CN113361637A - Potential safety hazard identification method and device for base station room - Google Patents
Potential safety hazard identification method and device for base station room Download PDFInfo
- Publication number
- CN113361637A CN113361637A CN202110743899.6A CN202110743899A CN113361637A CN 113361637 A CN113361637 A CN 113361637A CN 202110743899 A CN202110743899 A CN 202110743899A CN 113361637 A CN113361637 A CN 113361637A
- Authority
- CN
- China
- Prior art keywords
- potential safety
- safety hazard
- image
- preset
- hidden danger
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 231100001267 hazard identification Toxicity 0.000 title claims description 22
- 238000013145 classification model Methods 0.000 claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 26
- 238000012549 training Methods 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 8
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 5
- 230000006872 improvement Effects 0.000 claims description 5
- 238000000844 transformation Methods 0.000 claims description 4
- 239000002352 surface water Substances 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000015654 memory Effects 0.000 description 16
- 238000004364 calculation method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 8
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000007689 inspection Methods 0.000 description 3
- 238000010295 mobile communication Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- -1 branches Substances 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The application relates to the technical field of network operation and maintenance, and provides a method and a device for identifying potential safety hazards of a base station room, a computer readable storage medium and electronic equipment, wherein in the embodiment, the method comprises the following steps: acquiring an image to be identified, which is obtained by shooting the interior of a base station room by a shooting device, wherein the image to be identified comprises a potential safety hazard target; classifying images to be recognized through a preset hidden danger image classification model so as to determine the potential safety hazard class of a potential safety hazard target contained in the images to be recognized; and carrying out potential safety hazard judgment on potential safety hazard targets contained in the image to be recognized through a preset potential safety hazard recognition model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard targets contained in the image to be recognized have potential safety hazards. According to the technical scheme of the embodiment of the application, hidden danger judgment is carried out by adopting the unified hidden danger judgment rule learned by the model, the influence of subjective factors of people is avoided, and management tracking of the hidden danger can be carried out more conveniently.
Description
Technical Field
The application relates to the technical field of network operation and maintenance, in particular to a potential safety hazard identification method and device for a base station machine room.
Background
A base station, i.e., a common mobile communication base station, is a form of a radio station, which refers to a radio transceiver station for information transfer with a mobile phone terminal through a mobile communication switching center in a certain radio coverage area. The base station is generally divided into the following systems: transmission systems including SDH equipment, optical cables, and the like; power system, accumulator, commercial power, etc.; a moving loop monitoring system; an antenna feed system; a BTS master device; and other auxiliary equipment such as air conditioners, security doors, and the like. The main function of the base station is to provide wireless coverage, i.e. to enable wireless signal transmission between a wired communication network and a wireless terminal. As the number of base station rooms continues to increase, the maintenance problems that arise become more and more pronounced.
For the equipment and facilities management of the base station machine room, manual inspection is needed, time and labor are wasted in manual inspection, the subjective view of potential hazards is strong, the standards are different from person to person, and the management of the potential hazards is not facilitated.
Disclosure of Invention
The application provides a potential safety hazard identification method and device for a base station machine room, a computer readable storage medium and electronic equipment.
In a first aspect, an embodiment of the present application provides a method for identifying potential safety hazards of a base station room, including: acquiring an image to be identified, which is obtained by shooting the interior of a base station room by a shooting device, wherein the image to be identified comprises a potential safety hazard target; classifying the images to be recognized through a preset hidden danger image classification model to determine the potential safety hazard classes of potential safety hazard targets contained in the images to be recognized, wherein the preset hidden danger image classification model is obtained by training a first preset neural network based on a first sample set corresponding to a plurality of preset potential safety hazard classes containing the potential safety hazard classes; and carrying out potential safety hazard judgment on the potential safety hazard target contained in the image to be recognized through a preset potential safety hazard recognition model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard target contained in the image to be recognized has a potential safety hazard or not, and training a second preset neural network based on a second sample set corresponding to the potential safety hazard category by the preset potential safety hazard recognition model.
