CN116310922A - Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium - Google Patents
Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN116310922A CN116310922A CN202111555544.0A CN202111555544A CN116310922A CN 116310922 A CN116310922 A CN 116310922A CN 202111555544 A CN202111555544 A CN 202111555544A CN 116310922 A CN116310922 A CN 116310922A
- Authority
- CN
- China
- Prior art keywords
- risk identification
- monitoring video
- petrochemical plant
- identification method
- plant area
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 8
- 238000002372 labelling Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 16
- 239000013598 vector Substances 0.000 claims description 11
- 238000009826 distribution Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 5
- 239000000779 smoke Substances 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 3
- 239000000463 material Substances 0.000 claims description 3
- 231100001261 hazardous Toxicity 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 abstract description 4
- 238000013461 design Methods 0.000 abstract 1
- 238000005259 measurement Methods 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 8
- 230000002776 aggregation Effects 0.000 description 4
- 238000004220 aggregation Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000008447 perception Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000008521 reorganization Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a petrochemical plant area monitoring video risk identification method, which comprises the following steps: collecting public accident video data; labeling frame images with set intervals in the accident video data, wherein the labeled frame images form a sample set; randomly dividing the sample set into a training data set and a test data set; training a feature extractor taking the VGG model as a backbone network by adopting a training data set to obtain a single-classification network model; testing the recognition accuracy of the trained single-classification network model to the abnormal scene by adopting a test data set, and establishing a risk recognition model; and inputting the petrochemical plant monitoring video data into a risk identification model to obtain a risk identification result. The invention also discloses a petrochemical plant area monitoring video risk identification system, an electronic device and a storage medium. The invention is based on a single-classification image recognition algorithm based on a convolutional neural network, and takes the design of a loss measurement function as an improvement direction, thereby improving the real-time performance and the accuracy of risk scene recognition.
Description
Technical Field
The invention relates to the technical field of petrochemical plant area risk monitoring, in particular to a petrochemical plant area monitoring video risk identification method, a petrochemical plant area monitoring video risk identification system, electronic equipment and a storage medium.
Background
At present, the petrochemical enterprise safety production situation is stable, the safety risk management and control technical equipment is increasingly perfect, but the safety risk of the critical area of the factory is monitored and perceived only by the on-duty personnel inspection and the intelligent sensor, the risk perception means is single, the coverage surface is deficient, the response time is different, and the specific risk perception capability of the critical area of the factory is still to be further enhanced. With the rapid development of video acquisition and transmission technology, video monitoring is increasingly popular in the aspects of security protection, equipment monitoring, duty on duty and the like of petrochemical enterprises, but most monitoring videos are limited to on-site visual management, backtracking and evidence collection and other services, and lack of risk identification capability.
The computer vision technology is utilized to expand the enterprise video monitoring as a factory key area risk sensing means, and the real-time dynamic sensing and early warning capability of specific risks in the key area can be effectively improved. The fire disaster in the monitoring video is identified by a method based on deep learning, so that manual inspection can be replaced, the manpower consumption is reduced, and the risk perception instantaneity is improved.
In the prior art, patent document CN112907886A discloses a fire disaster identification method of a refinery based on a convolutional neural network, wherein a flame detection model of the method depends on detection of a moving target by a background differential algorithm and is limited by the change of the background such as illumination, brightness and the like; patent document CN110032977a discloses a safety early warning management system based on deep learning image fire disaster recognition, which needs to preprocess images through a foreground detection technology; patent document CN109886227a discloses an indoor fire video recognition method based on a multichannel convolutional neural network, which divides a flame region in an image, extracts the flame region and uses the flame region as a detection channel according to color characteristics, circularity characteristics, and the like.
The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person of ordinary skill in the art.
Disclosure of Invention
The invention aims to provide a petrochemical plant monitoring video risk identification method, a petrochemical plant monitoring video risk identification system, electronic equipment and a storage medium, so that the problems of structural enlargement, complex calculation and the like of a risk scene identification scheme in the prior art are solved.
