Abnormal state identification method of air shaft explosion-proof cover
Through image acquisition and feature vector analysis combined with target state classification model, the accuracy of the judgment of abnormal state of the explosion-proof cover of the air shaft is solved, and a more reliable safety warning is achieved.
Patent Information
- Application Number
- CN202411751356.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-14
AI Technical Summary
It is difficult for the prior art to accurately determine whether the explosion-proof cover of the air shaft is in an abnormal state, resulting in potential safety hazards.
Through image acquisition, Gaussian background modeling, feature vector extraction and cosine value calculation, combined with the trained target state classification model, a dual recognition mechanism is used to judge the abnormal state of the explosion-proof cover of the air shaft.
It realizes a more accurate judgment of whether the air shaft explosion-proof cover is in an abnormal state, and promptly issue early warnings to avoid potential safety hazards, and effectively filter false alarm information, improving the reliability of alarm information.
Smart Images
Figure CN119625416A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of coal mine production safety, and in particular to a method for identifying an abnormal state of a ventilation shaft explosion-proof cover. Background Art
[0002] The main function of the wind shaft explosion-proof cover is to protect the ventilator and reduce the airflow pressure in the mine to prevent damage to equipment and personnel in the event of an explosion. Under normal ventilation conditions, the wind shaft explosion-proof cover can isolate the ground atmosphere from the underground air, maintain air tightness, prevent airflow short circuit, and ensure the normal operation of the ventilation system. Once an explosion occurs underground, the explosion-proof cover can withstand the pressure of the explosion wave and release the pressure by deforming or opening, thereby protecting the ventilator from damage. Pressure relief also helps to balance the pressure inside and outside the mine, preventing the explosion airflow from causing greater damage to the equipment and personnel inside the mine.
[0003] When the main ventilation fan in the mine stops running, the explosion-proof cover of the air shaft can be opened to use natural wind pressure for ventilation, providing fresh air flow for the mine to meet the needs of underground personnel. During normal ventilation, the explosion-proof cover can maintain airtightness, prevent air flow short circuit, and ensure the effectiveness of the ventilation system.
[0004] By monitoring the abnormal status of the air shaft explosion-proof cover on the ground, the safe production of coal mines can be guaranteed. By timely discovering and handling potential problems, the reliability and safety of the ventilation system can be ensured, providing strong protection for the sustainable development of coal mining enterprises and the life safety of employees. Summary of the invention
[0005] The purpose of this application is to solve one of the technical problems in the related art at least to some extent.
[0006] To this end, the first purpose of the present application is to propose a method for identifying the abnormal state of an air shaft explosion-proof cover, so as to more accurately determine whether the air shaft explosion-proof cover is in an abnormal state, thereby issuing an early warning in time to avoid potential safety hazards.
[0007] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a method for identifying an abnormal state of an explosion-proof cover of a wind shaft, comprising: performing image acquisition on the explosion-proof cover of the wind shaft to obtain an image of the explosion-proof cover of the wind shaft, and performing Gaussian background modeling based on an area of interest in the image to determine a background image and a foreground image of the explosion-proof cover of the wind shaft; extracting a first feature vector of the explosion-proof cover of the wind shaft from the background image based on a first area of interest in the background image and a second area of interest in the foreground image, and extracting a second feature vector of the explosion-proof cover of the wind shaft from the foreground image; determining a cosine value of a feature vector based on the first feature vector and the second feature vector, and determining a first recognition state of the explosion-proof cover of the wind shaft based on the cosine value; determining a second recognition state of the explosion-proof cover of the wind shaft based on a trained target state classification model, the first area of interest and the second area of interest, wherein each of the first recognition state and the second recognition state is that the explosion-proof cover of the wind shaft is in a normal state, or that the explosion-proof cover of the wind shaft is in an abnormal state; in response to the first recognition state and the second recognition state being that the explosion-proof cover of the wind shaft is in an abnormal state, determining that the target state of the explosion-proof cover of the wind shaft is an abnormal state.
[0008] The abnormal state recognition method of the wind shaft explosion-proof cover provided in the present application is to collect an image of the wind shaft explosion-proof cover, and determine the background image and the foreground image from the image, so as to determine the feature vector of the wind shaft explosion-proof cover according to the area of interest in the background image and the foreground image. The first recognition state of the wind shaft explosion-proof cover is determined according to the cosine value between the feature vectors, and the second recognition state of the wind shaft explosion-proof cover is determined according to the trained target state classification model, so as to jointly determine the target state of the wind shaft explosion-proof cover according to the first recognition state and the second recognition state. Through the dual recognition mechanism, it is possible to more accurately judge whether the wind shaft explosion-proof cover is in an abnormal state, so as to issue an early warning in time to avoid potential safety hazards. At the same time, an alarm is only issued when both the first recognition state and the second recognition state identify an abnormality, which effectively filters out a large amount of false alarm information, thereby improving the reliability of the alarm information.
[0009] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0011] Figure 1 A schematic flow chart of a method for identifying an abnormal state of an explosion-proof cover of a wind shaft provided in an embodiment of the present application;
[0012] Figure 2A flowchart of a process of training a target state classification model in a method for identifying an abnormal state of an air shaft explosion-proof cover provided in an embodiment of the present application;
[0013] Figure 3 A schematic diagram of a twin network provided in an embodiment of the present application;
[0014] Figure 4 A schematic flow chart of another abnormal state identification method of an air shaft explosion-proof cover provided in an embodiment of the present application;
[0015] Figure 5 A schematic structural diagram of an abnormal state identification device for an air shaft explosion-proof cover provided in an embodiment of the present application. DETAILED DESCRIPTION
[0016] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0017] The following describes the abnormal state identification method and device of the wind shaft explosion-proof cover according to the embodiment of the present application with reference to the accompanying drawings.
[0018] Figure 1 is a flow chart showing a method for identifying an abnormal state of a wind shaft explosion-proof cover according to an exemplary embodiment. Figure 1 As shown, the abnormal state identification method of the wind shaft explosion-proof cover of the embodiment of the present application includes but is not limited to the following steps:
[0019] S101, performing image acquisition on the wind shaft explosion-proof cover to obtain an image of the wind shaft explosion-proof cover, and performing Gaussian background modeling based on a region of interest in the image to determine a background image and a foreground image of the wind shaft explosion-proof cover.
[0020] It should be noted that the execution subject of the abnormal state identification method of the wind shaft explosion-proof cover provided in the embodiment of the present application is an electronic device, and the electronic device can be a terminal device. Optionally, the terminal device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and the non-mobile electronic device can be a personal computer (personal computer, PC), a television, etc. The embodiment of the present application is not specifically limited.
[0021] In some implementations, an image acquisition device may be installed around the wind shaft explosion-proof cover to acquire images of the wind shaft explosion-proof cover. For example, a camera may be installed opposite the wind shaft explosion-proof cover, such as 30 meters away from the wind shaft explosion-proof cover and 5 meters above the ground, and the camera may acquire images of the area including the wind shaft explosion-proof cover to obtain images of the wind shaft explosion-proof cover.
[0022] Further, the region of interest is determined from the collected image of the wind shaft explosion-proof cover. Optionally, the position of the wind shaft explosion-proof cover in the image can be identified, and the region of interest is determined based on the position, that is, the region of interest includes the wind shaft explosion-proof cover.
[0023] Further, Gaussian background modeling is performed on the region of interest to obtain background and foreground images of the wind shaft explosion-proof cover. Optionally, multiple frames of images of the wind shaft explosion-proof cover can be collected, and Gaussian background modeling is performed based on the region of interest in the multiple frames of images to obtain accurate foreground and background images.
[0024] S102, based on the first region of interest in the background image and the second region of interest in the foreground image, extracting a first feature vector of the wind shaft explosion-proof cover from the background image, and extracting a second feature vector of the wind shaft explosion-proof cover from the foreground image.
[0025] In some implementations, the first feature vector can be extracted from the first region of interest and the second feature vector can be extracted from the second region of interest by using an image matching algorithm using a normalized cross-correlation coefficient. Alternatively, the first feature vector and the second feature vector can be obtained by matching the mean and variance of the image.
[0026] Optionally, the region of interest in the image can be divided into multiple neighborhoods, and the mean of the pixels can be determined based on the pixel information of each neighborhood, and the variance can be solved based on the mean, and the variance can be used as a feature vector. The feature vector has multiple dimensions. That is, the first feature vector is obtained by solving the pixel variance of the first region of interest, and the second feature vector is obtained by solving the pixel variance of the second region of interest.
[0027] S103, determining a cosine value of the eigenvector based on the first eigenvector and the second eigenvector, and determining a first recognition state of the air shaft explosion-proof cover based on the cosine value.
[0028] In some implementations, since the feature vector has multiple dimensions, the feature vector may be normalized first, and the cosine value may be determined based on the normalized feature vector. Optionally, the first feature vector and the second feature vector are normalized, and the cosine value between the first feature vector and the second feature vector is calculated based on the normalized first feature vector and the second feature vector.
[0029] Optionally, the cosine value may be calculated by obtaining the dot product of the normalized first eigenvector and the second eigenvector, and the modulus, that is, the length, of the first eigenvector and the second eigenvector.
[0030] Furthermore, the state of the wind shaft explosion-proof cover can be identified based on the cosine value to obtain a first identification state. By determining the cosine threshold, if the cosine value is greater than the cosine threshold, the first identification state is determined to be that the wind shaft explosion-proof cover is in a normal state; if the cosine value is less than the cosine threshold, the first identification state is determined to be that the wind shaft explosion-proof cover is in an abnormal state.
[0031] S104, based on the trained target state classification model, the first region of interest and the second region of interest, determine the second recognition state of the wind shaft explosion-proof cover, wherein each of the first recognition state and the second recognition state is that the wind shaft explosion-proof cover is in a normal state, or the wind shaft explosion-proof cover is in an abnormal state.