In the embodiment, images are classified through a preset potential safety hazard image classification model to obtain potential safety hazard categories corresponding to the images, potential safety hazard judgment is realized on the basis of the preset potential safety hazard identification model corresponding to the potential safety hazard categories, and characteristic differences among the images corresponding to different potential safety hazard types are fully considered, so that whether potential safety hazards exist in potential safety hazard targets contained in the images or not is judged accurately. Meanwhile, the model has relatively uniform potential safety hazard judgment rules, cannot be influenced by subjective factors of people, can objectively realize potential safety hazard judgment, can feed back potential safety hazard conditions of the base station machine room in time, and can more conveniently manage and track the potential safety hazards.
In a possible implementation manner, the first sample set includes a plurality of sample images corresponding to a plurality of preset potential safety hazard categories, and the sample images include potential safety hazard targets corresponding to any one of the plurality of preset potential safety hazard categories; the first set of samples comprises the second set of samples.
In this implementation, in consideration of the limited number of samples and the fact that the same sample can improve the model accuracy to a certain extent, the second preset neural network is trained using samples in the first sample set for training the first preset neural network.
In a possible implementation manner, the first preset neural network is a network obtained by improving a full connection layer in ResNet; the improvement in the fully-connected layer comprises: the full link layer is added with dropout in the training process, an activation function is added between linear transformation of the full link layer, and/or a classification regressor of the full link layer is a LogSoftmax function.
In the implementation mode, due to the fact that the number of samples is limited, an overfitting phenomenon of a model is easily caused, and therefore the accuracy of image classification is ensured by improving the full connection layer.
In one possible implementation, the activation function added between the linear transformations of the fully-connected layer is a ReLu function.
In one possible implementation, the number of convolution layers of the first preset neural network is 15 or 33.
Considering that the number of convolution layers is larger, the extracted information is larger, the network structure is more complex, and considering that the number of samples is smaller, therefore, the accuracy of image classification can be ensured by presetting the number of convolution layers of the neural network to be 15 or 33.
In a possible implementation manner, the second preset neural network is a network formed by replacing the residual block in the first preset neural network with a lightweight residual network.
In the implementation mode, in order to reduce the data volume, the preset neural network is improved by adopting the lightweight residual error network, so that the calculation amount is reduced, and the calculation efficiency is ensured.
In one possible implementation, the lightweight residual network is ShuffleNet-v2
In a possible implementation manner, the classifying the image to be recognized through a preset hidden danger image classification model to determine a potential safety hazard class of a potential safety hazard target included in the image to be recognized includes: inputting the image to be identified into the preset hidden danger image classification model to obtain the probability that the image to be identified output by the preset hidden danger image classification model corresponds to each preset hidden danger category in the plurality of preset hidden danger categories; and determining the potential safety hazard class corresponding to the maximum probability as the potential safety hazard class to which the potential safety hazard target contained in the image to be identified belongs.
In the implementation mode, the number of samples is limited, meanwhile, the model can classify images corresponding to multiple potential safety hazard categories, and in order to ensure the accuracy of classification, the preset potential safety hazard image classification model only takes the potential safety hazard category corresponding to the maximum probability as the potential safety hazard category to which the potential safety hazard target contained in the image to be identified belongs.
In a possible implementation manner, the acquiring an image to be recognized, which is obtained by shooting the inside of the base station room by the shooting device, includes: acquiring an image obtained by shooting the interior of a base station room by a shooting device; scaling or expanding the image in an equal proportion so that the side length of the shortest side of the scaled or expanded image is the maximum side length of the size of the input image of the preset hidden danger image classification model; performing center cutting on the zoomed or enlarged image to enable the size of the cut image to be in accordance with the size of the input image of the preset hidden danger image classification model; and determining the cut image as an image to be identified.