Another object of the present invention is to provide a petrochemical plant area monitoring video risk identification method, system, electronic device and storage medium, so as to improve the dependency of the prior art risk scene identification on image processing.
The invention further aims to provide a petrochemical plant monitoring video risk identification method, a petrochemical plant monitoring video risk identification system, electronic equipment and a storage medium, so that the real-time performance and the accuracy of risk scene identification are improved.
To achieve the above object, according to a first aspect of the present invention, there is provided a petrochemical plant area monitoring video risk identification method, comprising the steps of: collecting public accident video data; labeling frame images with set intervals in accident video data, wherein the frame image label in a normal scene is 1, the frame image label in an abnormal scene is 0, and the labeled frame images form a sample set; randomly dividing the sample set into a training data set and a test data set; training a feature extractor taking the VGG model as a backbone network by adopting a training data set to obtain a single-classification network model; testing the recognition accuracy of the trained single-classification network model to the abnormal scene by adopting a test data set, and establishing a risk recognition model; and inputting the petrochemical plant monitoring video data into a risk identification model to obtain a risk identification result.
Further, in the above technical solution, the vector distribution of different types of frame images in the feature space is constrained by combining a self-defined loss function when training the feature extractor using the VGG model as the backbone network.
Further, in the above technical solution, the custom loss function is:
wherein L is c Is a binary cross entropy loss function:
wherein y epsilon {0,1} represents the class corresponding to the classifier, O represents the abnormal scene class, and 1 represents the normal scene class; n represents the batch size of the current training batch; p represents a softmax probability score of y=0, 1-p represents a softmax probability score of y=1;
L n aggregate loss functions for class centers:
wherein z is i Representing the ith sample in the current batch,
represents the y i Class center point of class, i.e. [0,1 ]]Alpha is the equilibrium super parameter, punishment term->
Representing z i The feature vectors of the samples are far from the center points of the other classes, i.e. the abnormal scene class (y i =0) features of the image are far from the normal scene class (y) in euclidean spatial distance i =1) features of the image to learn an embedding space with discriminant;
beta=0.5 for controlling L n Is a loss of (2).
Further, in the above technical solution, when the feature extractor using the VGG model as the backbone network is trained by using the training data set, the Adam gradient descent algorithm is used to update the network structure parameters.
Further, in the above technical solution, the accident video data disclosed includes history monitoring video data and public network video data.
Further, in the above technical solution, the public network video data is obtained by searching a specific keyword on the open source video website.
Further, in the above-described embodiments, the interval is set to 10 to 15 frames.
Further, in the above technical solution, the petrochemical factory monitoring video risk identification method further includes: scaling the marked frame images in the sample set to the same size, and carrying out normalization processing; and expanding the sample set by adopting a data enhancement mode of random rotation, flipping or Gaussian noise addition.
Further, in the above technical scheme, 70% -75% of the sample sets form training data sets, and the rest form test data sets.
Further, in the above technical solution, the abnormal scene includes smoke, fire, intrusion in a dangerous area, or accumulation of materials.
According to a second aspect of the present invention, there is provided a petrochemical plant area monitoring video risk identification system, comprising: a video image acquisition unit for extracting frame images of the monitoring video in the target area at set intervals; a risk recognition unit that performs risk recognition on the frame image of the surveillance video in the extracted target area based on the risk recognition model; and an output unit for outputting a result of the risk identification, wherein the risk identification model is established based on the disclosed incident video data and the single classification neural network.
According to a third aspect of the present invention, there is provided a petrochemical plant monitoring video risk identification system, which adopts the petrochemical plant monitoring video risk identification method according to any one of the above technical solutions.
According to a fourth aspect of the present invention, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor performs the petrochemical plant area monitoring video risk identification method according to any one of the technical schemes.
According to a fifth aspect of the present invention, there is provided a non-transitory computer readable storage medium storing computer executable instructions for causing a computer to perform the petrochemical plant area monitoring video risk identification method according to any one of the above technical solutions.