[0032] It is understandable that when there is dim light at night, sudden changes in light during the day, or flashing of car lights, noise will be generated in the collected images, thus affecting the accuracy of image recognition. The state of the wind shaft explosion-proof cover can be further identified based on the state classification model of the twin neural network structure to determine the target state of the wind shaft explosion-proof cover.
[0033] In some implementations, a target state classification model is pre-trained, and the first region of interest and the second region of interest are input into the target state classification model. The twin neural network in the target state classification model classifies the state of the wind shaft explosion-proof cover based on the first region of interest and the second region of interest, so as to classify the state of the wind shaft explosion-proof cover into an abnormal state or a normal state, and thus a second recognition state can be obtained.
[0034] Optionally, the target state classification model can determine the second identification state by determining a first confidence level that the wind shaft explosion-proof cover is in a normal state and a second confidence level that the wind shaft explosion-proof cover is in an abnormal state, and by comparing the confidence levels. That is, if the first confidence level is greater than the second confidence level, it can be determined that the second identification state is that the wind shaft explosion-proof cover is in a normal state; if the first confidence level is less than the second confidence level, it can be determined that the second identification state is that the wind shaft explosion-proof cover is in an abnormal state.
[0035] S105, in response to the first identification state and the second identification state both indicating that the wind shaft explosion-proof cover is in an abnormal state, determining that the target state of the wind shaft explosion-proof cover is an abnormal state.
[0036] In some implementations, after the first identification state and the second identification state are determined, the target state of the wind shaft explosion-proof cover can be determined according to the first identification state and the second identification state. That is, if both the first identification state and the second identification state indicate that the wind shaft explosion-proof cover is in an abnormal state, the target state of the wind shaft explosion-proof cover is determined to be an abnormal state.
[0037] In some implementations, when it is determined that the ventilation shaft explosion-proof cover is in an abnormal state, an early warning message can be generated to alert coal mine workers that there is an abnormality in the ventilation shaft explosion-proof cover, and measures can be taken quickly to repair or replace it to prevent major accidents such as explosions and fires caused by failure of the explosion-proof cover, thereby ensuring the safety of mine personnel and equipment.
[0038] Optionally, if one of the first identification state and the second identification state is that the air shaft explosion-proof cover is in an abnormal state, the target state of the air shaft explosion-proof cover is determined to be a normal state; if both the first identification state and the second identification state are that the air shaft explosion-proof cover is in a normal state, the target state of the air shaft explosion-proof cover is determined to be a normal state.
[0039] In the abnormal state identification method of the wind shaft explosion-proof cover provided in the embodiment of the present application, by collecting an image of the wind shaft explosion-proof cover, and determining the background image and the foreground image from the image, the feature vector of the wind shaft explosion-proof cover is determined according to the area of interest in the background image and the foreground image. The first identification state of the wind shaft explosion-proof cover is determined according to the cosine value between the feature vectors, and the second identification state of the wind shaft explosion-proof cover is determined according to the trained target state classification model, so as to jointly determine the target state of the wind shaft explosion-proof cover according to the first identification state and the second identification state. Through the dual identification mechanism, it is possible to more accurately determine whether the wind shaft explosion-proof cover is in an abnormal state, so as to issue an early warning in time to avoid potential safety hazards. At the same time, an alarm is only issued when both the first identification state and the second identification state identify an abnormality, which effectively filters out a large amount of false alarm information, thereby improving the reliability of the alarm information.
[0040] Based on the above embodiments, the present application embodiment can explain the process of training the target state classification model, such as Figure 2 As shown, the process of training the target state classification model proposed in the embodiment of the present application includes but is not limited to the following steps:
[0041] S201, obtaining a training data set, and determining an anchor point sample, a first positive sample, and a first negative sample from the training data set, wherein the training data set includes a normal state image and an abnormal state image of a wind shaft explosion-proof cover.
[0042] In some implementations, a training data set may be obtained from a database, and / or a user may generate a training data set manually or through artificial intelligence, wherein samples in the training data set include normal state images and abnormal state images of the wind shaft explosion-proof cover.
[0043] Optionally, the training data set can be divided into positive samples and negative samples, and two samples of the same category are determined to form a sample pair as a positive sample. For example, the positive sample is (X i ,X j ,1), where 1 represents X i and X j are samples of the same category. By determining two samples of different categories to form a sample pair as a negative sample, such as a negative sample (X i ,X j ,0), where 0 represents X i and X j are samples of different categories.
[0044] Optionally, a sample may be determined from the multiple positive samples and the multiple negative samples as an anchor sample, that is, the anchor sample may be a positive sample or a negative sample. Then, a first positive sample and a first negative sample may be determined from the multiple positive samples and the multiple negative samples.
[0045] S202, based on the anchor point sample, the first positive sample and the first negative sample, the initial first state classification model is trained until the sample data in the training data set is trained, and a second state classification model is obtained.
