Target detection model construction method and substation foreign matter intrusion detection method

Abstract

The application relates to a target detection model construction method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring an initial target detection training set comprising a first sample set carrying label information and a second sample set not carrying label information; training a target detector based on the initial target detection training set; calling the target detector to perform target detection on the second sample set to obtain a target detection result; then, continuously updating the initial target detection training set based on the target detection result until a preset iteration termination condition is reached, obtaining a target detection training set, and constructing a target detection model based on the target detection training. The method can obtain a target detection model with higher precision and support high-precision detection. The application further provides a power transformation station foreign matter intrusion detection method and device, computer equipment, a storage medium and a computer program product. The method can accurately detect foreign matters intruding into a power transformation station.

Description

Technical Field

[0001] This application relates to the field of power equipment testing technology, and in particular to a target detection model construction method, device, computer equipment, storage medium and computer program product, as well as a substation foreign object intrusion detection method, device, computer equipment, storage medium and computer program product. Background Technology

[0002] Substations are a crucial component of power systems, primarily used for transforming, distributing, controlling, and protecting electrical energy. During substation operation, various equipment and systems need to operate stably to ensure the continuity and reliability of power supply. However, the substation environment often contains various foreign objects, such as oil stains, weeds, leaves, and dust, which may adversely affect the operation of substation equipment.

[0003] To promptly detect and resolve foreign object intrusion issues in substations and ensure the operational stability and reliability of power equipment, regular inspections are necessary. Traditional inspection methods typically rely on manual inspections, which are inefficient and prone to missing items. Therefore, intelligent inspection methods have emerged. Intelligent inspection primarily utilizes machine vision and image recognition technologies to monitor and identify the condition of substation equipment in real time, enabling rapid detection and location of foreign object intrusion problems.

[0004] However, current intelligent inspection solutions still suffer from low accuracy in detecting foreign object intrusion, which affects the safe and stable operation of substations. Summary of the Invention

[0005] Therefore, it is necessary to address the above-mentioned technical problems by providing a target detection model construction method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can support accurate foreign object intrusion detection, as well as a substation foreign object intrusion detection method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can achieve accurate foreign object intrusion detection.

[0006] Firstly, this application provides a method for constructing an object detection model. The method includes:

[0007] Obtain the initial object detection training set, which includes a first sample set with labeled information and a second sample set without labeled information;

[0008] The pre-trained target detection network is trained based on the initial target detection training set to obtain the target detector;

[0009] The target detector is invoked to perform target detection on the second sample set, and the target detection results are obtained, including the confidence scores of each sample in the second sample set.

[0010] From the second sample set, a positive sample prediction set with a confidence level higher than a preset first confidence level threshold and a negative sample prediction set with a confidence level lower than a preset second confidence level threshold are selected. The preset first confidence level threshold is greater than the preset second confidence level threshold.

[0011] Based on the positive and negative sample prediction sets, update the initial target detection training set, return to the step of training the pre-trained target detection network based on the initial target detection training set to obtain the target detector, until the preset iteration termination condition is reached.

[0012] The pre-trained object detection network is trained using the latest initial object detection training set to obtain the object detection model.

[0013] In one embodiment, the pre-trained object detection network is trained based on an initial object detection training set to obtain an object detector comprising:

[0014] The pre-trained target detection network is trained using the first sample set to obtain the initial target detector;

[0015] An initial target detector is used to perform target detection on the second sample set to obtain the initial target detection result. The initial target detection result includes the second sample set carrying annotation information, including confidence level.

[0016] From the second sample set carrying labeled information, a set of pseudo-negative samples with a confidence level higher than a preset third confidence level threshold is selected, where the preset third confidence level threshold is lower than the preset second confidence level threshold.

[0017] The initial target detector is trained based on the first sample set and the pseudo-negative sample set, thus obtaining the target detector.

[0018] In one embodiment, after selecting a positive sample prediction set with a confidence level higher than a preset first confidence level threshold and a negative sample prediction set with a confidence level lower than a preset second confidence level threshold from the second sample set, the method further includes:

[0019] Based on the confidence level of each sample in the second sample set, the scores of each predicted positive sample in the positive sample prediction set and the scores of each predicted negative sample in the negative sample prediction set are determined to update the positive sample prediction set and the negative sample prediction set.

[0020] In one embodiment, before updating the initial object detection training set based on the positive sample prediction set and the negative sample prediction set, the method further includes:

[0021] Based on the confidence level of each sample in the second sample set, determine the scores of each predicted positive sample in the positive sample prediction set and the scores of each predicted negative sample in the negative sample prediction set, so as to update the positive sample prediction set and the negative sample prediction set.

[0022] After obtaining the object detection model, the method also includes:

[0023] Based on the confidence level of each sample in the second sample set, determine the score of each predicted positive sample in the positive sample prediction set and the score of each predicted negative sample in the negative sample prediction set.

[0024] Based on the scores of the predicted positive samples determined in each iteration, the average score of the predicted positive samples is determined, and the average score of the predicted positive samples is used as the weight of the predicted positive samples. Similarly, based on the scores of the predicted negative samples determined in each iteration, the average score of the predicted negative samples is determined, and the average score of the predicted negative samples is used as the weight of the predicted negative samples.

[0025] Based on the weights of predicted positive samples and predicted negative samples, the loss function of the object detection model is recalculated, and the object detection model is updated.

[0026] Secondly, this application also provides an apparatus for constructing an object detection model. The apparatus includes:

[0027] The initial training set acquisition module is used to acquire the initial object detection training set, which includes a first sample set carrying labeled information and a second sample set without labeled information.