Compared with the prior art, the invention has one or more of the following beneficial effects:
1. according to the invention, the video recognition of the factory abnormal scene is regarded as a single-classification abnormal detection problem with a brand-new view angle, and aiming at the monitoring video of the petrochemical factory, the problems of structural enlargement, complex calculation and the like based on the traditional target detection scheme are greatly reduced based on a single-classification image recognition algorithm based on the convolutional neural network.
2. According to the invention, the self-defined loss function is combined during model training, the vector distribution of different types of frame images in the feature space is restrained, the learning of abnormal image features such as smoke, fire and the like is enhanced, the neural network can learn the features with the most discrimination, and the recognition precision is improved.
3. Aiming at the real-time dynamic sensing and early warning requirements of specific risks in key areas of factories, the application of the deep learning model in video image recognition is utilized, the monitoring video of enterprises is expanded to serve as a risk sensing means, and the specific risk hidden danger recognition capability and the dynamic monitoring and early warning capability of petrochemical enterprises are improved.
4. The invention can effectively utilize the existing monitoring video resources, public network data and the like, and avoid model distortion caused by insufficient data quantity.
5. The whole network uses an end-to-end model, does not need to intervene in the network by the traditional image recognition technology, reduces the complexity of the model, and realizes the rapid recognition of abnormal scenes.
The foregoing description is only an overview of the present invention, and it is to be understood that it is intended to provide a more clear understanding of the technical means of the present invention and to enable the technical means to be carried out in accordance with the contents of the specification, while at the same time providing a more complete understanding of the above and other objects, features and advantages of the present invention, and one or more preferred embodiments thereof are set forth below, together with the detailed description given below, along with the accompanying drawings.
Drawings
Fig. 1 is a flowchart of a petrochemical plant area monitoring video risk identification method according to an embodiment of the present invention.
Fig. 2 is a block diagram of a petrochemical plant area monitoring video risk identification system according to an embodiment of the present invention.
Fig. 3 is a schematic hardware structure of an electronic device for performing the petrochemical plant area monitoring video risk identification method according to an embodiment of the present invention.
Detailed Description
The following detailed description of embodiments of the invention is, therefore, to be taken in conjunction with the accompanying drawings, and it is to be understood that the scope of the invention is not limited to the specific embodiments.
Throughout the specification and claims, unless explicitly stated otherwise, the term "comprise" or variations thereof such as "comprises" or "comprising", etc. will be understood to include the stated element or component without excluding other elements or other components.
Spatially relative terms, such as "below," "beneath," "lower," "above," "upper," and the like, may be used herein for ease of description to describe one element's or feature's relationship to another element's or feature's in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the article in use or operation in addition to the orientation depicted in the figures. For example, if the article in the figures is turned over, elements described as "below" or "beneath" other elements or features would then be oriented "above" the elements or features. Thus, the exemplary term "below" may encompass both a direction of below and a direction of above. The article may have other orientations (rotated 90 degrees or other orientations) and the spatially relative descriptors used herein interpreted accordingly.
The terms "first," "second," and the like herein are used for distinguishing between two different elements or regions and are not intended to limit a particular position or relative relationship. In other words, in some embodiments, the terms "first," "second," etc. may also be interchanged with one another.
Example 1
As shown in fig. 1, the flow of the petrochemical plant area monitoring video risk identification method according to the embodiment of the invention is as follows:
s110 collects the disclosed accident video data.
The disclosed incident video data may include historical surveillance video data and public network video data. Public network video data can be obtained by searching specific keywords on an open source video website. For example, related public accident video data can be crawled and downloaded from an open source video website such as Youtube by using "factory explosion", "factory fire" and the like as keywords by using a Python crawler library beaufulSoup.
S120, marking frame images with set intervals in the accident video data, wherein the marked frame images form a sample set.
