Image state determination method and system for power inspection image

Abstract

This application provides a method and system for determining the image state of power line inspection images. The method is applied to an image processing terminal and includes: superimposing random impulse noise onto multiple normal images to obtain multiple first abnormal images; for each first abnormal image, replacing a replacement region in a target image with the noise-added region in the first abnormal image to obtain a second abnormal image; extracting first image features of each normal image and second image features of each second abnormal image using a trained image feature extraction model; training an initial prediction model using each first image feature, each second image feature, and a third image feature of each simulated power line inspection image to obtain a target prediction model; and determining the image state of the target power line inspection image using the target prediction model. This method improves the accuracy of the detected image state of the power line inspection images.

Description

Technical Field

[0001] This invention relates to the field of power equipment management, and more specifically, to a method and system for determining the image status of power inspection images. Background Technology

[0002] To promptly detect anomalies in power equipment and effectively eliminate potential safety hazards, power operation and maintenance departments need to conduct regular inspections. However, manual inspections are time-consuming, labor-intensive, and inefficient, and the accuracy of anomaly detection depends heavily on the expertise of the inspectors. With the rapid development of artificial intelligence technology, deep learning-based anomaly detection in power inspection images is gradually becoming a reality.

[0003] However, in actual industrial production environments, power line inspection images often exhibit an imbalance between positive and negative samples. That is, compared to normal samples, there are fewer defective and abnormal samples. This causes the model to tend to predict the majority category, making it easier to misclassify defective and abnormal images as normal images. Consequently, the model trained based on these samples cannot be properly used for anomaly detection in power line inspection images, thus reducing the accuracy of the detected image state when using the above methods for anomaly detection in power line inspection images. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for determining the image status of power inspection images, so as to improve the accuracy of the image status of the detected power inspection images.

[0005] In a first aspect, embodiments of this application provide a method for determining the image status of power inspection images, applied to an image processing terminal in an image status determination system. The image status determination system further includes an image acquisition terminal and an image simulation terminal. The image processing terminal is capable of communicating with both the image acquisition terminal and the image simulation terminal. The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images, and the image simulation terminal is used to simulate images of the power equipment to obtain multiple simulated power inspection images. The method includes:

[0006] Random impulse noise is used to superimpose noise onto multiple normal images to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal;

[0007] For each of the first abnormal images, the replacement region in the target image is replaced by the noise-added region in the first abnormal image to obtain a second abnormal image. The noise-added region is the region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is the region at the same position as the noise-added region in the first abnormal image.

[0008] The first image feature of each normal image and the second image feature of each second abnormal image are extracted using a trained image feature extraction model.

[0009] The target prediction model is obtained by training the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection images;

[0010] Using the target prediction model, the image state of the target power inspection image is determined based on the third image feature of the target power inspection image.

[0011] Optionally, before extracting the first image features of each normal image and the second image features of each of the second abnormal images using the trained image feature extraction model, the method further includes:

[0012] Binarization is performed on each of the normal images and each of the second abnormal images;

[0013] For each normal image and second abnormal image that has undergone binarization, a contrast-limited adaptive histogram equalization process is performed.

[0014] Optionally, the step of training the initial prediction model using each of the first image features, each of the second image features, and each of the simulated power inspection images to obtain the target prediction model includes:

[0015] The first training sample set is constructed by using each of the third image features as input samples of the first training sample set and using the third anomaly probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the first training sample set.

[0016] The target network model is obtained by training the initial network model using the first training sample set;

[0017] The second training sample set is constructed by using each of the first image features and each of the second image features as input samples of the second training sample set, and using the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the second training sample set.

[0018] Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

[0019] Optionally, before determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model, the method further includes:

[0020] The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

[0021] Optionally, determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model includes:

[0022] The third image feature of the target power inspection image is input into the target prediction model to obtain the target anomaly probability value of the target power inspection image;

[0023] Determine whether the target anomaly probability value exceeds the standard anomaly probability value;

[0024] If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as abnormal;

[0025] If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

[0026] Secondly, embodiments of this application provide an image status determination system for power inspection images. The system includes an image processing terminal, an image acquisition terminal, and an image simulation terminal. The image processing terminal is capable of communicating with the image acquisition terminal and the image simulation terminal, respectively.

