be tested and the target template image are subjected to difference operation and binarization; the binarized image of interest is divided into regions, and regions with an area smaller than a set threshold and regions located at the edge of the image are discarded; the remaining regions are combined and set as the abnormal alarm image to be output.

[0052] For images depicting instruments, respirators, insulators, enclosure doors, and personnel intrusion, this invention can employ the aforementioned method to perform image difference operations, obtaining the difference between the image to be tested and the target template. After obtaining the difference image, the image is binarized to obtain a binarized image. Since the pixel values ​​of the difference image are all 0 for parts of the target or background that are the same or similar, the obtained binarized image may contain multiple dissimilar, independent, and closed regions. Therefore, based on the size of each region, this invention can perform further filtering. If the area of ​​a closed region is too small, or if it is located at the edge of the image that was truncated or padded during image calibration, this closed region will be filtered out, thus ensuring the accuracy of the binarized image. Subsequently, the remaining regions are unioned and used as the output of the model of this invention.

[0053] Preferably, the region block with an area smaller than the set threshold is a region block whose area is less than 5% of the area of ​​the image to be tested.

[0054] Regarding the specific function of the model's output, as mentioned earlier, it can serve as direct alarm information or as an intermediate step in target detection, thereby adapting it to the typical operating state of the corresponding target, etc. This part will be further explained later.

[0055] In scenarios where the current environment is a cabinet-style or outdoor structure, the process of obtaining the region of interest will be relatively more complex.

[0056] In this embodiment of the invention, considering that screen cabinet scenarios often have shadow interference, and shadows are gray, while pressure plate components are red, green, and yellow, and handle components are black, the region of interest can be extracted based on the colors of key components such as pressure plates and handles, thereby removing shadow interference.

[0057] For example, in one embodiment of the present invention, if the image to be tested is detected to be in a cabinet scene, then the difference between the target template image and the image can be obtained and binarization can be performed. For ordinary images, the binarization process is simply based on the pixel difference between the two images. However, for the images in this application, the binarization process needs to consider the color of the original pixels in the image to be tested in advance. For example, in one embodiment of the present invention, the original image is an HSV format image. If the pixel in the HSV image is red, H: (0, 10), U: (170, 180), S: (43, 255), V: (46, 255), then the pixel in the binary image is set to 255; if the pixel in the HSV image is green, H: (35, 77), S:

[0058] If the pixel in the HSV image is yellow, H: (26, 77), S: (34, 255), V: (46, 255), then the pixel in the binary image is set to 255; if the pixel in the HSV image is black, H: (0, 180), S: (0, 255), V:

[0059] If (0, 46), then the pixel in the binary image is set to 255. In this case, the method not only considers the pixel difference between the two images, but also directly determines the value of these pixels in the binarized image based on some special colors in the image to be tested.

[0060] In another embodiment of the present invention, instead of calculating the difference between the two images, a rectangular box that can contain all colored pixels can be selected as the region of interest based on the location of the colored pixels.

[0061] For outdoor architectural scenes, the colors are often affected by weather and lighting from different angles, so outdoor architecture is usually depicted in black, white, and gray. Additionally, trees in outdoor scenes can appear yellow, green, or cyan depending on the season; therefore, aside from these colors, other colors in the image can be considered regions of interest, including red, blue, and purple. These areas may contain intrusive objects or other variations.

[0062] Based on the above approach, in one embodiment of the present invention, if the pixels in the HSV image are red, H:

[0063] Given a binary image with values ​​(0, 10)U(170, 180), S: (43, 255), V: (46, 255), the corresponding pixel value is set to 255. If the pixel in the HSV image is blue, with H: (100, 124), S: (43, 255), V: (46, 255), the corresponding pixel value is set to 255. If the pixel in the HSV image is purple, with H: (125, 155), S: (43, 255), V: (46, 255), the corresponding pixel value is set to 255. All other pixel values ​​in the binary image are set to 0. After obtaining the binarized image in this way, the smallest bounding box including all pixels with a value of 255 can be considered the region of interest for the current scene.