The accident video data obtained in the step S110 are read in by adopting a machine vision software library OpenCV and a frame image is reserved every ten frames according to a time sequence; and then, carrying out data annotation on the stored frame image by using a CoLabler image annotation tool, wherein the annotation format is as follows: the frame image label under normal scenes (including but not limited to factory daily, sky) is 1, and the video frame image label under abnormal scenes (including but not limited to smoke, fire) is 0.
Exemplary, the sample set is preprocessed, the size of each frame image is scaled to 480 x 480 pixels, the original information of the input image is fully reserved, the average value and standard deviation of the whole sample set are calculated through sampling, and then the average value and standard deviation parameters are used for normalizing each frame image, so that disturbance to the model caused by the oversized pixel value is reduced. In addition, the sample set is further expanded in a data enhancement mode of random rotation, flipping, gaussian noise addition and the like.
By way of example, abnormal scenarios may include smoke, fire, hazardous area intrusion, or material accumulation.
S130 randomly divides the sample set into a training data set and a test data set.
The sample set obtained in step S120 is randomly extracted 70% as a training data set, and the remaining 30% is a test data set. It should be understood that the present invention is not limited thereto, and those skilled in the art can select the proportion of the training data set according to the actual situation, for example, randomly extracting 70% -75% as the training data set and the rest as the test data set.
S140, training a feature extractor taking the VGG model as a backbone network by adopting a training data set to obtain a single-classification network model; and testing the recognition accuracy of the trained single-classification network model to the abnormal scene by adopting a test data set, and establishing a risk recognition model.
Further, in one or more exemplary embodiments of the present invention, vector distributions of different class frame images in a feature space are constrained in combination with a custom loss function when training a feature extractor using a VGG model as a backbone network, and model parameters are gradually optimized through a back propagation algorithm.
Further, in one or more exemplary embodiments of the invention, the binary cross entropy loss function of the model is set as:
wherein y epsilon {0,1} represents the class corresponding to the classifier, 0 represents the abnormal scene class, and 1 represents the normal scene class; n represents the batch size of the current training batch; p represents the softmax probability score of y=0, and 1-p represents the softmax probability score of y=1.
Further, setting a class center aggregation loss function places further constraints on the model. And (3) taking the feature vector of the global averaging layer, and sending the feature vector to a full-connection layer with 512-dimensional input and 2-dimensional output, wherein the feature vector with 2-dimensional represents the cluster center point of the normal scene class and the abnormal scene class in the feature space. The class center aggregation loss is used as a regular term in the total loss, the feature space distribution corresponding to the normal scene class and the abnormal scene class is constrained, the feature space distribution clusters of the normal scene class and the abnormal scene class are pulled apart as far as possible, and the distribution of the normal scene class and the abnormal scene class is monitored to be aggregated as far as possible. The image features thus learned are more discriminative, i.e. the spatial distribution of the feature vectors of the normal scene class image is distinguishable from the spatial distribution of the features of the abnormal scene image. The center aggregation loss function is as follows:
wherein z is i Representing the ith sample in the current batch,
represents the y i Class center point of class, i.e. [0,1 ]]Alpha is the equilibrium super parameter, punishment term->
Representing z i The feature vectors of the samples are far from the center points of the other classes, i.e. the abnormal scene class (y i =0) features of the image are far from the normal scene class (y) in euclidean spatial distance i =1) features of the image to learn an embedding space with discriminant.
Further, the customization of the modelThe loss function is:
where β=0.5 is used to control L n The loss of (2) is not excessive.
Further, in one or more exemplary embodiments of the invention, the network structure parameters are updated using Adam gradient descent algorithm when training the feature extractor using VGG model as the backbone network using the training data set.
S150, inputting the petrochemical plant monitoring video data into a risk identification model to obtain a risk identification result.
Example 2
Referring to fig. 2, the petrochemical plant area monitoring video risk identification system of the present embodiment includes: a video image acquisition unit 10 for extracting frame images of the monitoring video in the target area at set intervals; a risk recognition unit 20 that performs risk recognition on the frame image of the surveillance video in the extracted target area based on a risk recognition model established based on the disclosed accident video data and the single classification neural network; and an output unit 30 for outputting a result of the risk identification.