[0046] In some implementations, by inputting the anchor point sample, the first positive sample, and the first negative sample into the first state classification model, the first state classification model can calculate the loss function based on the anchor point sample, the first positive sample, and the first negative sample, and modify the model according to the loss function and continue training until the sample data in the training data set is trained, thereby obtaining the second state classification model.
[0047] In some implementations, since the first-state classification model is a twin network structure, the anchor sample, the first positive sample, and the first negative sample can be input into the same neural network in the initial first-state classification model to determine the first distance between the first positive sample and the anchor sample, and the second distance between the first negative sample and the anchor sample.
[0048] Optionally, the anchor sample, the first positive sample and the first negative sample are input into the same neural network, which extracts features from the anchor sample, the first positive sample and the first negative sample, and maps the extracted features to determine a first Euclidean distance between the first positive sample and the anchor sample, and the square of the first Euclidean distance is used as the first distance. The second Euclidean distance between the first negative sample and the anchor sample is determined, and the square of the second Euclidean distance is used as the second distance.
[0049] Optionally, the formula for calculating the first distance and the second distance is as follows:
[0050]
[0051]
[0052] Among them, d + represents the first distance, d - represents the second distance, x + represents the first positive sample, x a represents the anchor point sample, x - represents the first negative sample, f(x + ) indicates that the positive sample is mapped by the function, f(x - ) indicates that the negative sample is mapped by the function, f(x a ) indicates that the anchor point sample is mapped by the function.
[0053] like Figure 3 By inputting the anchor sample, the first positive sample and the first negative sample into the same neural network f, the neural network f extracts features from the anchor sample, the first positive sample and the first negative sample respectively, and maps the extracted features, we can get f(x + )、f(x - )、f(x a ), and then the first distance and the second distance can be calculated.
[0054] Furthermore, a first loss function of the first state classification model can be determined based on the first distance and the second distance, and the model can be corrected based on the first loss function and continued to be trained until the sample data in the training data set is trained to obtain a second state classification model.
[0055] Optionally, based on the distance between the positive sample and the anchor sample, and the distance between the negative sample and the anchor sample, the formula for calculating the loss function is as follows:
[0056] Loss(x a ,x + ,x - )=max{0,d + +ad -}(3)
[0057] Among them, Loss represents the loss function, d + Represents the distance between the positive sample and the anchor sample, d - represents the distance between the negative sample and the anchor sample, and a represents a hyperparameter greater than 0.
[0058] That is, the first distance and the second distance can be substituted into formula (3) to calculate the first loss function. If the difference between the second distance and the first distance is greater than a, the first loss function is 0. Otherwise, d + +ad - As the first loss function.
[0059] That is to say, only when the difference between the second distance and the first distance is less than a, the first loss function is used to correct the model and continue training until the sample data in the training data set is trained and the second state classification model is obtained.
[0060] S203, obtaining a second positive sample of the wind shaft explosion-proof cover, and determining a second negative sample based on the second positive sample and the second region of interest.
[0061] In some implementations, the training data set contains a large number of positive samples and negative samples. After the second state classification model is generated using the training data set, the second state classification model can be fine-tuned using on-site sample data to obtain a target state classification model.
[0062] Optionally, images of the wind shaft explosion-proof cover can be collected in real time as positive samples. That is, images of the wind shaft explosion-proof cover at different times can be collected as sample images, and based on the sample images, a second positive sample is formed. Optionally, images of different light intensities can be collected at different times as sample images. For example, images of the wind shaft explosion-proof cover during the day and images of the wind shaft explosion-proof cover at night can be collected as samples of the same type to form a second positive sample.
[0063] In some implementations, since the abnormal phenomenon of the wind shaft explosion-proof cover is rare, the second region of interest can be used to crop the second positive sample to obtain a second negative sample. For any sample image in the second positive sample, any sample image is intersected with the second region of interest to obtain an intersection ratio of any sample image and the region of interest.
[0064] Further, by determining a set intersection ratio threshold, comparing the intersection ratio with the set intersection threshold, and when the intersection ratio is less than the set intersection ratio threshold, cropping any sample image, a second negative sample can be obtained. That is, in response to the intersection ratio being less than the set intersection ratio threshold, cropping any sample image, obtaining an intersection image corresponding to any sample image, and generating a second negative sample based on the intersection image. For example, any region whose intersection ratio with the second region of interest is less than 0.2 can be cropped in any sample image as an intersection image.
[0065] S204: Training the second state classification model based on the anchor point sample, the second positive sample, and the second negative sample to obtain a target state classification model.
[0066] In some implementations, each state classification model is a twin network structure, and the anchor sample, the second positive sample, and the second negative sample can be input into the same neural network in the second state classification model to determine the third distance between the second positive sample and the anchor sample, and the fourth distance between the second negative sample and the anchor sample.
[0067] Optionally, the third distance and the fourth distance may be calculated based on the above formula (1) and formula (2).
[0068] Further, the second loss function of the second state classification model can be determined based on the third distance and the fourth distance, and the second state classification model can be adjusted based on the second loss function to obtain the target state classification model. Optionally, the second loss function can be calculated based on the above formula (3).