[0028] The object detector training module is used to train the pre-trained object detection network based on the initial object detection training set to obtain the object detector;

[0029] The target detection module is used to call the target detector to perform target detection on the second sample set and obtain the target detection results, which include the confidence scores of each sample in the second sample set.

[0030] The data filtering module is used to filter out positive sample prediction sets with confidence levels higher than a preset first confidence threshold and negative sample prediction sets with confidence levels lower than a preset second confidence threshold from the second sample set. The preset first confidence threshold is greater than the preset second confidence threshold.

[0031] The training set update module is used to update the initial target detection training set based on the positive sample prediction set and the negative sample prediction set, and re-control the target detector training module to train the pre-trained target detection network based on the initial target detection training set to obtain the target detector operation until the preset iteration termination condition is reached.

[0032] The model building module is used to train the pre-trained object detection network based on the latest initial object detection training set to obtain the object detection model.

[0033] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above-described target detection model construction method.

[0034] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps in the above-described target detection model construction method.

[0035] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in the above-described target detection model construction method.

[0036] The aforementioned object detection model construction method, apparatus, computer equipment, storage medium, and computer program products utilize an unlabeled second sample set (unlabeled data) to optimize the initial object detection training set. This helps the object detector better capture data distribution, improves the stability and robustness of the object detector, and differs from the traditional approach of only adding a second sample subset (positive sample data) with a confidence level greater than a preset first confidence threshold to update the initial object detection training set. By continuously optimizing and updating the initial object detection set based on the positive and negative sample prediction sets, this approach effectively addresses the problem of class imbalance in the training set, improves the accuracy of the object detector, and consequently makes the labeled data in the object detection training set obtained through iterative optimization of the object detector more accurate. Finally, training and constructing an object detection model based on the object detection training set results in higher performance and accuracy of the object detection model, supporting more precise object detection schemes.

[0037] Sixthly, this application provides a method for detecting foreign object intrusion in a substation. The method includes:

[0038] Acquire real-time monitoring images of the substation;

[0039] The trained foreign object intrusion detection model is invoked to perform foreign object intrusion detection on real-time monitoring images, and the foreign object intrusion detection results are obtained.

[0040] The foreign object intrusion detection model is constructed based on historical monitoring images of the substation, using the target detection model construction method described in any of the above embodiments.

[0041] In one embodiment, after obtaining the foreign object intrusion detection result, the method further includes:

[0042] If the foreign object intrusion detection result indicates the presence of an intruding foreign object in the real-time monitoring image, an early warning message will be sent.

[0043] Seventhly, this application also provides a foreign object intrusion detection device for substations. The device includes:

[0044] The monitoring image acquisition module is used to acquire real-time monitoring images of the substation.

[0045] The foreign object intrusion detection module is used to call the trained foreign object intrusion detection model to perform foreign object intrusion detection on real-time monitoring images and obtain foreign object intrusion detection results;

[0046] The foreign object intrusion detection model is constructed based on historical monitoring images of the substation, using the target detection model construction method described in any of the above embodiments.

[0047] Eighthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the aforementioned foreign object intrusion detection method.

[0048] Ninthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned foreign object intrusion detection method.

[0049] Tenthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in the aforementioned foreign object intrusion detection method.

[0050] The aforementioned methods, devices, computer equipment, storage media, and computer program products for detecting foreign objects in substations, by utilizing the foreign object intrusion detection model constructed using the aforementioned target detection model construction method, enable the foreign object intrusion detection model to have stronger performance and generalization ability, higher model accuracy and robustness. Correspondingly, by using this foreign object intrusion detection model to perform foreign object intrusion detection on real-time monitoring images of substations, it is possible to quickly and accurately identify intrusive foreign objects inside the substation and promptly detect potential safety hazards. Attached Figure Description

[0051] Figure 1 This is an application environment diagram of a foreign object intrusion detection method in a substation in one embodiment;

[0052] Figure 2 This is a flowchart illustrating a method for constructing a target detection model in one embodiment;

[0053] Figure 3 This is a flowchart illustrating the steps of training an object detector based on an initial object detection training set in one embodiment.

[0054] Figure 4 This is a detailed flowchart illustrating the object detection model construction method in another embodiment;

[0055] Figure 5 This is a detailed flowchart illustrating the object detection model construction method in yet another embodiment;

[0056] Figure 6 This is a detailed flowchart of a foreign object intrusion detection method for a substation in one embodiment;

[0057] Figure 7 This is a flowchart illustrating a foreign object intrusion detection method for substations in another embodiment;

[0058] Figure 8 This is a structural block diagram of a target detection model construction device in one embodiment;

[0059] Figure 9 This is a structural block diagram of the target detection model construction device in another embodiment;

[0060] Figure 10 This is a structural block diagram of a foreign object intrusion detection device for a substation in one embodiment;

[0061] Figure 11 This is a structural block diagram of a substation foreign object intrusion detection device in another embodiment;

[0062] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0065] The foreign object intrusion detection method for substations provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Specifically, the entire substation foreign object intrusion detection method can include a model building phase and a foreign object intrusion detection phase based on the built model.