The risk identification unit 20 includes a feature extraction module 21, a feature reorganization module 22, and an anomaly detection classifier 23.
The feature extraction module 21 takes the VGG16 feature extractor as a backbone, and consists of 13 convolution blocks, wherein each convolution block consists of a convolution layer, a BN layer and a ReLU activation layer. And adding a maximum pooling layer after the 2 nd, 4 th, 7 th, 10 th and 13 th convolution blocks, and performing dimension reduction and feature aggregation on the feature map after convolution. In order to train on limited labeling data to obtain a robust feature extractor, the module adopts the idea of transfer learning, and the feature extraction module loads model weights pre-trained on image data sets of tens of millions of image data sets of ImageNet.
The feature reorganization module 22 is composed of a layer of convolution blocks (Conv 1) with a convolution kernel size of 3*3, a step size of 2, and a padding of 0, and a maximum average pooling layer (GAP). The input feature map size of the feature re-assembly module 22 is 512×12×12, where 512 represents the number of feature channels, and the feature map size is reduced to 512×5×5 after Conv1, so as to further realize feature dimension reduction. The GAP layer greatly reduces parameters of the network by spatial information and average value of the feature map, so that the feature map of 512 x 5 can be reduced to 512 x 1 without any parameter fitting, and the compression of the information is completed. The definition is as follows:
where input represents the input of the GAP layer, i.e., a feature map with a size of 512×5×5, h×w represents the feature map size, and x, y represents the channel value behind each pixel point on the feature map.
Abnormality detection classifier 23: the original VGG classifier is required to be mapped and classified through three full-connection layers, and because the number of the finally recognized classes is only abnormal scene classes and other classes, complex mapping for many times is not required, the VGG original classifier is changed into a full-connection layer which is 512-dimensional in one layer and 2-dimensional in the output and provided with a dropout function. After the full connection layer, using a softmax classification function, the abnormal scene class is classified into 0, the normal scene class is classified into 1, and the model parameters are iteratively optimized with the real labels through a back propagation algorithm.
Example 3
The present embodiment provides a non-transitory (non-volatile) computer storage medium storing computer-executable instructions that can perform the method of any of the above-described method embodiments and achieve the same technical effects.
Example 4
The present embodiments provide a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, cause the computer to perform the methods of the above aspects and achieve the same technical effects.
Example 5
Fig. 3 is a schematic hardware structure of an electronic device for executing the petrochemical plant area monitoring video risk identification method according to the present embodiment. The device includes one or more processors 610 and memory 620. Take one processor 610 as an example. The apparatus may further include: an input device 630 and an output device 640.
The processor 610, memory 620, input devices 630, and output devices 640 may be connected by a bus or other means, for example in fig. 3.
The input device 630 may receive input numeric or character information and generate signal inputs. The output device 640 may include a display device such as a display screen.
One or more modules are stored in the memory 620 that, when executed by the one or more processors 610, perform:
collecting public accident video data;
labeling frame images with set intervals in accident video data, wherein the frame image label in a normal scene is 1, the frame image label in an abnormal scene is 0, and the labeled frame images form a sample set;
randomly dividing the sample set into a training data set and a test data set;
training a feature extractor taking the VGG model as a backbone network by adopting a training data set to obtain a single-classification network model;
testing the recognition accuracy of the trained single-classification network model to the abnormal scene by adopting a test data set, and establishing a risk recognition model; and
and inputting the petrochemical plant monitoring video data into a risk identification model to obtain a risk identification result.
The product can execute the method provided by the embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment may be found in the methods provided in other embodiments of the present invention.
The apparatus embodiments described above are merely illustrative, wherein elements illustrated as separate elements may or may not be physically separate, and elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
From the above description of embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus a general purpose hardware platform, or may be implemented by hardware. Based on such understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the related art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method of the respective embodiments or some parts of the embodiments.