In the implementation mode, the image is cut after being enlarged in equal proportion, so that the image is ensured to meet the size requirement of model output on the premise of reducing information loss.
In one possible implementation manner, the preset potential safety hazards are any of plugging through-wall holes, water leakage on wall surfaces, damage to doors or door locks, sanitation and mess.
In a second aspect, an embodiment of the present application provides a potential safety hazard identification device for a base station room, including: the image acquisition module is used for acquiring an image to be identified, which is obtained by shooting the interior of the base station machine room by the shooting device; the hidden danger classification module is used for classifying the images to be recognized through a preset hidden danger image classification model so as to determine the potential safety hazard class of a potential safety hazard target contained in the images to be recognized, and the preset hidden danger image classification model is obtained by training a first preset neural network based on a first sample set corresponding to a plurality of preset potential safety hazard classes containing the potential safety hazard class; and the hidden danger identification module is used for judging the potential safety hazard of the potential safety hazard target contained in the image to be identified through a preset potential safety hazard identification model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard target contained in the image to be identified has the potential safety hazard, and the preset potential safety hazard identification model is obtained by training a second preset neural network based on a second sample set corresponding to the potential safety hazard category.
In a third aspect, the present application provides a computer-readable storage medium comprising executable instructions which, when executed by a processor of an electronic device, perform the method according to any one of the first aspect.
In a fourth aspect, the present application provides an electronic device comprising a processor and a memory storing execution instructions, wherein when the processor executes the execution instructions stored in the memory, the processor performs the method according to any one of the first aspect.
Further effects of the above-mentioned unconventional preferred modes will be described below in conjunction with specific embodiments.
Drawings
In order to more clearly illustrate the embodiments or prior art solutions of the present application, the drawings needed for describing the embodiments or prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present application, and that other drawings can be obtained by those skilled in the art without inventive exercise.
Fig. 1 is a schematic flow chart of a method for identifying potential safety hazards in a base station room according to an embodiment of the present application;
FIG. 2a is a schematic view of a sanitation mess provided by an embodiment of the present application;
fig. 2b is a schematic view of a wall surface leaking water according to an embodiment of the present disclosure;
FIG. 2c is a schematic view of a door lock according to an embodiment of the present application showing a broken door lock;
fig. 2d is a schematic view of a hole plugging provided in the present embodiment;
fig. 3 is a schematic diagram of a ResNet structural parameter provided in an embodiment of the present application;
FIG. 4 is a schematic diagram of structural parameters of ShuffleNet-v2 provided in an embodiment of the present application;
FIG. 5a is a first schematic diagram of the structure of ShuffleNet-v2 provided in the embodiments of the present application;
FIG. 5b is a second schematic diagram of the structure of ShuffleNet-v2 provided in the embodiments of the present application;
fig. 6 is a schematic structural diagram of a potential safety hazard identification apparatus for a base station room according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be described in detail and completely with reference to the following embodiments and accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The base station machine room may have potential safety hazards, for example, the potential safety hazards may be sanitation messy, wall surface water leakage, door or door lock damage, incomplete blocking of through-wall holes and the like, and the potential safety hazards may be determined by combining actual conditions. For example, the image shown in fig. 2a can be referred to for hygienic mess, and the ground has more sundries, such as cigarette ends, dust, branches, stones, etc.; the wall surface leakage can refer to the image shown in fig. 2b, the wall surface in the oval area in the image has cracks, peeling or stains, and when the wall surface has cracks or peeling, the wall surface is likely to leak water; door lock damage may refer to the image shown in fig. 2 c; the through-wall hole may not be completely blocked, referring to the image shown in fig. 2d, in actual reference, the through-wall hole may not be completely blocked, or the blocked through-wall hole may not be tight, resulting in light leakage, which is a possible hidden danger.