[0066] During the model building phase, an operator can upload the initial object detection training set to the server 104 via terminal 102 and send a training set optimization message to the server 104 via terminal 102. The server 104 responds to the message, obtains the initial object detection training set, which includes a first sample set with labeled information and a second sample set without labeled information. Based on the initial object detection training set, a pre-trained object detection network is trained to obtain an object detector. The object detector is then used to perform object detection on the second sample set to obtain the object detection results. The object detection results include the confidence scores of each sample in the second sample set. Based on the confidence scores, a positive sample prediction set and a negative sample prediction set are selected from the second sample set. In this prediction set, the negative sample prediction set is the second sample subset with a confidence level lower than a preset second confidence level threshold, and the positive sample prediction set is the second sample subset with a confidence level higher than a preset first confidence level threshold. The object detector is trained based on the first sample set and the third sample set carrying labeled information. According to the positive and negative sample prediction sets, the initial object detection training set is updated, and the process of training the pre-trained object detection network based on the initial object detection training set to obtain the object detector continues until a preset iteration termination condition is reached. Then, the iteration stops, the current initial object detection training set is determined as the object detection training set, and the pre-trained object detection network is trained based on the object detection training set to obtain the object detection model.

[0067] In the foreign object intrusion detection stage based on the constructed foreign object intrusion detection model, the substation's operation and maintenance personnel can pre-construct the foreign object intrusion detection model using the target detection model construction method described in any of the above embodiments, deploy the foreign object intrusion detection model on server 104, and preset the triggering conditions for foreign object intrusion detection. In practical applications, terminal 102 synchronously uploads the received real-time monitoring images of the substation to server 104. When server 104 receives the real-time monitoring images, it triggers foreign object intrusion detection, calls the trained foreign object intrusion detection model to perform foreign object intrusion detection on the real-time monitoring images, and obtains the foreign object intrusion detection results.

[0068] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0069] In one embodiment, such as Figure 2 As shown, a method for constructing an object detection model is provided, which can be applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0070] Step S100: Obtain the initial object detection training set, which includes a first sample set carrying labeled information and a second sample set without labeled information.

[0071] The initial object detection training set (hereinafter referred to as the initial training set) is a raw, unoptimized training set used for object detection. Specifically, the initial object detection training set includes a first sample set carrying annotation information (label data) and a second sample set without annotation information (label data). Typically, the samples in the first sample set contain target objects and their annotation information; therefore, the first sample set can be considered a positive sample set. The second sample set does not carry annotation data, but this does not mean that all samples in it are negative; it may contain target objects or not, and therefore can be considered a pseudo-negative sample set. In this embodiment, the samples in the first and second sample sets can be images or videos. The annotation information specifically includes the location of the target and the confidence level, where the confidence level can be considered as the probability of the target object belonging to a certain category.

[0072] In practice, the server may respond to a training set optimization message sent by the terminal, randomly selecting a portion of samples from a pre-stored dataset containing labeled information as the first sample set, and randomly selecting the same number of samples from a pre-stored dataset without labeled information as the second sample set. Taking substation foreign object detection as an example, the first sample set may consist of substation images containing intruding objects and labeled with information manually marked by inspection personnel. The labeled information specifically includes the location of the intruding foreign object and its confidence level, and can be recorded in a labeling file. The second sample set consists of substation images without labeled information. Specifically, the images may have a background similar to the foreign object or contain images of intruding foreign objects that have been omitted from the labeling. In practical applications, since the labeling information is done manually, labelers are often affected by background interference and other factors, resulting in the problem of forgetting to add labels to images containing target objects (referred to as "missed labels" in the industry). Therefore, the number of samples in the second sample set is usually relatively large.

[0073] Step S200: Train the pre-trained target detection network based on the initial target detection training set to obtain the target detector.

[0074] An object detector can be viewed as an object classifier used to identify and locate different categories of target objects in images or videos. It typically uses deep learning algorithms, such as convolutional neural networks, to achieve object detection through multi-level feature extraction and classification of input data. In this embodiment, after obtaining the initial object detection training set, a pre-trained object detection network, such as RetinaNet, can be trained based on the first and second sample sets to obtain the object detector. Specifically, the training process can be as follows:

[0075] (1) Data preparation: Image enhancement and other processing are performed on the image or video data in the initial target detection training set.

[0076] (2) Feature extraction: Use a pre-trained target detection network to extract features from the image or video data in the initial target detection training set to obtain multi-level feature maps.

[0077] (3) Candidate box generation: Multiple candidate boxes are generated on the feature map by means of sliding windows or anchor points to represent regions that may contain target objects.

[0078] (4) Label matching: For each candidate box, assign corresponding category labels and location information according to the degree of overlap between it and the real target.

[0079] (5) Loss function calculation and backpropagation: The loss value is calculated based on the difference between the predicted and true values ​​of classification and regression, and the network parameters are updated through backpropagation.

[0080] (6) NMS (Non-Maximum Suppression) processing: Since the same target object may be detected by multiple candidate boxes, non-maximum suppression (NMS) processing is required to filter out the final target box.

[0081] It is understood that in other embodiments, the object detector may also be Faster R-CNN (Region-based Convolutional Neural Network) and SSD (Single Shot MultiBox Detector), etc.

[0082] Step S300: Call the target detector to perform target detection on the second sample set and obtain the target detection results, which include the confidence scores of each sample in the second sample set.

[0083] In practice, after obtaining the second sample set, a pre-trained object detector can be invoked to perform object detection on the samples in the second sample set, identifying different categories of target objects and determining the location of each category, the probability of belonging to that category (i.e., the confidence level), etc., to obtain the object detection results. Specifically, the object detection process of the object detector may include the following steps:

[0084] (1) Preprocessing: Preprocess the images or videos in the second sample set, such as scaling, cropping, normalization, etc., to facilitate subsequent processing.

[0085] (2) Feature extraction: Deep learning algorithms such as convolutional neural networks (CNN) are used to extract features from the processed images or videos to obtain multi-level feature maps.

[0086] (3) Candidate box generation: Multiple candidate boxes are generated on the feature map by means of sliding windows or anchor points to represent regions that may contain target objects.