[0027] The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images;

[0028] The image simulation terminal is used to perform image simulation on the power equipment to obtain multiple simulated power inspection images;

[0029] The image processing terminal is used to superimpose noise onto multiple normal images using random impulse noise to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal.

[0030] The image processing terminal is configured to, for each of the first abnormal images, use the noise-added region in the first abnormal image to replace the replacement region in the target image to obtain a second abnormal image, wherein the noise-added region is a region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is a region at the same position as the noise-added region in the first abnormal image.

[0031] The image processing terminal is used to extract the first image feature of each normal image and the second image feature of each second abnormal image using a trained image feature extraction model.

[0032] The image processing terminal is used to train the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model;

[0033] The image processing terminal is used to determine the image state of the target power inspection image based on the third image feature of the target power inspection image using the target prediction model.

[0034] Optionally, before extracting the first image features of each normal image and the second image features of each second abnormal image using the trained image feature extraction model, the image processing terminal is further configured to:

[0035] Binarization is performed on each of the normal images and each of the second abnormal images;

[0036] For each normal image and second abnormal image that has undergone binarization, a contrast-limited adaptive histogram equalization process is performed.

[0037] Optionally, when the image processing terminal trains the initial prediction model using each of the first image features, each of the second image features, and each of the simulated power inspection images to obtain the target prediction model, it specifically performs the following:

[0038] The first training sample set is constructed by using each of the third image features as input samples of the first training sample set and using the third anomaly probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the first training sample set.

[0039] The target network model is obtained by training the initial network model using the first training sample set;

[0040] The second training sample set is constructed by using each of the first image features and each of the second image features as input samples of the second training sample set, and using the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the second training sample set.

[0041] Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

[0042] Optionally, before determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model, the image processing terminal is further configured to:

[0043] The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

[0044] Optionally, when the image processing terminal determines the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model, it specifically performs the following:

[0045] The third image feature of the target power inspection image is input into the target prediction model to obtain the target anomaly probability value of the target power inspection image;

[0046] Determine whether the target anomaly probability value exceeds the standard anomaly probability value;

[0047] If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as abnormal;

[0048] If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

[0049] The technical solution provided in this application includes, but is not limited to, the following beneficial effects:

[0050] This application utilizes random impulse noise to superimpose noise onto multiple normal images, resulting in multiple first abnormal images. The normal images are those whose image state has been pre-marked as normal from among the multiple actual power inspection images. For each first abnormal image, a replacement region in a target image is replaced using the noisy region in the first abnormal image, resulting in a second abnormal image. The noisy region is the area superimposed by the random impulse noise, the target image is a normal image randomly selected from the multiple normal images, and the replacement region is the region located in the same position as the noisy region in the first abnormal image. Through these steps, the same number of abnormal samples can be obtained based on normal samples in the power inspection images, avoiding the problem of an imbalance between normal and abnormal samples, and providing higher-quality sample data for subsequent model training.

[0051] This application utilizes a trained image feature extraction model to extract the first image feature of each normal image and the second image feature of each second abnormal image; it then uses the first image feature, the second image feature, and the third image feature of each simulated power inspection image to train an initial prediction model to obtain a target prediction model. Through the above steps, by using a balanced number of normal and abnormal samples, as well as simulated samples, to train the network model for anomaly detection, the problem of the trained model being unable to accurately detect anomalies due to imbalanced sample classification can be avoided, thereby improving the performance and detection accuracy of the model used for anomaly detection.

[0052] This application utilizes the target prediction model to determine the image state of the target power inspection image based on the third image features of the target power inspection image. Through the above steps, the image state of the target power inspection image is determined using a model with high detection accuracy and more accurate anomaly detection, which can improve the accuracy of the detected image state of the power inspection image.

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1A flowchart of a method for determining the image status of a power inspection image provided in Embodiment 1 of the present invention is shown;

[0056] Figure 2 A flowchart of an image processing method provided in Embodiment 1 of the present invention is shown;

[0057] Figure 3 The flowchart of a target prediction model determination method provided in Embodiment 1 of the present invention is shown;

[0058] Figure 4 A flowchart of an image state determination method provided in Embodiment 1 of the present invention is shown;

[0059] Figure 5 The diagram shows a schematic of the image status determination system for power inspection images provided in Embodiment 2 of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] To facilitate understanding of this application, the following is combined with... Figure 1 The flowchart illustrating the method for determining the image state of a power inspection image provided in Embodiment 1 of the present invention will be described in detail for Embodiment 1 of this application.