[0069] That is to say, when the second loss function is greater than 0, the second state classification model is adjusted to obtain the target state classification model. If the second loss function is 0, the second state classification model can be determined as the target state classification model.
[0070] In the abnormal state identification method of the wind shaft explosion-proof cover provided in the embodiment of the present application, a large-scale training data set is obtained, and the anchor point sample, the first positive sample and the first negative sample are determined from the training data set to train the first state classification model to obtain a second state classification model. Further, a small data sample is obtained on site, and a second positive sample and a second negative sample are determined to train the second state classification model to obtain a target state classification model. In the embodiment of the present application, the state classification model is a twin network structure, which can learn the subtle differences between normal and abnormal wind shaft explosion-proof covers, thereby improving the accuracy and efficiency of abnormal identification of wind shaft explosion-proof covers and improving the adaptability of the state classification model.
[0071] Figure 4 is a flow chart showing a method for identifying an abnormal state of a wind shaft explosion-proof cover according to an exemplary embodiment. Figure 4 As shown, the abnormal state identification method of the wind shaft explosion-proof cover of the embodiment of the present application includes but is not limited to the following steps:
[0072] S401, performing image acquisition on the wind shaft explosion-proof cover to obtain an image of the wind shaft explosion-proof cover, and performing Gaussian background modeling based on a region of interest in the image to determine a background image and a foreground image of the wind shaft explosion-proof cover.
[0073] In the embodiment of the present application, the implementation method of step S401 can be implemented by any method in the embodiments of the present application, which is not limited here and will not be repeated.
[0074] S402, based on the first region of interest in the background image and the second region of interest in the foreground image, extracting a first feature vector of the wind shaft explosion-proof cover from the background image, and extracting a second feature vector of the wind shaft explosion-proof cover from the foreground image.
[0075] In some implementations, for any image in the background image and the foreground image, pixel information of any image is obtained, and based on the pixel information, a region of interest corresponding to any image is divided into regions to obtain multiple neighborhoods.
[0076] Furthermore, based on the pixel information of each neighborhood, the pixel mean of each neighborhood is calculated, and based on the pixel mean, the pixel variance of each neighborhood is determined as the feature vector of any image. That is, when any image is a background image, the feature vector of any image is the first feature vector. When any image is a foreground image, the feature vector of any image is the second feature vector.
[0077] For example, taking the first region of interest as an example, the mean value of the 3*3 surrounding neighborhood of each pixel in the first region of interest may be calculated, and then the pixel variance may be calculated based on the mean value to obtain 9 first eigenvectors.
[0078] S403, determining a cosine value of the eigenvector based on the first eigenvector and the second eigenvector, and determining a first recognition state of the air shaft explosion-proof cover based on the cosine value.
[0079] In some implementations, since there are multiple first eigenvectors and multiple second eigenvectors, the first eigenvectors and the second eigenvectors may be normalized first, and the cosine value of the eigenvector may be calculated based on the normalized first eigenvectors and the second eigenvectors.
[0080] Furthermore, by obtaining a set cosine threshold and comparing the cosine threshold with the cosine value, the first identification state can be determined. That is, by obtaining the set cosine threshold, in response to the cosine value being greater than the cosine threshold, the first identification state of the wind shaft explosion-proof cover is determined to be a normal state; in response to the cosine value being less than the cosine threshold, the first identification state of the wind shaft explosion-proof cover is determined to be an abnormal state.
[0081] S404: Determine an intersecting image corresponding to the second region of interest and the foreground image.
[0082] In some implementations, the second region of interest may be used to crop the foreground image to obtain an intersection image, and the intersection ratio of the foreground image and the region of interest may be obtained by intersecting the foreground image with the second region of interest.
[0083] Furthermore, by determining a set intersection ratio threshold, comparing the intersection ratio with the set intersection threshold, and when the intersection ratio is less than the set intersection ratio threshold, the foreground image is cropped to obtain an intersection image. For example, any region whose intersection ratio with the second region of interest is less than 0.2 can be cropped in the foreground image as the intersection image.
[0084] S405: The intersection image, the first region of interest, and the second region of interest are input as input information into a plurality of different neural networks in the target state classification model, the neural network determines the confidence corresponding to the input information, and determines the second recognition state based on the confidence.
[0085] In some implementations, the target state classification model receives input information consisting of an intersection image, a first region of interest, and a second region of interest. A neural network can calculate a first target distance between the intersection image and the first region of interest, and a second target distance between the intersection image and the second region of interest, and determine the confidence level corresponding to the input information based on the first target distance and the second target distance.
[0086] Optionally, the first target distance and the second target distance may be calculated based on the above formula (1) and formula (2). Optionally, the confidence level corresponding to the input information may be calculated based on the above formula (3).
[0087] Further, if the confidence level is greater than 0, it can be determined that the second identification state is that the wind shaft explosion-proof cover is in an abnormal state; if the confidence level is 0, it can be determined that the second identification state is that the wind shaft explosion-proof cover is in a normal state.
[0088] S406: In response to the first identification state and the second identification state both indicating that the wind shaft explosion-proof cover is in an abnormal state, determining that the target state of the wind shaft explosion-proof cover is an abnormal state.
[0089] In the embodiment of the present application, the implementation method of step S406 can be implemented by any method in the embodiments of the present application, which is not limited here and will not be repeated.
[0090] In some implementations, when the first identification state and the second identification state indicate that the wind shaft explosion-proof cover is in a normal state, the target state of the wind shaft explosion-proof cover can be determined to be a normal state. That is, in response to the first identification state that the wind shaft explosion-proof cover is in a normal state and the second identification state that the wind shaft explosion-proof cover is in an abnormal state, the target state of the wind shaft explosion-proof cover is determined to be a normal state. Alternatively, in response to the first identification state that the wind shaft explosion-proof cover is in an abnormal state and the second identification state that the wind shaft explosion-proof cover is in a normal state, the target state of the wind shaft explosion-proof cover is determined to be a normal state.
[0091] In the abnormal state identification method of the wind shaft explosion-proof cover provided in the embodiment of the present application, by collecting an image of the wind shaft explosion-proof cover and determining the background image and the foreground image from the image, the feature vector of the wind shaft explosion-proof cover is determined according to the region of interest in the background image and the foreground image. The first identification state of the wind shaft explosion-proof cover is determined according to the cosine value between the feature vectors. By inputting the intersecting image, the first region of interest, and the second region of interest into the target state classification model, the second identification state of the wind shaft explosion-proof cover can be determined, so as to jointly determine the target state of the wind shaft explosion-proof cover according to the first identification state and the second identification state. Through the dual identification mechanism, it is possible to more accurately judge whether the wind shaft explosion-proof cover is in an abnormal state, so as to issue an early warning in time to avoid potential safety hazards. At the same time, an alarm is only issued when both the first identification state and the second identification state identify an abnormality, which effectively filters out a large amount of false alarm information, thereby improving the reliability of the alarm information.
[0092] In order to implement the above-mentioned embodiment, the present application also proposes an abnormal state identification device for an air shaft explosion-proof cover.
[0093] Figure 5 A schematic structural diagram of an abnormal state identification device for an air shaft explosion-proof cover provided in an embodiment of the present application.
[0094] like Figure 5 As shown, the abnormal state identification device 500 of the wind shaft explosion-proof cover includes:
[0095] The acquisition module 501 is used to acquire images of the wind shaft explosion-proof cover to obtain an image of the wind shaft explosion-proof cover, and perform Gaussian background modeling based on the region of interest in the image to determine the background image and foreground image of the wind shaft explosion-proof cover;
[0096] An extraction module 502 is used to extract a first feature vector of the air shaft explosion-proof cover from the background image and a second feature vector of the air shaft explosion-proof cover from the foreground image based on a first region of interest in the background image and a second region of interest in the foreground image;
[0097] A first determination module 503 is used to determine a cosine value of a feature vector based on the first feature vector and the second feature vector, and to determine a first recognition state of the air shaft explosion-proof cover based on the cosine value;
[0098] A second determination module 504 is used to determine a second identification state of the air shaft explosion-proof cover based on the trained target state classification model, the first region of interest, and the second region of interest, wherein each of the first identification state and the second identification state is that the air shaft explosion-proof cover is in a normal state, or the air shaft explosion-proof cover is in an abnormal state;
[0099] The third determination module 505 is configured to determine that the target state of the air shaft explosion-proof cover is an abnormal state in response to the first identification state and the second identification state both indicating that the air shaft explosion-proof cover is in an abnormal state.
[0100] In a possible implementation of an embodiment of the present application, the second determination module 504 is also used to: obtain a training data set, and determine an anchor sample, a first positive sample and a first negative sample from the training data set, wherein the training data set includes a normal state image and an abnormal state image of the wind shaft explosion-proof cover; based on the anchor sample, the first positive sample and the first negative sample, train the initial first state classification model until the sample data in the training data set is trained to obtain a second state classification model; obtain a second positive sample of the wind shaft explosion-proof cover, and determine a second negative sample based on the second positive sample and the second region of interest; based on the anchor sample, the second positive sample and the second negative sample, train the second state classification model to obtain the target state classification model; wherein each state classification model is a twin network structure.
[0101] In a possible implementation of an embodiment of the present application, the second determination module 504 is further used to: input the anchor point sample, the first positive sample and the first negative sample into the same neural network in the initial first state classification model to determine the first distance between the first positive sample and the anchor point sample, and the second distance between the first negative sample and the anchor point sample; determine the first loss function of the first state classification model based on the first distance and the second distance, and correct the model based on the first loss function and continue training until the sample data in the training data set is trained, and the second state classification model is obtained.
[0102] In a possible implementation manner of the embodiment of the present application, the second determination module 504 is further used to: collect images of the wind shaft explosion-proof cover at different times as sample images; and construct the second positive sample based on the sample images.
[0103] In a possible implementation of the embodiment of the present application, the second determination module 504 is further used to: for any sample image in the second positive sample, intersect the any sample image with the second region of interest to obtain an intersection ratio between the any sample image and the region of interest; in response to the intersection ratio being less than a set intersection ratio threshold, cropping the any sample image to obtain an intersection image corresponding to the any sample image, and generating the second negative sample based on the intersection image.
[0104] In a possible implementation manner of an embodiment of the present application, the second determination module 504 is further used to: input the anchor point sample, the second positive sample and the second negative sample into the same neural network in the second state classification model to determine a third distance between the second positive sample and the anchor point sample, and a fourth distance between the second negative sample and the anchor point sample; determine a second loss function of the second state classification model based on the third distance and the fourth distance, and adjust the second state classification model based on the second loss function to obtain the target state classification model.
[0105] In a possible implementation of an embodiment of the present application, the second determination module 504 is further used to: determine an intersection image corresponding to the second region of interest and the foreground image; input the intersection image, the first region of interest, and the second region of interest as input information into multiple different neural networks in the target state classification model, and the neural network determines the confidence corresponding to the input information, and determines the second recognition state based on the confidence.
[0106] In a possible implementation of an embodiment of the present application, the extraction module 502 is also used to: for any image between the background image and the foreground image, obtain pixel information of the any image, and based on the pixel information, perform region division on the region of interest corresponding to the any image to obtain multiple neighborhoods; based on the pixel information of each neighborhood, calculate the pixel mean of each neighborhood, and based on the pixel mean, determine the pixel variance of each neighborhood as a feature vector of the any image; wherein, if the any image is the background image, the feature vector of the any image is a first feature vector; if the any image is the foreground image, the feature vector of the any image is a second feature vector.
[0107] In a possible implementation of an embodiment of the present application, the first determination module 503 is also used to: obtain a set cosine threshold, and in response to the cosine value being greater than the cosine threshold, determine that the first identification state of the air shaft explosion-proof cover is a normal state; in response to the cosine value being less than the cosine threshold, determine that the first identification state of the air shaft explosion-proof cover is an abnormal state.
[0108] In a possible implementation of the embodiment of the present application, the device also includes: in response to the first identification state that the wind shaft explosion-proof cover is in a normal state and the second identification state that the wind shaft explosion-proof cover is in an abnormal state, determining that the target state of the wind shaft explosion-proof cover is a normal state; or, in response to the first identification state that the wind shaft explosion-proof cover is in an abnormal state and the second identification state that the wind shaft explosion-proof cover is in a normal state, determining that the target state of the wind shaft explosion-proof cover is a normal state.
[0109] In the abnormal state recognition device for the wind shaft explosion-proof cover provided in the embodiment of the present application, by collecting an image of the wind shaft explosion-proof cover and determining the background image and the foreground image from the image, the feature vector of the wind shaft explosion-proof cover is determined according to the region of interest in the background image and the foreground image. The first recognition state of the wind shaft explosion-proof cover is determined according to the cosine value between the feature vectors, and the second recognition state of the wind shaft explosion-proof cover is determined according to the trained target state classification model, so as to jointly determine the target state of the wind shaft explosion-proof cover according to the first recognition state and the second recognition state. Through the dual recognition mechanism, it is possible to more accurately judge whether the wind shaft explosion-proof cover is in an abnormal state, so as to issue an early warning in time to avoid potential safety hazards. At the same time, an alarm is only issued when both the first recognition state and the second recognition state identify an abnormality, which effectively filters out a large amount of false alarm information, thereby improving the reliability of the alarm information.
[0110] It should be noted that the above explanation of the embodiment of the abnormal state identification method of the wind shaft explosion-proof cover is also applicable to the abnormal state identification device of the wind shaft explosion-proof cover of this embodiment, and will not be repeated here.
[0111] The collection, storage, use, processing, transmission, provision and application of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.
[0112] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.
[0113] This application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, this application is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, risks can be minimized by limiting data collection and deleting data. In addition, when applicable, personal identifiers are removed from such personal information to protect the privacy of users.
[0114] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0115] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0116] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0117] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.
[0118] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0119] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0120] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0121] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A method for identifying an abnormal state of an explosion-proof cover of a wind shaft, characterized in that: The method comprises: Performing image acquisition on the wind shaft explosion-proof cover to obtain an image of the wind shaft explosion-proof cover, and performing Gaussian background modeling based on an area of interest in the image to determine a background image and a foreground image of the wind shaft explosion-proof cover; Based on the first region of interest in the background image and the second region of interest in the foreground image, extracting a first feature vector of the air shaft explosion-proof cover from the background image, and extracting a second feature vector of the air shaft explosion-proof cover from the foreground image; Determine a cosine value of a characteristic vector based on the first characteristic vector and the second characteristic vector, and determine a first identification state of the air shaft explosion-proof cover based on the cosine value; Based on the trained target state classification model, the first region of interest and the second region of interest, determining a second recognition state of the wind shaft explosion-proof cover, wherein each of the first recognition state and the second recognition state is that the wind shaft explosion-proof cover is in a normal state, or the wind shaft explosion-proof cover is in an abnormal state; In response to the first identification state and the second identification state both being that the wind shaft explosion-proof cover is in an abnormal state, it is determined that the target state of the wind shaft explosion-proof cover is an abnormal state.
2. The method according to claim 1, characterized in that The training process of the target state classification model includes: Acquire a training data set, and determine an anchor sample, a first positive sample, and a first negative sample from the training data set, wherein the training data set includes a normal state image and an abnormal state image of the wind shaft explosion-proof cover; Based on the anchor point sample, the first positive sample and the first negative sample, the initial first state classification model is trained until the sample data in the training data set is trained, thereby obtaining a second state classification model; Acquire a second positive sample of the air shaft explosion-proof cover, and determine a second negative sample based on the second positive sample and the second region of interest; Based on the anchor point sample, the second positive sample and the second negative sample, training the second state classification model to obtain the target state classification model; Among them, each state classification model is a twin network structure.
3. The method according to claim 2, characterized in that The initial first state classification model is trained based on the anchor point sample, the first positive sample and the first negative sample until the sample data in the training data set is trained to obtain a second state classification model, including: Inputting the anchor sample, the first positive sample and the first negative sample into the same neural network in the initial first state classification model to determine a first distance between the first positive sample and the anchor sample, and a second distance between the first negative sample and the anchor sample; Based on the first distance and the second distance, a first loss function of the first state classification model is determined, and the model is corrected based on the first loss function and continues to be trained until the sample data in the training data set is trained to obtain the second state classification model.
4. The method according to claim 2, characterized in that: The obtaining of a second positive sample of the air shaft explosion-proof cover comprises: Collecting images of the air shaft explosion-proof cover at different times as sample images; Based on the sample image, the second positive sample is constructed.
5. The method according to claim 4, characterized in that The determining a second negative sample based on the second positive sample and the second region of interest includes: For any sample image in the second positive sample, intersect the any sample image with the second region of interest to obtain an intersection ratio between the any sample image and the region of interest; In response to the intersection ratio being less than a set intersection ratio threshold, any one of the sample images is cropped to obtain an intersection image corresponding to the any one of the sample images, and the second negative sample is generated based on the intersection image.
6. The method according to claim 3, characterized in that The training of the second state classification model based on the anchor point sample, the second positive sample and the second negative sample to obtain the target state classification model includes: Inputting the anchor sample, the second positive sample, and the second negative sample into the same neural network in the second state classification model to determine a third distance between the second positive sample and the anchor sample, and a fourth distance between the second negative sample and the anchor sample; Based on the third distance and the fourth distance, a second loss function of the second state classification model is determined, and the second state classification model is adjusted based on the second loss function to obtain the target state classification model.
7. The method according to claim 5, characterized in that The determining the second recognition state of the air shaft explosion-proof cover based on the trained target state classification model, the first region of interest, and the second region of interest includes: Determine an intersecting image corresponding to the second region of interest and the foreground image; The intersection image, the first region of interest, and the second region of interest are input as input information into a plurality of different neural networks in the target state classification model, the neural network determines the confidence corresponding to the input information, and determines the second recognition state based on the confidence.
8. The method according to claim 1, characterized in that The extracting a first feature vector of the air shaft explosion-proof cover from the background image and a second feature vector of the air shaft explosion-proof cover from the foreground image based on the first region of interest in the background image and the second region of interest in the foreground image comprises: For any image among the background image and the foreground image, pixel information of the image is obtained, and based on the pixel information, a region of interest corresponding to the image is divided into regions to obtain a plurality of neighborhoods; Based on the pixel information of each neighborhood, a pixel mean of each neighborhood is calculated, and based on the pixel mean, a pixel variance of each neighborhood is determined as a feature vector of any one of the images; Wherein, if any one of the images is the background image, the feature vector of any one of the images is the first feature vector; if any one of the images is the foreground image, the feature vector of any one of the images is the second feature vector.
9. The method according to claim 1, characterized in that: The determining, based on the cosine value, a first identification state of the air shaft explosion-proof cover comprises: Obtaining a set cosine threshold, and in response to the cosine value being greater than the cosine threshold, determining that the first identification state of the air shaft explosion-proof cover is a normal state; In response to the cosine value being less than the cosine threshold, it is determined that the first identification state of the air shaft explosion-proof cover is an abnormal state.
10. The method according to claim 1, characterized in that The method further comprises: In response to the first identification state being that the wind shaft explosion-proof cover is in a normal state and the second identification state being that the wind shaft explosion-proof cover is in an abnormal state, determining that the target state of the wind shaft explosion-proof cover is a normal state; or, In response to the first identification state being that the wind shaft explosion-proof cover is in an abnormal state and the second identification state being that the wind shaft explosion-proof cover is in a normal state, it is determined that the target state of the wind shaft explosion-proof cover is a normal state.