[0064] Since some images may represent more general scenarios, this invention does not exclude the possibility of further early warning analysis for such images. Therefore, if it is neither a cabinet scenario nor an outdoor structure scenario, the entire image is taken as the region of interest, according to general scenarios.

[0065] Preferably, the abnormal alarm image is determined based on the abnormal target template image; when it is determined to be abnormal, an abnormal alarm containing the abnormal alarm image is output.

[0066] This invention can pre-define abnormal target template images. For example, a fault database can be pre-proposed, which may include typical image sets of various target devices and scenarios when abnormalities occur. For instance, the fault database may include sample images of normal markings, blurred dials, and damaged dials, as well as samples of normal respirators, damaged respirators, and discolored respirators. Similarly, for insulator scenarios, samples of normal insulators and broken insulators can be provided; for box door scenarios, typical images of normal and abnormal box door closure can be provided. In personnel intrusion scenarios, different samples and sample features are provided, such as personnel not wearing safety helmets, personnel not wearing work clothes, normal safety helmets, and normal work clothes.

[0067] With these pre-set parameters, in different scenarios, after the image under test is registered and the region of interest is acquired, it can be matched with typical faults in the fault database using image matching. If a fault occurs, an alarm message will be issued immediately.

[0068] This invention integrates deep learning and traditional image processing techniques, which can effectively identify changes in substation scenes and has practical application feasibility. Using this algorithm can effectively reduce the false negative rate and the false positive rate, and improve efficiency and accuracy.

[0069] Figure 2 This is a schematic diagram of the module structure of a substation image detection system with region-of-interest self-adjustment according to the present invention. Figure 2As shown, in a second aspect, the present invention relates to a substation image detection system with self-adjusting region of interest (ROI). The system implements the steps of the substation image detection method with self-adjusting ROI in the first aspect of the present invention. The system includes an extraction module, a calibration module, and a detection module. The extraction module extracts a target template image from a substation image and performs scene segmentation of the substation based on the target template image. The calibration module calibrates the image to be tested against the target template image and determines the scene of the image to be tested. The detection module selects the region of interest (ROI) in the image to be tested based on the scene of the image to be tested, performs YOLO target detection on the ROI, and outputs an anomaly alarm when an anomaly is detected.

[0070] It is understood that the substation image detection system, in order to achieve the various functions in the methods provided in the embodiments of this application, includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, based on the algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0071] This application embodiment can divide the substation image detection system into functional modules based on the above method example. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0072] The substation image detection system can be implemented by one or more server devices connected via communication. Each device includes at least one processor, a bus system, and at least one communication interface. The processor can be a Central Processing Unit (CPU), or it can be replaced by a Field Programmable Gate Array (FPGA), an Application-Specific Integrated Circuit (ASIC), or other hardware. Alternatively, an FPGA or other hardware can be used in conjunction with a CPU as the processor.

[0073] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a bus. The memory can also be integrated with the processor.

[0074] The hard drive can be a mechanical hard drive or a solid-state drive (SSD), etc. The interface card can be a host bus adapter (HBA), a redundant array of independent disks (RID), an expander card, or a network interface controller (NIC), etc., and this embodiment of the invention is not limited to any particular type. The interface card in the hard drive module communicates with the hard drive. The storage node communicates with the interface card of the hard drive module to access the hard drive in the hard drive module.

[0075] The hard drive interface can be Serial Attached Small Computer System Interface (SAS), Serial Advanced Technology Attachment (SATA), or Peripheral Component Interconnect Express (PCIe), etc.

[0076] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0077] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0078] The beneficial effects of this invention are that, compared with the prior art, the substation image detection method and system with self-adjusting region of interest (ROI) can extract target template images, pre-divide scenes, calibrate and allocate scenes for the image under test, and automatically adjust the ROI according to the scene of the image under test, thereby effectively achieving target detection and anomaly alarm. This invention is ingeniously conceived and yields accurate results, providing targeted ROI extraction for the operating scene and abnormal state of the target, thus achieving precise target anomaly alarm.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A substation image detection method with region-of-interest self-adjustment, characterized in that, The method includes the following steps: Step 1: Extract the target template image from the substation image, and perform scene division of the substation based on the target template image; The target template image includes one or more of the following: daytime image, nighttime image, visible light image, and infrared image of the scene target in the standard coordinate direction; wherein, the scene target includes instruments, transformer breathers, insulators, cabinets, enclosure doors, substation personnel, and outdoor structures; Step 2: Calibrate the image to be tested against the closest target template image, and determine the scene of the image to be tested; Step 2.1: Extract image feature points from the image to be tested and the target template image respectively; Step 2.2: The nearest neighbor algorithm is used to match the image feature points of the image to be tested and the target template image, and a feature point transition matrix is ​​generated; Step 2.3: Use a random sampling algorithm to iteratively filter the feature point transition matrix for incorrect matching until a filtered transition matrix with an error probability less than a set threshold is obtained. Step 2.4: Perform an affine transformation on the image to be tested based on the screening transition matrix to calibrate the image to be tested; Step 3: Based on the scene of the image to be tested, select the region of interest in the image to be tested, and perform YOLO target detection on the region of interest. When an abnormality is detected, output an abnormality alarm. The region of interest is selected for the current scene; wherein, if the current scene is an instrument scene, transformer breather scene, insulator scene, box door scene, or personnel intrusion scene, the region of interest is the area where the target is located; if the current scene is a cabinet scene or outdoor structure scene, the region of interest is obtained based on the area where the colored pixels in the image are located.

2. The substation image detection method with region-of-interest self-adjustment as described in claim 1, characterized in that: The substation scene division is generated based on different scene targets within the substation.

3. The substation image detection method with region-of-interest self-adjustment as described in claim 1, characterized in that: When the target template image is a handle or pressure plate in a substation, the current scene of the image to be tested is identified as a cabinet scene; When the target template image is a box door in a substation, the current scene of the image to be tested is identified as a box door scene; When the target template image is personnel in a substation, the current scene of the image to be tested is identified as a personnel intrusion scene; When the target template image is an instrument, a breather, or an insulator in a substation, the proportion of the area occupied by the target in the image to be tested is determined. When the proportion is greater than 10%, the current scene of the image to be tested is identified as an instrument scene, a breather scene, or an insulator scene, respectively. When the target template image is an insulator and the target occupies less than 1% of the area in the image under test, the current scene of the image under test is identified as an outdoor elevated scene.

4. The substation image detection method with region-of-interest self-adjustment as described in claim 1, characterized in that: The image to be tested and the target template image are subjected to difference operation and binarization; The binarized image of interest is divided into regions, and regions with an area smaller than a set threshold and regions located at the edge of the image are discarded. The remaining area blocks are merged and set together to form the abnormal alarm image to be output.

5. The substation image detection method with region-of-interest self-adjustment as described in claim 4, characterized in that: Determine the abnormal alarm image based on the abnormal target template image; When an anomaly is detected, an anomaly alarm containing an anomaly alarm image is output.

6. The substation image detection method with region-of-interest self-adjustment as described in claim 5, characterized in that: The region block with an area smaller than the set threshold is a region block whose area is less than 5% of the area of ​​the image to be tested.

7. A substation image detection system with region-of-interest self-adjustment, characterized in that: The system is used to implement the steps of the substation image detection method with region-of-interest self-adjustment as described in any one of claims 1-6; and... The system includes an extraction module, a calibration module, and a detection module; wherein, The extraction module is used to extract a target template image from the substation image and to perform scene division of the substation based on the target template image; The calibration module is used to calibrate the image to be tested with the target template image and determine the scene of the image to be tested. The detection module is used to select the region of interest in the image to be tested based on the scene of the image to be tested, and to perform YOLO target detection on the region of interest. When an abnormality is detected, an abnormality alarm is output.

Images