[0063] See Figure 1 The above, Figure 1A flowchart of a method for determining the image status of power inspection images provided in Embodiment 1 of the present invention is shown. The method includes an image processing terminal applied to an image status determination system, which further includes an image acquisition terminal and an image simulation terminal. The image processing terminal can communicate with both the image acquisition terminal and the image simulation terminal. The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images. The image acquisition terminal is installed on a drone. Drone inspection has several advantages in power equipment inspection. First, it has a wide coverage area, enabling rapid coverage of large areas of power equipment, saving time and manpower costs. Second, drone inspection has high safety, effectively reducing personnel safety risks. Furthermore, drone inspection exhibits excellent efficiency; it can quickly acquire image data and perform comprehensive visual monitoring of power equipment, improving inspection efficiency. The image simulation terminal is used to simulate images of the power equipment to obtain multiple simulated power inspection images. The method includes steps S101 to S105:

[0064] S101: Random impulse noise is used to superimpose noise onto multiple normal images to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal.

[0065] Specifically, the image processing terminal randomly generates multiple small-area salt-and-pepper noises, and then randomly selects a preset number of images from multiple actual power inspection images obtained by the image acquisition terminal from the power equipment, which are pre-marked as normal. Each of these preset number of normal images is superimposed with each randomly generated small-area salt-and-pepper noise (the correspondence between each normal image and each small-area salt-and-pepper noise is randomly set) to obtain a first abnormal image that meets the preset number.

[0066] S102: For each of the first abnormal images, the replacement region in the target image is replaced by the noise-added region in the first abnormal image to obtain a second abnormal image, wherein the noise-added region is the region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is the region at the same position as the noise-added region in the first abnormal image.

[0067] Specifically, when superimposing small-area salt-and-pepper noise onto each normal image, since the range of the small-area salt-and-pepper noise is not completely covered and has regional characteristics, the area covered by the small-area salt-and-pepper noise (the noisy area) in the superimposed normal image is extracted. Then, the replacement area in a normal image randomly selected from the multiple normal images, which is in the same position as the noisy area, is replaced. This results in the second abnormal image, thereby achieving image enhancement.

[0068] S103: Using the trained image feature extraction model, extract the first image feature of each normal image and the second image feature of each second abnormal image.

[0069] Specifically, the trained image feature extraction model is the ResNet50 model, which has a deep network structure, stronger representation capabilities, and can capture more abstract and high-level features. Its skip connections also make the network easier to train. Skip connections can also help transfer low-level detail information to high-level features, thereby improving feature extraction performance. Therefore, the trained image feature extraction model is chosen to extract features from the image.

[0070] S104: The initial prediction model is trained using each of the first image features, each of the second image features, and each of the simulated power inspection images to obtain the target prediction model.

[0071] Specifically, the initial prediction model is an untrained initial model, while the target prediction model is a trained model used to determine the probability of image anomalies based on image features.

[0072] S105: Using the target prediction model, determine the image state of the target power inspection image based on the third image features of the target power inspection image.

[0073] Specifically, before determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model, the model quality of the target prediction model can be evaluated (the model can be tested) so that the target prediction model used for step S105 can better predict the anomaly probability and image state of the target power inspection image.

[0074] In one feasible implementation plan, see Figure 2 The above, Figure 2The flowchart of an image processing method provided in Embodiment 1 of the present invention is shown. Before extracting the first image features of each normal image and the second image features of each second abnormal image using a trained image feature extraction model, the method further includes steps S201 to S202:

[0075] S201: Binarize each of the normal images and each of the second abnormal images.

[0076] Specifically, in order to improve the image quality when entering the model for training, each normal image and each second abnormal image need to be preprocessed. The first step of the preprocessing is to binarize each normal image and each second abnormal image.

[0077] S202: Perform contrast-limited adaptive histogram equalization on each normal image and second abnormal image that has completed binarization.

[0078] Specifically, after binarizing each normal image and each second abnormal image, a contrast-limited adaptive histogram equalization (HSE) method is used to preprocess each binarized normal and second abnormal images. The HSE method involves setting a threshold for the original image's histogram; if the histogram grayscale value exceeds the threshold, it is cropped, and then the portion exceeding the threshold is evenly distributed across different grayscale levels, making the mapping function smoother. To avoid blockiness in the mapped sample images, interpolation is performed on the result of the mapping transformation for each pixel. First, the image is divided into blocks, and a histogram mapping function is calculated for each block. For non-boundary pixels, the mapping values ​​of the four adjacent windows are calculated, and then bilinear interpolation is performed to obtain the final pixel mapping value. For boundary pixels, the histogram mapping function mapping value of the nearest window is used for linear interpolation.

[0079] In one feasible implementation plan, see Figure 3 The above, Figure 3 The flowchart illustrates a target prediction model determination method provided in Embodiment 1 of the present invention, wherein the step of training an initial prediction model using each first image feature, each second image feature, and each third image feature of the simulated power inspection image to obtain a target prediction model includes steps S301 to S304:

[0080] S301: Use each of the third image features as input samples of the first training sample set, and use the third anomaly probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the first training sample set to construct the first training sample set.

[0081] S302: The target network model is obtained by training the initial network model using the first training sample set.

[0082] S303: Use each of the first image features and each of the second image features as input samples of the second training sample set, and use the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the second training sample set to construct the second training sample set.

[0083] S304: Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

[0084] Specifically, due to the unique characteristics of power line inspection images, pre-trained models alone are insufficient to learn the rich features of the inspection data. To improve the model's generalization ability in power line inspection images, the network needs to be trained. First, the initial network model is trained using the first training sample set. Then, the model parameters of the target network model obtained from the training are used to co-train the outdoor feature extraction network using the second training sample set.

[0085] The input to the first training sample set is each of the third image features, and the output of the first training sample set is the third anomaly probability value of the simulated power inspection image corresponding to each of the third image features. The input to the second training sample set is each of the first image features and each of the second image features, and the output of the second training sample set is the first anomaly probability value of the normal image corresponding to each of the first image features and the second anomaly probability value of the second abnormal image corresponding to each of the second image features.

[0086] First, perform the first model training:

[0087] The first training sample set is input into the initial network model (using the ResNet50 model) for training. The network parameters are adjusted using the Focal-Loss loss function to obtain the target network model, thereby improving the network's extraction capability.

[0088] Focal-Loss loss function FL(p) t The details are as follows:

[0089] FL(p t )=-α t (1-p t ) γ log(p t )

[0090] Where, p tα is the probability that the model classifies the t-th sample as belonging to the positive class. t It represents the sample weights, where γ is an adjustable parameter, 1-p. t This represents the model's confidence in the negative class. When dealing with imbalanced classification problems, the Focal-Loss loss function reduces the contribution of easily classified negative samples to the loss, thus enabling the model to focus on those difficult-to-classify samples. Because it can be dynamically adjusted based on the sample distribution, the model can adapt to different imbalanced situations. By adjusting the parameters, the model can focus more on difficult samples while maintaining the original distribution of easily classified samples.

[0091] Then, a second model training is performed:

[0092] Using the Sim_meta-Loss loss function, the model parameters in the initial prediction model are adjusted based on the model parameters obtained in the first model training to obtain the target prediction model.

[0093] Sim_meta-Loss loss function L sim_meta (θ,D out D in The details are as follows:

[0094]

[0095] Where θ is the first model parameter, D in and D out These are the first training sample set and the second training sample set, respectively. Indicated by the second model parameter θ i The model obtained below, FL out Let θ0 represent the Focal-Loss on the second training sample set, where β is a hyperparameter controlling the weight of the penalty term, θ0 is the second training sample set, and ||θ||0 is the value of the second training sample set. i -θ0‖ 2 Indicates the current second model parameter θ i The squared Euclidean distance between the parameters obtained from training the second training sample set θ0 and the second training sample set θ0.

[0096] Finally, a target prediction model is obtained that can be used to predict the probability value of image anomalies based on image features.

[0097] In one feasible implementation, before determining the image state of the target power line inspection image based on the third image features of the target power line inspection image using the target prediction model, the method further includes:

[0098] The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

[0099] Specifically, the target power inspection image that needs to be determined in terms of image state is input into the trained image feature extraction model to obtain the third image feature.

[0100] In one feasible implementation plan, see Figure 4 The above, Figure 4 The flowchart of an image state determination method provided in Embodiment 1 of the present invention is shown. The step of determining the image state of the target power inspection image based on the third image feature of the target power inspection image using the target prediction model includes steps S401 to S404:

[0101] S401: Input the third image feature of the target power inspection image into the target prediction model to obtain the target anomaly probability value of the target power inspection image.

[0102] S402: Determine whether the target anomaly probability value exceeds the standard anomaly probability value.

[0103] S403: If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as anomaly.

[0104] S404: If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

[0105] Specifically, based on the trained target prediction model, the target anomaly probability value of the target power inspection image that needs to be determined in terms of image state can be obtained. Then, the image state of the target power inspection image is determined based on the comparison between the target anomaly probability value and the preset standard anomaly probability value.

[0106] Example 2

[0107] See Figure 5 As shown, Figure 5 The diagram shows a schematic of an image status determination system for power inspection images provided in Embodiment 2 of the present invention. The system includes an image processing terminal 501, an image acquisition terminal 502, and an image simulation terminal 503. The image processing terminal is capable of communicating with the image acquisition terminal and the image simulation terminal, respectively.

[0108] The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images;

[0109] The image simulation terminal is used to perform image simulation on the power equipment to obtain multiple simulated power inspection images;

[0110] The image processing terminal is used to superimpose noise onto multiple normal images using random impulse noise to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal.

[0111] The image processing terminal is configured to, for each of the first abnormal images, use the noise-added region in the first abnormal image to replace the replacement region in the target image to obtain a second abnormal image, wherein the noise-added region is a region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is a region at the same position as the noise-added region in the first abnormal image.

[0112] The image processing terminal is used to extract the first image feature of each normal image and the second image feature of each second abnormal image using a trained image feature extraction model.

[0113] The image processing terminal is used to train the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model;

[0114] The image processing terminal is used to determine the image state of the target power inspection image based on the third image feature of the target power inspection image using the target prediction model.

[0115] In one feasible implementation, before extracting the first image features of each normal image and the second image features of each second abnormal image using the trained image feature extraction model, the image processing terminal is further configured to:

[0116] Binarization is performed on each of the normal images and each of the second abnormal images;

[0117] For each normal image and second abnormal image that has undergone binarization, a contrast-limited adaptive histogram equalization process is performed.

[0118] In one feasible implementation, when the image processing terminal trains the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model, it specifically performs the following:

[0119] The first training sample set is constructed by using each of the first image features and each of the second image features as input samples of the first training sample set, and the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the first training sample set. The second training sample set is constructed by using each of the third image features as input samples of the second training sample set, and the third abnormal probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the second training sample set.

[0120] The target network model is obtained by training the initial network model using the first training sample set;

[0121] Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

[0122] In one feasible implementation, before determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model, the image processing terminal is further configured to:

[0123] The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

[0124] In one feasible implementation, when the image processing terminal determines the image state of the target power line inspection image based on the third image features of the target power line inspection image using the target prediction model, it specifically performs the following:

[0125] The third image feature of the target power inspection image is input into the target prediction model to obtain the target anomaly probability value of the target power inspection image;

[0126] Determine whether the target anomaly probability value exceeds the standard anomaly probability value;

[0127] If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as abnormal;

[0128] If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

[0129] The image status determination system for power inspection images provided in this embodiment of the invention can be specific hardware on a device or software or firmware installed on the device. The system provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiments. For the sake of brevity, any parts not mentioned in the system embodiments can be referred to the corresponding content in the aforementioned method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0130] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some communication interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] In addition, the functional units in the embodiments provided by the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0133] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0134] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0135] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for determining the image state of power line inspection images, characterized in that, An image processing terminal is applied to an image state determination system. The image state determination system further includes an image acquisition terminal and an image simulation terminal. The image processing terminal is capable of communicating with both the image acquisition terminal and the image simulation terminal. The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images. The image simulation terminal is used to simulate images of the power equipment to obtain multiple simulated power inspection images. The method includes: Random impulse noise is used to superimpose noise onto multiple normal images to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal; For each of the first abnormal images, the replacement region in the target image is replaced by the noise-added region in the first abnormal image to obtain a second abnormal image. The noise-added region is the region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is the region at the same position as the noise-added region in the first abnormal image. The first image feature of each normal image and the second image feature of each second abnormal image are extracted using a trained image feature extraction model. The target prediction model is obtained by training the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection images; Using the target prediction model, the image state of the target power inspection image is determined based on the third image features of the target power inspection image; The step of training the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model includes: The first training sample set is constructed by using each of the third image features as input samples of the first training sample set and using the third anomaly probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the first training sample set. The target network model is obtained by training the initial network model using the first training sample set; The second training sample set is constructed by using each of the first image features and each of the second image features as input samples of the second training sample set, and using the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the second training sample set. Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

2. The method according to claim 1, characterized in that, Before extracting the first image features of each normal image and the second image features of each of the second abnormal images using the trained image feature extraction model, the method further includes: Binarization is performed on each of the normal images and each of the second abnormal images; For each normal image and second abnormal image that has undergone binarization, a contrast-limited adaptive histogram equalization process is performed.

3. The method according to claim 1, characterized in that, Before determining the image state of the target power line inspection image based on the third image features using the target prediction model, the method further includes: The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

4. The method according to claim 1, characterized in that, The step of determining the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model includes: The third image feature of the target power inspection image is input into the target prediction model to obtain the target anomaly probability value of the target power inspection image; Determine whether the target anomaly probability value exceeds the standard anomaly probability value; If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as abnormal; If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

5. A system for determining the image status of power line inspection images, characterized in that, The system includes an image processing terminal, an image acquisition terminal, and an image simulation terminal, wherein the image processing terminal is capable of communicating with the image acquisition terminal and the image simulation terminal, respectively. The image acquisition terminal is used to acquire images of power equipment to obtain multiple actual power inspection images; The image simulation terminal is used to perform image simulation on the power equipment to obtain multiple simulated power inspection images; The image processing terminal is used to superimpose noise onto multiple normal images using random impulse noise to obtain multiple first abnormal images, wherein the normal images are images whose image states in the multiple actual power inspection images have been pre-marked as normal. The image processing terminal is configured to, for each of the first abnormal images, use the noise-added region in the first abnormal image to replace the replacement region in the target image to obtain a second abnormal image, wherein the noise-added region is a region superimposed by the random impulse noise, the target image is a normal image randomly selected from a plurality of normal images, and the replacement region is a region at the same position as the noise-added region in the first abnormal image. The image processing terminal is used to extract the first image feature of each normal image and the second image feature of each second abnormal image using a trained image feature extraction model. The image processing terminal is used to train the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model; The image processing terminal is used to determine the image state of the target power inspection image based on the third image features of the target power inspection image using the target prediction model. When the image processing terminal trains the initial prediction model using each of the first image features, each of the second image features, and each of the third image features of the simulated power inspection image to obtain the target prediction model, it is specifically used for: The first training sample set is constructed by using each of the first image features and each of the second image features as input samples of the first training sample set, and the first abnormal probability value of the normal image corresponding to each of the first image features and the second abnormal probability value of the second abnormal image corresponding to each of the second image features as output samples of the first training sample set. The second training sample set is constructed by using each of the third image features as input samples of the second training sample set, and the third abnormal probability value of the simulated power inspection image corresponding to each of the third image features as output samples of the second training sample set. The target network model is obtained by training the initial network model using the first training sample set; Using the second training sample set, the initial prediction model is trained according to the model parameters in the target network model to obtain the target prediction model.

6. The system according to claim 5, characterized in that, Before the image processing terminal extracts the first image features of each normal image and the second image features of each second abnormal image using the trained image feature extraction model, it is further configured to: Binarization is performed on each of the normal images and each of the second abnormal images; For each normal image and second abnormal image that has undergone binarization, a contrast-limited adaptive histogram equalization process is performed.

7. The system according to claim 5, characterized in that, Before determining the image state of the target power inspection image based on the third image feature of the target power inspection image using the target prediction model, the image processing terminal is further configured to: The image features of the target power inspection image are extracted using the image feature extraction model to obtain the third image feature.

8. The system according to claim 5, characterized in that, When the image processing terminal uses the target prediction model to determine the image state of the target power line inspection image based on the third image features, it is specifically used for: The third image feature of the target power inspection image is input into the target prediction model to obtain the target anomaly probability value of the target power inspection image; Determine whether the target anomaly probability value exceeds the standard anomaly probability value; If the target anomaly probability value exceeds the standard anomaly probability value, then the image status of the target power inspection image is marked as abnormal; If the target anomaly probability value does not exceed the standard anomaly probability value, then the image status of the target power inspection image is marked as normal.

Images