[0087] (4) Feature classification: For each candidate box, extract its feature vector and use a classifier to classify it to determine whether it contains the target object.

[0088] (5) Bounding box regression: For candidate boxes that are classified as containing the target object, the position and size of the regressor are further adjusted to more accurately locate the bounding box of the target object.

[0089] (6) NMS processing: Since the same target object may be detected by multiple candidate boxes, non-maximum suppression (NMS) processing is required to filter out the final target box.

[0090] (7) Post-process the target detection results according to application requirements, such as identifying the target category.

[0091] Step S400: Select positive sample prediction sets with confidence levels higher than a preset first confidence level threshold and negative sample prediction sets with confidence levels lower than a preset second confidence level threshold from the second sample set.

[0092] Following the previous embodiment, after obtaining the target detection results for each sample in the second sample set, corresponding annotation information is added to each sample. At this point, a portion of the samples in the second sample set can be selected based on the confidence level of each sample. Specifically, samples with high confidence levels can be predicted as positive samples. For example, a second sample with a confidence level higher than a preset first confidence threshold can be determined as a predicted positive sample. In this embodiment, the first preset confidence threshold can be 0.7, and the predicted positive sample is the sample that is predicted as a positive sample. Simultaneously, samples with low confidence levels can be predicted as negative samples. For example, a second sample with a confidence level lower than a preset second confidence threshold can be determined as a predicted negative sample. In this embodiment, the preset second confidence threshold can be 0.2, and the predicted negative sample is the sample that is predicted as a negative sample. Subsequently, the predicted positive samples are aggregated to obtain a positive sample prediction set, and the predicted negative samples are aggregated to obtain a negative sample prediction set. It is understood that the first confidence threshold and the second confidence threshold can also be other values, depending on the actual situation.

[0093] Step S500: Update the initial target detection training set based on the positive sample prediction set and the negative sample prediction set, return to step S200, and continue until the preset iteration termination condition is met to enter step S600.

[0094] The preset iteration termination condition may include a preset number of iterations, or the number of positive sample prediction sets obtained after multiple iterations and reaching a preset number. In this embodiment, unlike the traditional PU-bagging method, which only adds unlabeled samples predicted as positive samples with confidence scores higher than a certain threshold to the labeled dataset, after obtaining the positive and negative sample prediction sets, the positive and negative sample prediction sets can be added to the first sample set. Then, the updated first sample set and the original second sample set are merged to obtain the updated initial object detection training set.

[0095] Specifically, the process can involve adding positive and negative sample prediction sets to the first sample set, adding the label data of predicted positive samples from the positive sample prediction set and the label data of predicted negative samples from the negative sample prediction set to the corresponding label data or annotation file of the first sample set to update the first sample set. Then, the updated first sample set and the original second sample set are merged to update the initial object detection training set. The updated initial object detection set is then used as the training set for the next object detector training, and the object detector is trained again. Based on the trained object detector, object detection is performed on the unlabeled second sample set to obtain the label data for the samples in the second sample set. Steps S200-S500 are repeated to continuously use the newly trained, more accurate object detector to perform object detection on the unlabeled second sample set until a preset iteration termination condition is reached, at which point the iteration stops. At this point, the label data for the second sample set will be more accurate, and the quality of the updated initial object detection training set will also be higher. It should be noted that since the predicted negative samples carrying labeled information are used in the training of the target detector, and the predicted negative samples include not only real negative samples but also data that are actually positive samples, during the target detection process, the data that detects positive samples in the unlabeled second sample set is more likely to be positive samples.

[0096] Step S600: Train the pre-trained object detection network based on the latest initial object detection training set to obtain the object detection model.

[0097] After the iteration terminates, the latest initial object detection training set can be used to determine the final object detection training set. This training set is then used again as the training set for the pre-trained object detection network to train the network and obtain the final accurate object detection model. Specifically, the model training process can be found in the training methods described in the above embodiments, and will not be repeated here.

[0098] In the above object detection model construction method, the optimization of the initial object detection training set using an unlabeled second sample set (unlabeled data) helps the object detector better capture the data distribution, improves the stability and robustness of the object detector, and differs from the traditional approach of only adding a second sample subset (positive sample data) with a confidence level greater than a preset first confidence threshold to the initial object detection training set to update the initial object detection training set. This method continuously optimizes and updates the initial object detection set based on the positive and negative sample prediction sets, which can effectively address the problem of class imbalance in the training set, improve the accuracy of the object detector, and thus make the labeled data in the object detection training set obtained through iterative optimization of the object detector more accurate. Finally, the object detection model is trained and constructed based on the object detection training set, which can make the performance and accuracy of the object detection model higher, so as to support more accurate object detection schemes.

[0099] like Figure 3 As shown, in one embodiment, step S200 includes:

[0100] Step S220: Use the first sample set to train the pre-trained target detection network to obtain the initial target detector.

[0101] Step S240: Use the initial target detector to perform target detection on the second sample set to obtain the initial target detection result. The initial target detection result includes the second sample set carrying annotation information, including confidence level.

[0102] Step S260: Select a set of pseudo-negative samples with a confidence level higher than a preset third confidence level threshold from the second sample set carrying labeled information. The preset third confidence level threshold is lower than the preset second confidence level threshold.

[0103] Step S280: Based on the first sample set and the pseudo-negative sample set, train the initial target detector to obtain the target detector.

[0104] The pseudo-negative sample set is a sample set that carries coarse annotation information and can be approximated as negative samples (though not necessarily actually negative samples). In this embodiment, unlike the approach of directly training the target detector using the first sample set and the second sample set without annotation information, to improve the accuracy of the target detector, preliminary target detection can be performed on the second sample set without annotation information to roughly obtain the annotation information, i.e., label data, for each sample in the second sample set. Then, the target detector is trained using the first sample set with annotation information and the second sample set with annotation information.

[0105] In specific implementation, the pre-trained target detection network can be trained first using a first sample set (i.e., a positive sample set) to obtain an initial target detector. Then, the initial target detector is used to detect targets on a second sample set to obtain initial target detection results. These results include the annotation information of the target objects in each sample of the second sample set, including their location and confidence level. At this point, all samples in the second sample set carry coarse annotation information. Then, samples in the second sample set with a confidence level higher than a preset third confidence threshold are selected. These samples may be identified as containing target objects, and the selected samples can be identified as pseudo-negative samples carrying annotation information. Pseudo-negative samples are then aggregated to obtain a pseudo-negative sample set. In this embodiment, the preset third confidence threshold can be 0.01. Then, the initial target detector is further trained based on the first sample set and the pseudo-negative sample set to obtain the target detector. The specific training process can refer to the target detector training process described in the above embodiment, and will not be repeated here. In this embodiment, an initial object detector is first trained using a first sample set. Then, the initial object detector is further trained and optimized using a pseudo-negative sample set carrying coarse labeling information and the first sample set, thereby improving the accuracy and classification ability of the object detector. It is understood that the third confidence threshold can also be other values, depending on the specific circumstances.

[0106] like Figure 4 As shown, in one embodiment, before step S500, the method further includes:

[0107] Step S420: Based on the confidence level of each sample in the second sample set, determine the score of each predicted positive sample in the positive sample prediction set and the score of each predicted negative sample in the negative sample prediction set, so as to update the positive sample prediction set and the negative sample prediction set.

[0108] In this embodiment, the sample score can be determined based on the sample's confidence level, and the sample score can be understood as the sample's weight value. After each predicted positive sample and predicted negative sample is selected, the score of each sample can be determined based on the sample's confidence level to update the positive sample prediction set and the negative sample prediction set. Specifically, if the sample's confidence level is represented by p, and if the probability of a sample being predicted as a positive sample is p, then the score of the predicted positive sample can be determined as p, and the score of the predicted negative sample can be determined as 1-p. In other embodiments, if the probability of a sample being predicted as a positive sample is p, then the score of the predicted positive sample can be determined as log(p / (1-p)), where log is the natural logarithm, used to convert the probability value into a score. The score of the predicted negative sample can be 1-log(p / (1-p)). It is important to note that when p = 0 or p = 1, log(p / (1-p)) will become negative infinity or positive infinity. To avoid this, p can be restricted to a small interval, such as [epsilon, 1-epsilon], where epsilon is a small positive number close to 0. It is understood that in other embodiments, different methods can be used to determine the sample score based on confidence. In this embodiment, the sample score is determined by the sample's confidence, which more accurately assesses the importance and reliability of each unlabeled sample, thereby allowing for targeted selection of samples for training and improving model performance. Furthermore, the sample score can better reflect its contribution to the object detection task, enabling the model to focus more on samples with higher scores that are more likely to contain the target object, thus improving the recall and precision of the detection.

[0109] like Figure 5 As shown, in one embodiment, before step S500, the method further includes: step S420, determining the score of each predicted positive sample in the positive sample prediction set and the score of each predicted negative sample in the negative sample prediction set based on the confidence level.

[0110] The method also includes:

[0111] Step S700: Based on the scores of the predicted positive samples determined in each iteration, determine the average score of the predicted positive samples and use the average score of the predicted positive samples as the weight of the predicted positive samples; and based on the scores of the predicted negative samples determined in each iteration, determine the average score of the predicted negative samples and use the average score of the predicted negative samples as the weight of the predicted negative samples.

[0112] Step S720: Based on the weights of the predicted positive samples and the weights of the predicted negative samples, recalculate the loss function of the object detection model and update the object detection model.

[0113] In this embodiment, after selecting the positive and negative sample prediction sets during each iteration, the sample score can be determined based on the sample's confidence level. The specific score determination method can be found in the previous embodiment and will not be repeated here. Following this method, after multiple iterations are completed and the iteration stops, each sample records the score determined by the target detector during each iteration. Then, based on the predicted positive sample scores determined in each iteration, the average score of the predicted positive samples is determined, and this average score is used as the weight for the predicted positive samples. Similarly, based on the predicted negative sample scores determined in each iteration, the average score of the predicted negative samples is determined, and this average score is used as the weight for the predicted negative samples. Based on the weights of the predicted positive and negative samples, the loss function of the target detection model is recalculated, and the target detection model is updated.

[0114] Specifically, the loss function of the object detection model can be Focal Loss. In this embodiment, different Focal Losses can be used in the object detection model for confidence and category prediction, respectively. It should be noted that in the process of calculating the loss function, the loss function of all added positive and negative samples needs to be multiplied by their corresponding scores p or (1-p), while the score of the positive samples in the original first sample set is 1.

[0115] In this embodiment, by using sample weighting, the score corresponding to the sample is used as its weight, the loss function is recalculated, and the object detection model is updated. This allows the object detection model to pay more attention to those samples with higher scores, which are more likely to be real positive samples, thereby improving the recall and accuracy of the object detection model.

[0116] In one embodiment, such as Figure 6 As shown, a foreign object intrusion detection model method for substations is provided, which is then applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0117] Step S800: Obtain real-time monitoring images of the substation.

[0118] The real-time monitoring images of the substation contain substation environmental information. In this embodiment, the real-time monitoring images can be multiple monitoring images or multiple monitoring images extracted from real-time monitoring video data. Generally, substation environments often contain foreign objects such as oil stains, weeds, leaves, and dust. These foreign objects may affect the heat dissipation of power equipment, leading to excessively high equipment temperatures, thereby affecting the operational stability and lifespan of the equipment. They may also cause insulation breakdown, resulting in accidents such as short circuits and explosions, causing power grid failures and power outages. In addition, they may block the ventilation holes and heat sinks of the equipment, preventing the equipment from dissipating heat properly, and thus causing problems such as overheating and burnout. Therefore, this application adopts an intelligent inspection method to detect the presence of foreign objects in the substation. Specifically, maintenance personnel can collect historical monitoring images of the substation, manually annotate some of these images, and then construct an initial foreign object intrusion detection training set based on both annotated and unannotated historical monitoring images. Then, based on this initial training set, a foreign object intrusion detection model is constructed using the target detection model construction method described in the above embodiments. This model is then deployed on a server, and the server's foreign object intrusion triggering conditions are set. For example, the server can trigger foreign object intrusion detection when it receives real-time monitoring images of the substation uploaded by a terminal.

[0119] In practice, the substation can be captured in real time using image acquisition devices such as cameras. The cameras can then send the captured monitoring images to the terminal in real time, and the terminal can then simultaneously upload the received real-time monitoring images of the substation to the server.

[0120] Step S820: Call the trained foreign object intrusion detection model to perform foreign object intrusion detection on the real-time monitoring image and obtain the foreign object intrusion detection result.

[0121] In this embodiment, the foreign object intrusion detection model can be based on an initial foreign object intrusion detection training set. The target detection model construction method described in the previous embodiment is used to train and test the YOLOv5 (You Only Look Once version 5) model to obtain the foreign object intrusion detection model. The YOLOv5 model is a real-time target detection algorithm based on deep learning that can detect multiple targets in an image simultaneously and output their category and location information. In this embodiment, YOLOv5 uses a new model architecture called CSPNet (CrossStage Parallel Network). This architecture utilizes cross-layer connections and channel separation to improve the model's performance and efficiency. The entire model consists of three parts: a backbone network, a detection head, and a loss function. Specifically, the YOLOv5 backbone network adopts the CSPNet structure, which consists of a preprocessing layer, a convolutional layer, a CSPDarknet53 (YOLO V3 backbone network) network, and two SPP layers (spatial pyramid pooling layers). The CSPDarknet53 network consists of several residual blocks, each internally employing a CSPNet structure, effectively improving the performance and efficiency of feature extraction. The YOLOv5 detector head comprises several convolutional layers and a final convolutional layer, used to generate the confidence score, class probability, and location information for each predicted bounding box. The output of the detector head is a three-dimensional tensor containing information from all predicted bounding boxes. YOLOv5 uses a novel loss function called Focal Loss, which effectively addresses the imbalance between positive and negative samples. The core idea of ​​Focal Loss is to increase the penalty for easily classified negative samples, thereby improving the model's ability to identify difficult samples. YOLOv5 uses two different Focal Losses, one for confidence score and the other for class prediction. It's also important to note that the loss functions for all additional positive and negative predictions need to be multiplied by the corresponding score p or (1-p), while the score for the original labeled positive samples is 1.

[0122] The specific model building process can involve collecting historical monitoring images of the substation, manually annotating some of these images with information including the category and location of the target object, and then, based on both annotated and unannotated historical monitoring images, employing the target detection training set optimization process described in the target detection model building method above, to obtain an initial foreign object intrusion detection training set and a foreign object intrusion test set. Next, a YOLOv5 model is built. This can be done using a pre-trained model for initialization or by rebuilding the YOLOv5 model from scratch for training. The model building process includes defining the model structure, selecting a loss function, and setting hyperparameters. The model training process involves selecting a suitable optimizer, such as Adam (Adaptive Moment Estimation) or SGD (Stochastic Gradient Descent), setting the learning rate and number of training iterations, and using the PyTorch deep learning framework for parameter optimization.

[0123] In practical applications, when the server acquires real-time monitoring images of the substation, it triggers the foreign object intrusion detection condition and enters the foreign object intrusion detection program. This program can call a pre-trained foreign object intrusion detection model to perform foreign object intrusion detection on the real-time monitoring images and obtain the foreign object intrusion detection result. The specific foreign object intrusion detection process can be found in the target detection process of the target detector in the above embodiment, and will not be repeated here.

[0124] In the above-mentioned method for detecting foreign objects in substations, the foreign object intrusion detection model constructed by the above-mentioned target detection model construction method can make the foreign object intrusion detection model more powerful and have stronger generalization ability, higher model accuracy and robustness. Accordingly, by using this foreign object intrusion detection model to perform foreign object intrusion detection on real-time monitoring images of substations, foreign objects intruding inside the substation can be quickly and accurately identified, and potential safety hazards can be detected in a timely manner.

[0125] like Figure 7 As shown, in one embodiment, after obtaining the foreign object intrusion detection result, the method further includes: step S840, if the foreign object intrusion detection result indicates that there is an intruding foreign object in the real-time monitoring image, a warning message is sent.

[0126] Following the above embodiments, if the foreign object intrusion detection model outputs a foreign object intrusion detection result indicating the presence of an intruding foreign object in the current real-time monitoring image, an early warning message is sent to the terminal; otherwise, foreign object intrusion detection is performed on the next real-time monitoring image. In this embodiment, by sending an early warning message when an intruding foreign object is detected, relevant personnel can be promptly notified to handle the situation, thereby enhancing the safe and stable operation of the substation.

[0127] In another embodiment, the method further includes: receiving foreign object intrusion detection result feedback data sent by the terminal, and updating the foreign object intrusion detection model based on the foreign object intrusion detection result feedback data.

[0128] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0129] Based on the same inventive concept, this application also provides a target detection model construction device for implementing the target detection model construction method described above, and a substation foreign object intrusion detection device for implementing the substation foreign object intrusion detection method described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more target detection model construction device embodiments provided below can be found in the limitations of the target detection model construction method above, and the specific limitations of multiple substation foreign object intrusion detection device embodiments can be found in the limitations of the substation foreign object intrusion detection method above, and will not be repeated here.

[0130] In one embodiment, such as Figure 8 As shown, an object detection model building device is provided, including: an initial training set acquisition module 810, an object detector training module 820, an object detection module 830, a data filtering module 840, a training set update module 850, and a model building module 860, wherein:

[0131] The initial training set acquisition module 810 is used to acquire the initial object detection training set, which includes a first sample set carrying labeled information and a second sample set without labeled information.

[0132] The object detector training module 820 is used to train the pre-trained object detection network based on the initial object detection training set to obtain the object detector;

[0133] The target detection module 830 is used to call the target detector to perform target detection on the second sample set and obtain the target detection results, which include the confidence scores of each sample in the second sample set.

[0134] The data filtering module 840 is used to filter out positive sample prediction sets with confidence levels higher than a preset first confidence threshold and negative sample prediction sets with confidence levels lower than a preset second confidence threshold from the second sample set. The preset first confidence threshold is greater than the preset second confidence threshold.

[0135] The training set update module 850 is used to update the initial target detection training set based on the positive sample prediction set and the negative sample prediction set, and to re-control the target detector training module to train the pre-trained target detection network based on the initial target detection training set to obtain the target detector operation until the preset iteration termination condition is reached; and to determine the target detection training set.

[0136] The model building module 860 is used to train the pre-trained object detection network based on the latest initial object detection training set to obtain an object detection model.

[0137] In the aforementioned object detection model construction device, the initial object detection training set is optimized using an unlabeled second sample set (unlabeled data). This helps the object detector better capture data distribution, improving its stability and robustness. Furthermore, unlike the traditional approach of only adding a subset of second samples (positive sample data) with a confidence level greater than a preset first confidence threshold to the initial object detection training set to update it, this device continuously optimizes and updates the initial object detection set based on both positive and negative sample prediction sets. This effectively addresses the problem of class imbalance in the training set, improving the accuracy of the object detector. Consequently, the labeled data in the object detection training set obtained through iterative optimization of the object detector becomes more accurate. Finally, training and constructing the object detection model based on the training set results in higher performance and accuracy, supporting more precise object detection schemes.

[0138] In one embodiment, the target detector training module 820 is further configured to call the first sample set to train the pre-trained target detection network to obtain an initial target detector, use the initial target detector to perform target detection on the second sample set to obtain an initial target detection result, the initial target detection result including the second sample set carrying annotation information, the annotation information including confidence, select pseudo-negative sample sets from the second sample set carrying annotation information whose confidence is higher than a preset third confidence threshold, the preset third confidence threshold being lower than a preset second confidence threshold, and train the initial target detector based on the first sample set and the pseudo-negative sample set to obtain the target detector.

[0139] like Figure 9As shown, in one embodiment, the apparatus further includes a sample score determination module 842, which is used to determine the score of each predicted positive sample in the positive sample prediction set and the score of each predicted negative sample in the negative sample prediction set based on the confidence level of each sample in the second sample set, so as to update the positive sample prediction set and the negative sample prediction set.

[0140] like Figure 9 As shown, in one embodiment, the apparatus further includes: a model update module 870, configured to determine the scores of each predicted positive sample in the positive sample prediction set and the scores of each predicted negative sample in the negative sample prediction set based on the confidence level; determine the average score of the predicted positive samples based on the scores of the predicted positive samples determined in each iteration, and determine the average score of the predicted positive samples as the weight of the predicted positive samples; determine the average score of the predicted negative samples based on the scores of the predicted negative samples determined in each iteration, and determine the average score of the predicted negative samples as the weight of the predicted negative samples; recalculate the loss function of the target detection model based on the weights of the predicted positive samples and the weights of the predicted negative samples, and update the target detection model.

[0141] In one embodiment, such as Figure 10 As shown, this application also provides a substation foreign object intrusion detection device, including: a monitoring image acquisition module 910 and a foreign object intrusion detection module 920, wherein:

[0142] The monitoring image acquisition module 910 is used to acquire real-time monitoring images of the substation.

[0143] The foreign object intrusion detection module 920 is used to call the trained foreign object intrusion detection model to perform foreign object intrusion detection on real-time monitoring images and obtain foreign object intrusion detection results;

[0144] The foreign object intrusion detection model is constructed based on historical monitoring images of the substation, using the target detection model construction method described in any of the above embodiments.

[0145] like Figure 11 As shown, in one embodiment, the substation foreign object intrusion detection device further includes an early warning message sending module 940, which sends an early warning message if the foreign object intrusion detection result indicates that there is an intruding foreign object in the real-time monitoring image.

[0146] The modules in the aforementioned target detection model construction device and substation foreign object intrusion detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0147] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 12 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores initial target detection training sets or real-time monitoring images of substations, etc. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a target detection model construction method or a substation foreign object intrusion detection method.

[0148] Those skilled in the art will understand that Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0149] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the target detection model construction method or the substation foreign object intrusion detection method described above.

[0150] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps in the target detection model construction method or the substation foreign object intrusion detection method described above.

[0151] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the target detection model construction method or the substation foreign object intrusion detection method described above.

[0152] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0153] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0154] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for detecting foreign object intrusion in a substation, characterized in that, The method includes: An initial target detection training set is obtained, which includes a first sample set carrying labeled information and a second sample set without labeled information. The first sample set includes substation images containing intruding objects and carrying the location of the intruding objects and the confidence level of the intruding objects. The second sample set includes substation images without labeled information. The pre-trained target detection network is trained based on the initial target detection training set to obtain the target detector; The target detector is invoked to perform target detection on the second sample set to obtain target detection results, which include the confidence scores of each sample in the second sample set. From the second sample set, a positive sample prediction set with a confidence level higher than a preset first confidence level threshold and a negative sample prediction set with a confidence level lower than a preset second confidence level threshold are selected. The preset first confidence level threshold is greater than the preset second confidence level threshold. The initial target detection training set is updated based on the positive and negative sample prediction sets. The process of training the pre-trained target detection network based on the initial target detection training set to obtain a target detector continues until a preset iteration termination condition is met. The process of training the pre-trained target detection network based on the initial target detection training set to obtain a target detector includes: using the first sample set to train the pre-trained target detection network to obtain an initial target detector; using the initial target detector to perform target detection on the second sample set to obtain an initial target detection result, where the initial target detection result includes a second sample set carrying labeled information, the labeled information including confidence levels; selecting a pseudo-negative sample set from the second sample set carrying labeled information whose confidence levels are higher than a preset third confidence threshold, where the preset third confidence threshold is lower than a preset first confidence threshold; and training the initial target detector based on the first sample set and the pseudo-negative sample set to obtain the target detector. The pre-trained object detection network is trained using the latest initial object detection training set to obtain an object detection model; Acquire real-time monitoring images of the substation; The trained target detection model is invoked to perform foreign object intrusion detection on the real-time monitoring image, and the foreign object intrusion detection result is obtained.

2. The method according to claim 1, characterized in that, Before updating the initial object detection training set based on the positive sample prediction set and the negative sample prediction set, the method further includes: Based on the confidence level of each sample in the second sample set, the scores of each predicted positive sample in the positive sample prediction set and the scores of each predicted negative sample in the negative sample prediction set are determined to update the positive sample prediction set and the negative sample prediction set.

3. The method according to claim 1 or 2, characterized in that, Before updating the initial object detection training set based on the positive sample prediction set and the negative sample prediction set, the method further includes: Based on the confidence level, the scores of each predicted positive sample in the positive sample prediction set and the scores of each predicted negative sample in the negative sample prediction set are determined to update the positive sample prediction set and the negative sample prediction set. After obtaining the target detection model, the method further includes: Based on the scores of the predicted positive samples determined in each iteration, the average score of the predicted positive samples is determined, and the average score of the predicted positive samples is used as the weight of the predicted positive samples. Similarly, based on the scores of the predicted negative samples determined in each iteration, the average score of the predicted negative samples is determined, and the average score of the predicted negative samples is used as the weight of the predicted negative samples. Based on the weights of the predicted positive samples and the weights of the predicted negative samples, the loss function of the target detection model is recalculated, and the target detection model is updated.

4. The method according to claim 1, characterized in that, After obtaining the foreign object intrusion detection result, the method further includes: If the foreign object intrusion detection result indicates that there is an intruding foreign object in the real-time monitoring image, an early warning message is sent.

5. The method according to claim 1, characterized in that, The method further includes: The system receives feedback data on foreign object intrusion detection results sent by the receiving terminal and updates the target detection model based on the feedback data.

6. A foreign object intrusion detection device, characterized in that, The device includes: The monitoring image acquisition module is used to acquire real-time monitoring images of the substation. The foreign object intrusion detection module is used to call the trained target detection model to perform foreign object intrusion detection on the real-time monitoring image and obtain the foreign object intrusion detection result. The device is further configured to: acquire an initial target detection training set, the initial target detection training set including a first sample set carrying labeled information and a second sample set not carrying labeled information, the first sample set including substation images containing intruding objects and carrying information on the location of the intruding objects and the confidence level of the intruding objects, the second sample set including substation images not carrying labeled information; train a pre-trained target detection network based on the initial target detection training set to obtain a target detector; call the target detector to perform target detection on the second sample set to obtain a target detection result, the target detection result including the values ​​of each sample in the second sample set. The confidence level is determined by selecting positive sample prediction sets with confidence levels higher than a preset first confidence threshold and negative sample prediction sets with confidence levels lower than a preset second confidence threshold from the second sample set. The preset first confidence threshold is greater than the preset second confidence threshold. Based on the positive sample prediction sets and the negative sample prediction sets, the initial target detection training set is updated. The process of training the pre-trained target detection network based on the initial target detection training set to obtain the target detector is repeated until a preset iteration termination condition is reached. Finally, the pre-trained target detection network is trained based on the latest initial target detection training set to obtain the target detection model. The process of training a pre-trained target detection network based on an initial target detection training set to obtain a target detector includes: training the pre-trained target detection network using the first sample set to obtain an initial target detector; using the initial target detector to perform target detection on the second sample set to obtain an initial target detection result, wherein the initial target detection result includes a second sample set carrying annotation information, the annotation information including confidence level; selecting a pseudo-negative sample set from the second sample set carrying annotation information whose confidence level is higher than a preset third confidence level threshold, wherein the preset third confidence level threshold is lower than a preset first confidence level threshold; and training the initial target detector based on the first sample set and the pseudo-negative sample set to obtain the target detector.

7. The apparatus according to claim 6, characterized in that, The device also includes a warning message sending module, which sends a warning message if the foreign object intrusion detection result indicates that there is an intruding foreign object in the real-time monitoring image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images