The foregoing descriptions of specific exemplary embodiments of the present invention are presented for purposes of illustration and description. It is not intended to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application to thereby enable one skilled in the art to make and utilize the invention in various exemplary embodiments and with various modifications as are suited to the particular use contemplated. Any simple modifications, equivalent variations and modifications of the above-described exemplary embodiments should fall within the scope of the present invention.
Claims (14)
1. The petrochemical plant area monitoring video risk identification method is characterized by comprising the following steps of:
collecting public accident video data;
labeling frame images with set intervals in the accident video data, wherein the frame image label in a normal scene is 1, the frame image label in an abnormal scene is 0, and the labeled frame images form a sample set;
randomly dividing the sample set into a training data set and a test data set;
training a feature extractor taking the VGG model as a backbone network by adopting a training data set to obtain a single-classification network model;
testing the recognition accuracy of the trained single-classification network model to the abnormal scene by adopting a test data set, and establishing a risk recognition model; and
and inputting the petrochemical plant monitoring video data into the risk identification model to obtain a risk identification result.
2. The petrochemical plant area monitoring video risk identification method according to claim 1, wherein vector distribution of different types of frame images in a feature space is constrained by combining a self-defined loss function when training a feature extractor taking a VGG model as a main network.
3. The petrochemical plant area monitoring video risk identification method according to claim 2, wherein the custom loss function is:
wherein L is c Is a binary cross entropy loss function:
wherein y epsilon {0,1} represents the class corresponding to the classifier, 0 represents the abnormal scene class, and 1 represents the normal scene class; n represents the batch size of the current training batch; p represents a softmax probability score of y=0, 1-p represents a softmax probability score of y=1;
L n aggregate loss functions for class centers:
wherein z is i Representing the ith sample in the current batch,
represents the y i Class center point of class, i.e. [0,1 ]]Alpha is the equilibrium super parameter, punishment term->
Representing z i The feature vectors of the samples are far from the center points of the other classes, i.e. the abnormal scene class (y i =0) features of the image are far from the normal scene class (y) in euclidean spatial distance i =1) features of the image to learn an embedding space with discriminant; beta=0.5 for controlling L n Is a loss of (2).
4. The petrochemical plant area monitoring video risk identification method according to claim 1, wherein when a feature extractor taking a VGG model as a main network is trained by adopting a training data set, network structure parameters are updated by adopting an Adam gradient descent algorithm.
5. The petrochemical plant area monitoring video risk identification method of claim 1, wherein the disclosed incident video data comprises historical monitoring video data and public network video data.
6. The petrochemical plant area monitoring video risk identification method according to claim 5, wherein the public network video data is obtained through specific keyword search on an on-source video website.
7. The petrochemical plant area monitoring video risk identification method according to claim 1, wherein the set interval is 10-15 frames.
8. The petrochemical plant area monitoring video risk identification method according to claim 1, further comprising: scaling the marked frame images in the sample set to the same size, and carrying out normalization processing; and expanding the sample set by adopting a data enhancement mode of random rotation, flipping or Gaussian noise addition.
9. The petrochemical plant area monitoring video risk identification method according to claim 1, wherein 70% -75% of the sample sets form training data sets, and the rest form test data sets.
10. The petrochemical plant area monitoring video risk identification method of claim 1, wherein the abnormal scene comprises smoke, fire, hazardous area intrusion, or material accumulation.
11. A petrochemical plant area monitoring video risk identification system, comprising:
a video image acquisition unit for extracting frame images of the monitoring video in the target area at set intervals;
a risk recognition unit that performs risk recognition on the frame image of the surveillance video in the extracted target area based on the risk recognition model; and
an output unit for outputting a result of the risk identification,
wherein the risk identification model is established based on the disclosed incident video data and a single-classification neural network.
12. A petrochemical plant monitoring video risk identification system, characterized in that the system adopts the petrochemical plant monitoring video risk identification method according to any one of claims 1-10.
13. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,,
the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the petrochemical plant area monitoring video risk identification method of any one of claims 1-10.
14. A non-transitory computer-readable storage medium storing computer-executable instructions for causing a computer to perform the petrochemical plant area monitoring video risk identification method according to any one of claims 1 to 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111555544.0A CN116310922A (en) | 2021-12-17 | 2021-12-17 | Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111555544.0A CN116310922A (en) | 2021-12-17 | 2021-12-17 | Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116310922A true CN116310922A (en) | 2023-06-23 |
Family
ID=86787407
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111555544.0A Pending CN116310922A (en) | 2021-12-17 | 2021-12-17 | Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116310922A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117253196A (en) * | 2023-11-17 | 2023-12-19 | 本溪钢铁（集团）信息自动化有限责任公司 | Video-based security risk monitoring method and device in steel industry |
| CN117579625A (en) * | 2024-01-17 | 2024-02-20 | 中国矿业大学 | Inspection task pre-distribution method for double prevention mechanism |
-
2021
- 2021-12-17 CN CN202111555544.0A patent/CN116310922A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117253196A (en) * | 2023-11-17 | 2023-12-19 | 本溪钢铁（集团）信息自动化有限责任公司 | Video-based security risk monitoring method and device in steel industry |
| CN117253196B (en) * | 2023-11-17 | 2024-02-02 | 本溪钢铁（集团）信息自动化有限责任公司 | Video-based security risk monitoring method and device in steel industry |
| CN117579625A (en) * | 2024-01-17 | 2024-02-20 | 中国矿业大学 | Inspection task pre-distribution method for double prevention mechanism |
| CN117579625B (en) * | 2024-01-17 | 2024-04-09 | 中国矿业大学 | Inspection task pre-distribution method for double prevention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111800507A (en) | Traffic monitoring method and traffic monitoring system | |
| Piciarelli et al. | Surveillance-oriented event detection in video streams | |
| CN111738054B (en) | Behavior anomaly detection method based on space-time self-encoder network and space-time CNN | |
| CN110827505A (en) | Smoke segmentation method based on deep learning | |
| Kumar et al. | Automatic vehicle number plate recognition system using machine learning | |
| CN116310922A (en) | Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium | |
| Ghosh et al. | A hybrid deep learning model by combining convolutional neural network and recurrent neural network to detect forest fire | |
| CN110852222A (en) | Campus corridor scene intelligent monitoring method based on target detection | |
| Li et al. | Improved YOLOv4 network using infrared images for personnel detection in coal mines | |
| CN112163572A (en) | Method and device for identifying object | |
| CN115294528A (en) | Pedestrian safety monitoring method and device | |
| US10860865B2 (en) | Predictive security camera system | |
| CN116846059A (en) | Edge detection system for power grid inspection and monitoring | |
| CN111860457A (en) | Fighting behavior recognition early warning method and recognition early warning system thereof | |
| Yandouzi et al. | Investigation of combining deep learning object recognition with drones for forest fire detection and monitoring | |
| CN113191274A (en) | Oil field video intelligent safety event detection method and system based on neural network | |
| CN113095160A (en) | Power system personnel safety behavior identification method and system based on artificial intelligence and 5G | |
| CN117423157A (en) | Mine abnormal video action understanding method combining migration learning and regional invasion | |
| Ali et al. | Abnormal activity detection in surveillance video scenes | |
| CN116682162A (en) | Robot detection algorithm based on real-time video stream | |
| CN115083229B (en) | Intelligent recognition and warning system of flight training equipment based on AI visual recognition | |
| CN116543333A (en) | Target recognition method, training method, device, equipment and medium of power system | |
| CN116419059A (en) | Automatic monitoring method, device, equipment and medium based on behavior label | |
| Cai et al. | Multistage analysis of abnormal human behavior in complex scenes | |
| US10650275B2 (en) | Method for detection of temporal pattern anomalies in video streams |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |