Substation inspection image identification method and system based on image processing

Abstract

The invention discloses a transformer substation inspection image recognition method and system based on image processing, and relates to the technical field of transformer substation inspection, and the method comprises the steps: remotely obtaining an inspection image of each device in a transformer substation, and marking the collection place and collection time; identifying the equipment type in each inspection image, and grouping the inspection images of different equipment types according to the acquisition place and the acquisition time of the images; sequentially identifying whether defective equipment exists in each group or not by adopting a restricted domain comparison identification algorithm, if not, ignoring the defective equipment, and if so, entering the next processing step; based on the defect database, the identified defects are diagnosed, and an inspection report is generated according to a diagnosis result and forwarded to operation and maintenance personnel; operation and maintenance reports returned by the operation and maintenance personnel are regularly integrated into the defect database, so that the accuracy of a diagnosis result is ensured. According to the restricted domain comparison recognition algorithm, influence factors caused by acquisition time and acquisition places are reduced to a great extent.

Description

Technical Field

[0001] The present invention relates to the technical field of substation inspection, and particularly relates to a substation inspection image recognition method and system based on image processing. Background Art

[0002] The recognition method of substation inspection images is one of the key technologies in the power industry to ensure the safe operation of equipment. For example, image processing technology is used to detect the flame image of a power plant boiler, an infrared thermal imager is used to diagnose electrical equipment faults, and a defect recognition method based on deep learning, etc. These methods can significantly reduce costs and labor intensity and improve the effectiveness of the inspection process. Although significant progress has been made in image processing and recognition technology, complex image analysis and state recognition problems are still in the experimental research stage. There are many types of substation equipment with large appearance differences and are easily affected by factors such as environmental light and shooting angle, which increases the difficulty of image recognition. Moreover, the defect recognition method based on deep learning consumes a large amount of resources. To solve the above problems, researchers in this field urgently need to develop an inspection image recognition method with high recognition accuracy, high stability, and saving computing resources. Summary of the Invention

[0003] The present invention provides a substation inspection image recognition method based on image processing, including:

[0004] Step1. Remotely obtain the inspection images of each device in the substation and mark their collection locations and collection times;

[0005] Step2. Identify the device types in each inspection image and group the inspection images of different device types according to the collection location and collection time of the images;

[0006] Step3. Use the restricted domain contrast recognition algorithm to sequentially identify whether there are defective devices in each group. If not, ignore them. If so, enter the next processing step;

[0007] Step4. Diagnose the identified defects based on the defect database and generate an inspection report according to the diagnosis result and forward it to the operation and maintenance personnel;

[0008] Step5. Regularly integrate the operation and maintenance reports returned by the operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis result.

[0009] For the substation inspection image recognition method based on image processing as described above, where identifying the device types in each inspection image is specifically divided into the following sub-steps:

[0010] Extract the recognition features of various devices in the historical inspection images;

[0011] Associate the identification features of various devices with the device type and the device placement location;

[0012] Traverse the received inspection images, and query the device types whose placement locations are the same as the current acquisition location;

[0013] Calculate the similarity of the identification features between the current inspection image and each device type in the query result, and use the device type with the highest similarity to label the current inspection image.

[0014] A substation inspection image recognition method based on image processing as described above, wherein the inspection images of different device types are grouped according to the image acquisition location and acquisition time, and specifically divided into the following sub-steps:

[0015] Divide the entire substation area into grids to obtain multiple sub-areas;

[0016] Perform the first grouping of the obtained inspection images according to the acquisition time;

[0017] Taking the same sub-area where the acquisition location belongs as the grouping condition, perform the second grouping of the inspection images to obtain the final grouped data.

[0018] A substation inspection image recognition method based on image processing as described above, wherein the restricted domain contrast recognition algorithm is used to sequentially identify whether there are defective devices in each group, and specifically divided into the following sub-steps:

[0019] Define the first inspection image in the group as the algorithm starting point;

[0020] Judge whether the device type of the starting point image has been completely recognized. If so, reset the algorithm starting point to the next inspection image. If not, extract the inspection images with the same device type as the starting point image;

[0021] Traverse the extracted inspection images, and calculate the average difference of each identification feature between the current inspection image and other inspection images;

[0022] If the average difference is greater than the threshold, output the current inspection image as the defective device image;

[0023] After traversing all the extracted inspection images, mark the device type of this batch of inspection images as completely recognized, reset the algorithm starting point to the next image after the current starting point image, and continue the recognition until all the inspection images in the group are recognized.

[0024] A substation inspection image recognition method based on image processing as described above, wherein the identified defects are diagnosed based on the defect database, and an inspection report is generated according to the diagnosis result and forwarded to the operation and maintenance personnel, and specifically divided into the following sub-steps:

[0025] Train a defect diagnosis model based on the data in the defect database;

[0026] Input the features of the defective inspection images into the defect diagnosis model to obtain a diagnosis result;

[0027] Create an inspection report template, and fill the diagnosis result and the inspection images into the template to form a complete inspection report.

[0028] An image recognition method for substation inspection based on image processing as described above, wherein the operation and maintenance reports returned by operation and maintenance personnel are regularly integrated into the defect database, which is specifically divided into the following sub-steps:

[0029] Judge whether the defect diagnosis results in the returned operation and maintenance report are consistent with those in the inspection report;

[0030] If they are consistent, ignore them; if they are inconsistent, take the defect diagnosis results in the operation and maintenance report as the standard and correct the diagnosis results in the defect database;

[0031] If there is no diagnosis result in the inspection report, add the inspection photos in the inspection report and the diagnosis results in the operation and maintenance report to the defect database.

[0032] The present invention also provides an image recognition system for substation inspection based on image processing, which is characterized by including: an inspection image acquisition module, an inspection image grouping module, a defect recognition module, a defect diagnosis module, and a defect data update module;

[0033] The inspection image acquisition module is used to remotely acquire the inspection images of each device in the substation and mark its acquisition location and acquisition time;

[0034] The inspection image grouping module is used to identify the device types in each inspection image and group the inspection images of different device types according to the acquisition location and acquisition time of the images;

[0035] The defect recognition module is used to sequentially identify the defective devices in each group by using the restricted domain contrast recognition algorithm;

[0036] The defect diagnosis module is used to diagnose the identified defects based on the defect database and generate an inspection report according to the diagnosis results and forward it to the operation and maintenance personnel;

[0037] The defect data update module is used to regularly integrate the operation and maintenance reports returned by the operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis results.

[0038] The beneficial effects achieved by the present invention are as follows: The use of the acquisition location reduces the recognition scope of device types, thereby improving the recognition speed of devices. The restricted domain comparison recognition algorithm greatly reduces the influencing factors brought by the acquisition time and acquisition location. The defect database and defect diagnosis model are updated regularly, ensuring stability while improving the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0040] Figure 1 It is a flowchart of a substation inspection image recognition method based on image processing provided in Embodiment 1 of the present application;

[0041] Figure 2 It is a schematic diagram of a substation inspection image recognition system based on image processing provided in Embodiment 2 of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] Embodiment 1

[0044] As Figure 1 shown, Embodiment 1 of the present application provides a substation inspection image recognition method based on image processing, including:

[0045] Step S10: Remotely obtain the inspection images of each device in the substation and mark its acquisition location and acquisition time;

[0046] A data center is set up in the substation to obtain the inspection images collected in real time by the inspection equipment. The obtained inspection images are marked with the acquisition location and acquisition time and encrypted and stored in the data center.

[0047] Step S20: Identify the device types in each inspection image and group the inspection images of different device types according to the acquisition location and acquisition time of the images;

[0048] Current image recognition calculations can identify the types of objects based on the features of the objects in the image. Using this technology, the types of equipment can be identified from the inspection images. However, considering that the equipment in the substation is rarely moved in terms of its placement, it is entirely feasible to shorten the image recognition time by combining the factor of location in the substation. Specifically:

[0049] I. Extract the recognition features of various types of equipment in the historical inspection images;

[0050] Use image processing technology to extract the recognition features of various types of equipment from the historical inspection images, including color features, shape features, texture features, and spatial relationship features. Subsequently, a recognition weight is generated for each recognition feature through a support vector machine. The feature with a higher weight plays a greater role in type recognition.

[0051] II. Associate the recognition features of various types of equipment with the equipment type and the equipment placement location;

[0052] III. Traverse the received inspection images and query the equipment types with the placement location consistent with the current acquisition location;

[0053] IV. Calculate the similarity of the recognition features between the current inspection image and each equipment type in the query result, and use the equipment type with the highest similarity to label the current inspection image;

[0054] The calculation formula for calculating the similarity between the inspection image and the equipment type recognition features is expressed as:

[0055] where a i is the i-th recognition feature of the inspection image, μ i is the recognition weight of the i-th recognition feature, b i is the i-th recognition feature of a certain equipment type, i takes values from 1 to n, n is the total number of recognition features, and S is the similarity calculation result.

[0056] After identifying the equipment type of the inspection image, start grouping the inspection images. Specifically, it is divided into the following sub-steps:

[0057] I. Divide the entire substation area into grids to obtain multiple sub-areas;

[0058] The parameters for dividing the grids are defined according to the actual application scenario, but it is necessary to ensure that the number of equipment of the same type in the sub-area is more than two. That is to say, it is necessary to ensure that each piece of equipment has a reference for comparison with each other, and this reference must be equipment of the same type.

[0059] II. Perform the first grouping of the obtained inspection images according to the acquisition time;

[0060] This step is to ensure that even if the inspection images are received with a delay due to problems in the data transmission link, it will not affect the (defect) recognition effect of the inspection images;

[0061] III. Group the inspection images for the second time with the same sub-region where the collection location belongs as the grouping condition to obtain the final grouped data;

[0062] At this time, the inspection pictures in one group all come from one sub-region and the collection time is also consistent.

[0063] Step S30: Use the restricted domain comparison and recognition algorithm to sequentially identify whether there are defective devices in each group. If not, ignore it. If so, enter the next processing step;

[0064] The restricted domain comparison and recognition algorithm means restricting the comparison and recognition domain within each group, minimizing the influencing factors caused by different collection times and collection positions to the greatest extent, so as to achieve the effect of improving the defect recognition accuracy and speed. The specific process can be divided into the following sub-steps:

[0065] I. Define the first inspection image in the group as the starting point of the algorithm;

[0066] II. Judge whether the device type of the starting point image has been recognized. If so, reset the starting point of the algorithm to the next inspection image. If not, extract the inspection images with the same device type as the starting point image;

[0067] Persons in this field should be aware that this step is to reset the starting point of the algorithm to the inspection image of the next device type, so as to extract all the inspection images of this device type;

[0068] The extracted inspection images and the current starting point image are stored in an ordered sequence together, and the arrangement order is the same as the arrangement order in the group.

[0069] III. Traverse the extracted inspection images and calculate the difference mean value of each recognition feature between the current inspection image and other inspection images;

[0070] When humans distinguish defects on the surface of a device, they will first notice the positions where the color or texture is significantly different from other parts, such as positions with obvious damage like paint peeling, weathering, and bumps. Therefore, when this algorithm calculates the difference between the texture feature and the color feature, it will also consider the difference changes between different parts of the device itself to improve the accuracy. The specific difference mean value calculation formula is expressed as:

[0071]

[0072] Where h 0 、p 0respectively represent the shape feature and the spatial relationship feature of the current inspection image, h j and p j respectively represent the shape feature and the spatial relationship feature of the j-th inspection image except the current inspection image, e jk and c jk respectively represent the texture feature and the color feature of the k-th sub-region in the j-th inspection image, e jk+1 and c jk+1 respectively represent the texture feature and the color feature of the (k + 1)-th sub-region in the j-th inspection image, e 0k and c 0k respectively represent the texture feature and the color feature of the k-th sub-region in the current inspection image, e 0k+1 and c 0k+1 respectively represent the texture feature and the color feature of the (k + 1)-th sub-region in the current inspection image. k takes values from 1 to w - 1, where w is the total number of sub-regions segmented from the inspection image, j takes values from 1 to m - 1, where m is the total number of inspection images extracted this time, and D is the calculation result of the average difference.

[0073] IV. If the average difference is greater than the threshold, output the current inspection image as a defective device image;

[0074] V. After traversing all the extracted inspection images, mark the device type of this batch of inspection images as recognized, reset the algorithm starting point to the next inspection image of the current starting image, and continue to execute step II until all the inspection images in the group are recognized;

[0075] Each inspection image in each group is processed through the above steps. It should be noted that the defective inspection images output in this step will be transmitted in real time to step S40 for processing, without waiting until all inspection images are processed.

[0076] Step S40: Diagnose the recognized defects based on the defect database, and generate an inspection report according to the diagnosis result and forward it to the operation and maintenance personnel;

[0077] The defect database maintains all the discovered device defects and their related inspection images in the substation. Use these data to train a defect diagnosis model to achieve the effect of automatic diagnosis of inspection images. Specifically:

[0078] I. Train a defect diagnosis model based on the data in the defect database;

[0079] Extract the device recognition features in each inspection image in the defect database as inputs, including shape features, color features, texture features, and spatial relationship features; the corresponding historical diagnosis results as outputs, and train a defect diagnosis model in a supervised manner, denoted as:

[0080] where Y is the model output, and x v is the v-th recognition feature of the input, and ε v is the autoregressive coefficient of the v-th recognition feature, is the diagnostic coefficient of the v-th recognition feature, and θ v is the diagnostic bias of the v-th recognition feature. v ranges from 1 to V, where V is the length of the input set, and δ u is the output weight of the u-th layer of the model, is the output bias of the u-th layer of the model. u ranges from 1 to U, where U is the total number of layers of the model.

[0081] II. Input the defective inspection image features into the defect diagnosis model to obtain a diagnosis result;

[0082] III. Create an inspection report template, and fill the diagnosis result and the inspection image into the template to form a complete inspection report;

[0083] The inspection report template presets two sections: diagnosis result and details. The details section is used to display details fields such as the inspection image, collection location, and collection time, while the diagnosis result is used to display the output result of the model.

[0084] Step S50: Regularly integrate the operation and maintenance reports sent back by the operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis results;

[0085] The operation and maintenance reports sent back by the operation and maintenance personnel contain accurate defect diagnosis content. By integrating this data into the defect database and then using the new integrated content to perform incremental training on the defect diagnosis model, the diagnosis results can become more comprehensive and accurate. Specifically, the integration process is divided into the following sub-steps:

[0086] I. Determine whether the defect diagnosis results in the sent-back operation and maintenance report are consistent with those in the inspection report;

[0087] II. If they are consistent, ignore them. If they are inconsistent, correct the diagnosis results in the defect database based on the defect diagnosis results in the operation and maintenance report;

[0088] III. If there is no diagnosis result in the inspection report, add the inspection photos in the inspection report and the diagnosis results in the operation and maintenance report to the defect database;

[0089] If the defect diagnosis model encounters a device defect it has never seen, that is, there is no information about this type of defect in the defect database, the output result will be filled back into the inspection report as an empty string. At this time, the processing method of this step will be triggered.

[0090] Embodiment 2

[0091] Such as Figure 2As shown in the figure, Embodiment 2 of the present application provides a substation inspection image recognition system based on image processing, including: an inspection image acquisition module 21, an inspection image grouping module 22, a defect recognition module 23, a defect diagnosis module 24, and a defect data update module 25;

[0092] The inspection image acquisition module 21 is used to remotely acquire inspection images of various devices in the substation and mark their acquisition locations and acquisition times.

[0093] The inspection image grouping module 22 is used to identify the device types in each inspection image and group the inspection images of different device types according to the acquisition location and acquisition time of the images; specifically including: a device type identification sub-module, a substation division sub-module, and a grouped data acquisition sub-module;

[0094] The device type identification sub-module is used to identify the type of device in the image in combination with the acquisition location of the inspection image;

[0095] Current image recognition calculations can identify the types of objects based on the features of the objects in the image. Using this technology, the types of devices can be identified from the inspection images. However, considering that the positions of the devices in the substation are rarely changed in practice, then combining the factor of location to shorten the image recognition time is completely feasible in the substation. Specifically:

[0096] I. Extract the recognition features of various devices in the historical inspection images;

[0097] Use image processing technology to extract the recognition features of various devices from the historical inspection images, including color features, shape features, texture features, and spatial relationship features. Subsequently, a recognition weight is generated for each recognition feature through a support vector machine. The higher the weight, the greater the role of the feature in type recognition.

[0098] II. Associate the recognition features of various devices with the device types and device placement locations;

[0099] III. Traverse the received inspection images and query the device types with the same placement location as the current acquisition location;

[0100] IV. Calculate the similarity between the current inspection image and the recognition features of each device type in the query result, and use the device type with the highest similarity to label the current inspection image;

[0101] The calculation formula for calculating the similarity between the inspection image and the device type recognition features is expressed as:

[0102] where a i is the i-th recognition feature of the inspection image, μ iis the recognition weight of the i-th recognition feature, b i is the i-th recognition feature of a certain device type, where i ranges from 1 to n, n is the total number of recognition features, and S is the similarity calculation result.

[0103] The sub-module for sub-dividing the substation is used to divide the grid for the entire substation area to obtain multiple sub-areas;

[0104] The parameters for dividing the grid are defined according to the actual application scenario, but it is necessary to ensure that the number of devices of the same type within the sub-area is more than two. That is to say, it is necessary to ensure that each device has a reference for comparison, and this reference must be a device of the same type.

[0105] The sub-module for obtaining grouped data is used to perform the first grouping of the obtained inspection images according to the acquisition time, and then perform the second grouping of the inspection images with the same sub-area where the acquisition location belongs as the grouping condition to obtain the final grouped data;

[0106] Grouping by acquisition time can ensure that even if the inspection images are received late due to problems in the data transmission link, it will not affect the (defect) recognition effect of the inspection images; finally, the inspection images within each group are from one sub-area and the acquisition time is also the same.

[0107] The defect recognition module 23 is used to sequentially identify the defective devices in each group by using the restricted domain comparison recognition algorithm;

[0108] The restricted domain comparison recognition algorithm means restricting the comparison recognition domain within each group to minimize the influencing factors caused by different acquisition times and acquisition positions, so as to achieve the effect of improving the defect recognition accuracy and speed. The specific process can be divided into the following sub-steps:

[0109] I. Define the first inspection image in the group as the starting point of the algorithm;

[0110] II. Determine whether the device type of the starting point image has been recognized. If so, reset the starting point of the algorithm to the next inspection image. If not, extract the inspection images with the same device type as the starting point image;

[0111] The extracted inspection images and the current starting point image are stored in an ordered sequence together, and the arrangement order is the same as the arrangement order within the group.

[0112] III. Traverse the extracted inspection images and calculate the average difference of each recognition feature between the current inspection image and other inspection images;

[0113] When humans identify defects on the surface of a device, they first notice areas where the color or texture is significantly different from other parts, such as areas with paint peeling, weathering, or bumps that are clearly damaged at a glance. Therefore, when this algorithm calculates the difference between texture features and color features, it also considers the differences and changes between various parts of the device itself to improve accuracy. The specific formula for calculating the difference mean is expressed as:

[0114]

[0115] Where h 0 and p 0 represent the shape feature and spatial relationship feature of the current inspection image respectively, h j and p j represent the shape feature and spatial relationship feature of the j-th inspection image except the current inspection image respectively, e jk and c jk represent the texture feature and color feature of the k-th sub-domain in the j-th inspection image respectively, e jk+1 and c jk+1 represent the texture feature and color feature of the (k + 1)-th sub-domain in the j-th inspection image respectively, e 0k and c 0k represent the texture feature and color feature of the k-th sub-domain in the current inspection image respectively, e 0k+1 and c 0k+1 represent the texture feature and color feature of the (k + 1)-th sub-domain in the current inspection image respectively. k takes values from 1 to w - 1, where w is the total number of sub-domains segmented from the inspection image, j takes values from 1 to m - 1, where m is the total number of inspection images extracted this time, and D is the calculation result of the difference mean.

[0116] IV. If the difference mean is greater than the threshold, output the current inspection image as a defective device image;

[0117] V. After traversing all the extracted inspection images, mark the device type of this batch of inspection images as recognized, reset the algorithm starting point to the next inspection image of the current starting image, and continue to execute step II until all the inspection images in the group are recognized;

[0118] Each inspection image in each group is processed through the above steps. It should be noted that the defective inspection images output by this module will be transmitted to the defect diagnosis module 24 in real time for processing, without waiting for all the inspection images to be processed.

[0119] The defect diagnosis module 24 is used to diagnose the identified defects based on the defect database and generate an inspection report according to the diagnosis results and forward it to the operation and maintenance personnel; specifically including: the defect diagnosis model training sub-module, the diagnosis result acquisition sub-module, and the inspection report generation sub-module;

[0120] A defect diagnosis model training sub-module, which is used to train a defect diagnosis model using the data in the defect database;

[0121] Extract the device recognition features in each inspection image in the defect database as inputs, including shape features, color features, texture features, and spatial relationship features; the corresponding historical diagnosis results as outputs, and a defect diagnosis model is trained in a supervised manner, expressed as:

[0122] Where Y is the model output, x v is the v-th recognition feature of the input, εv is the autoregressive coefficient of the v-th recognition feature, is the diagnosis coefficient of the v-th recognition feature, θ v is the diagnosis bias of the v-th recognition feature, v takes values from 1 to V, V is the length of the input set, δ u is the output weight of the u-th layer of the model, is the output bias of the u-th layer of the model, u takes values from 1 to U, and U is the total number of layers of the model.

[0123] A diagnosis result acquisition sub-module, which is used to input the features of defective inspection images into the defect diagnosis model to obtain the diagnosis result.

[0124] An inspection report generation sub-module, which is used to create an inspection report template and fill back the diagnosis result and the inspection image into the template to form a complete inspection report;

[0125] The inspection report template presets two sections: diagnosis result and details. The details section is used to display details fields such as the inspection image, collection location, and collection time, and the diagnosis result is used to display the output result of the model.

[0126] The defect data update module 25 is used to regularly integrate the operation and maintenance reports sent back by the operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis results;

[0127] The operation and maintenance reports sent back by the operation and maintenance personnel contain accurate defect diagnosis content. By integrating this data into the defect database and then using the integrated new content to perform incremental training on the defect diagnosis model, the diagnosis results can be made more comprehensive and accurate. Specifically, the integration process is divided into the following sub-steps:

[0128] I. Judge whether the defect diagnosis results in the operation and maintenance report sent back are consistent with those in the inspection report;

[0129] II. If they are consistent, ignore them. If they are inconsistent, use the defect diagnosis result in the operation and maintenance report as the standard to correct the diagnosis result in the defect database;

[0130] III. If there is no diagnostic result in the inspection report, add the inspection photos in the inspection report and the diagnostic results in the operation and maintenance report to the defect database;

[0131] When the defect diagnosis model encounters a device defect it has never seen, that is, when there is no information about this type of defect in the defect database, the output result will be filled back into the inspection report as an empty string, and this will trigger the processing method of this step.

[0132] Corresponding to the above embodiments, an embodiment of the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0133] The memory is used to store one or more program instructions;

[0134] The processor is used to run one or more program instructions to execute a substation inspection image recognition method based on image processing.

[0135] Corresponding to the above embodiments, an embodiment of the present invention provides a computer-readable storage medium. The computer storage medium contains one or more program instructions, and the one or more program instructions are used to be executed by the processor to execute a substation inspection image recognition method based on image processing.

[0136] An embodiment disclosed by the present invention provides a computer-readable storage medium. Computer program instructions are stored in the computer-readable storage medium. When the computer program instructions run on a computer, the computer is enabled to execute the above-mentioned substation inspection image recognition method based on image processing.

[0137] In the embodiments of the present invention, the processor may be an integrated circuit chip with signal processing capabilities. The processor may be a general-purpose processor, a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0138] The various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. The processor reads the information in the storage medium and combines its hardware to complete the steps of the above method.

[0139] The storage medium can be a memory, for example, it can be a volatile memory or a non-volatile memory, or it can include both volatile and non-volatile memories.

[0140] Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0141] The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM).

[0142] The storage medium described in the embodiments of the present invention is intended to include but not be limited to these and any other suitable types of memories.

[0143] Those skilled in the art should be able to realize that in one or more of the above examples, the functions described in the present invention can be implemented by a combination of hardware and software. When applying software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. The computer-readable medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transmission of a computer program from one place to another. The storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0144] The specific embodiments described above have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the present invention should be included in the protection scope of the present invention.

Claims

1. A substation inspection image recognition method based on image processing, characterized in that: include: Step 1: Remotely obtain inspection images of each device in the substation and mark the collection location and time; Step 2: Identify the device type in each inspection image, and group the inspection images of different device types according to the image acquisition location and acquisition time; Step 3: Use the restricted domain comparison and recognition algorithm to identify whether there are defective devices in each group in turn. If not, ignore it. If so, proceed to the next processing step; Step 4: Diagnose the identified defects based on the defect database, and generate an inspection report based on the diagnosis results and forward it to the operation and maintenance personnel; Step 5. Regularly integrate the operation and maintenance reports sent back by operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis results.

2. A substation inspection image recognition method based on image processing according to claim 1, characterized in that: Identify the device type in each inspection image, which is divided into the following sub-steps: Extract identification features of various types of equipment in historical inspection images; Associating the identification features of various types of equipment with the equipment type and equipment placement; Traverse the received inspection images and query the equipment type whose placement location is consistent with the current collection location; Calculate the similarity between the identification features of the current inspection image and each device type in the query result, and use the device type with the highest similarity to annotate the current inspection image.

3. The substation inspection image recognition method based on image processing according to claim 1 is characterized in that: Group inspection images of different equipment types according to the location and time of image acquisition. The specific steps are as follows: Divide the entire substation area into grids to obtain multiple sub-areas; The acquired inspection images are grouped for the first time according to the acquisition time; The inspection images are grouped for the second time based on the same sub-area to which the acquisition locations belong, to obtain the final grouping data.

4. The substation inspection image recognition method based on image processing according to claim 1 is characterized in that: The restricted domain comparison and recognition algorithm is used to identify whether there are defective devices in each group in turn. The specific steps are as follows: The first inspection image in the group is defined as the algorithm starting point; Determine whether the device type of the starting image has been identified. If so, reset the algorithm starting point to the next inspection image. If not, extract the inspection image that is consistent with the device type of the starting image. Traverse the extracted inspection images and calculate the mean difference of each identification feature between the current inspection image and other inspection images; If the difference mean is greater than the threshold, the current inspection image is output as a defective device image; After the extracted inspection images are traversed, the device type of the batch of inspection images is marked as recognized, and the algorithm starting point is reset to the next inspection image of the current starting point image, and recognition continues until all inspection images in the group are recognized.

5. The substation inspection image recognition method based on image processing according to claim 1 is characterized in that: The identified defects are diagnosed based on the defect database, and an inspection report is generated based on the diagnosis results and forwarded to the operation and maintenance personnel. The specific steps are as follows: Train a defect diagnosis model based on the data in the defect database; Input the defective inspection image features into the defect diagnosis model to obtain the diagnosis result; Create an inspection report template and fill the diagnosis results and inspection images into the template to form a complete inspection report.

6. The substation inspection image recognition method based on image processing according to claim 1 is characterized in that: Regularly integrate the operation and maintenance reports sent back by the operation and maintenance personnel into the defect database, which is divided into the following sub-steps: Determine whether the defect diagnosis results in the returned operation and maintenance report are consistent with those in the inspection report; If they are consistent, they are ignored. If they are inconsistent, the defect diagnosis results in the operation and maintenance report shall prevail, and the diagnosis results in the defect database shall be corrected; If there is no diagnosis result in the inspection report, the inspection photos in the inspection report and the diagnosis result in the operation and maintenance report are added to the defect database.

7. A substation inspection image recognition system based on image processing, characterized in that: include: Inspection image acquisition module, inspection image grouping module, defect recognition module, defect diagnosis module, defect data update module; Inspection image acquisition module, used to remotely acquire inspection images of each device in the substation and mark the acquisition location and time; The inspection image grouping module is used to identify the device type in each inspection image and group the inspection images of different device types according to the image acquisition location and acquisition time; A defect identification module is used to sequentially identify defective devices in each group using a restricted domain comparison identification algorithm; The defect diagnosis module is used to diagnose the identified defects based on the defect database, and generate an inspection report based on the diagnosis results and forward it to the operation and maintenance personnel; The defect data update module is used to regularly integrate the operation and maintenance reports sent back by operation and maintenance personnel into the defect database to ensure the accuracy of the diagnosis results.

8. A substation inspection image recognition system based on image processing according to claim 7, characterized in that: The inspection image grouping module specifically includes: equipment type identification submodule, substation segmentation submodule, and grouping data acquisition submodule; The equipment type identification submodule is used to identify the type of equipment in the image in combination with the collection location of the inspection image; Substation subdivision module, used to divide the entire substation area into grids to obtain multiple sub-areas; The grouping data acquisition submodule is used to group the acquired inspection images for the first time according to the acquisition time, and then group the inspection images for the second time based on the same sub-area to which the acquisition location belongs as the grouping condition to obtain the final grouping data.

9. A computer storage medium, characterized in that: include: at least one memory and at least one processor; A memory for storing one or more program instructions; A processor is used to run one or more program instructions to execute the substation inspection image recognition method based on image processing as described in any one of claims 1 to 6.

Images