The conventional method is to regularly inspect according to a classified inspection system, manually record the potential safety hazard conditions in a base station machine room, maintain the machine room, a switching power supply, a storage battery, mobile communication equipment, transmission equipment, an optical cable, an integrated cabinet, an air conditioner, a distribution box, a power supply monitoring system, a fire extinguisher and the like, which are important equipment and facilities, and ensure the normal operation of the equipment and the facilities. This kind of mode usually needs the manual work of field maintenance personnel back to the scene to inspect and record, leads to the potential safety hazard condition of base station room not in time to feed back the renewal to judge the standard differs, leads to the hidden danger problem to fail to carry out the unified follow-up management of closed loop, the base station room equipment facility trouble or the damage that all kinds of hidden dangers of preconditioning probably lead to.
In order to solve the problems, the embodiment of the application provides a potential safety hazard identification method for a base station room, the method classifies images through a preset potential safety hazard image classification model to obtain potential safety hazard types corresponding to the images, and potential safety hazard judgment is realized based on the preset potential safety hazard identification model corresponding to the potential safety hazard types.
As shown in fig. 1, a method for identifying a potential safety hazard of a base station room according to an embodiment of the present application is provided. The method provided by the embodiment of the application can be applied to electronic equipment, such as a server, a mobile phone or a general computer. In this embodiment, an electronic device is used as an execution subject to be described, and the method specifically includes the following steps:
In this step, the electronic device obtains an image to be identified, which is obtained by shooting the interior of the base station room by the shooting device. In one example, a shooting device is arranged on the electronic device, that is, the shooting device is arranged integrally with the electronic device, and the electronic device can control the shooting device to shoot pictures or videos inside the base station room at regular time, and determine an image to be identified from the pictures or videos shot by the shooting device. In one example, the shooting device is separately arranged from the electronic equipment, and the communication device is arranged on the electronic equipment, so that the electronic equipment can communicate with the shooting device through the communication device, and thus, a picture or a video shot by the shooting device to the inside of the base station room is obtained. Illustratively, the photographing apparatus may be a device having a photographing function, such as a mobile phone; for example, the photographing device may be a device with a photographing function, such as a camera, fixedly installed at different corners inside the base station room. In one example, the camera may store the captured picture or video in a database or an external memory. The external storage device may include a floppy disk, a removable hard disk, a usb disk, and the like, which is not limited herein. The database can be arranged on the electronic equipment and can also be arranged on other electronic equipment. Correspondingly, the electronic equipment acquires the image or the video through the database or the external memory, so as to acquire the image to be identified. It should be understood that if the electronic device acquires a video, the video needs to be decimated to acquire an image to be identified.
In practical application, the image to be recognized usually includes one potential safety hazard target, and certainly may also include a plurality of potential safety hazard targets, where the potential safety hazard targets indicate objects which have potential safety hazards and can be recorded by the shooting device in the base station room, such as SDH equipment, an optical cable, a storage battery, BTS main equipment, an air conditioner, an anti-theft door, the ground, terrestrial sundries, a wall surface, a wall through hole and the like, and the specific needs are determined by combining with actual needs, and no specific limitation is made here. In addition, a plurality of shooting devices corresponding to the potential safety hazard targets can be arranged in the machine room, so that one shooting device can shoot one potential safety hazard target in the visual field at regular time to obtain an image to be identified. Subsequently, when hidden danger management needs to be performed on newly-added potential safety hazard targets, shooting devices corresponding to the newly-added potential safety hazard targets can be added. Of course, in order to reduce the cost, one photographing device may be used to photograph multiple potential safety hazard targets, but the images need to be cut subsequently to obtain the images to be recognized corresponding to each potential safety hazard target. It should be understood that the shooting device shoots at a fixed angle, and at this time, if the positions of the shooting device and the plurality of potential safety hazard targets are fixed, the positions of the potential safety hazard targets in the image are also unchanged, so that the image cropping can be realized as long as the positions of the plurality of potential safety hazard targets in the image are preset, and the image cropping can be realized quickly without target detection.
In one embodiment, the electronic equipment acquires an image obtained by shooting the inside of a base station room by a shooting device; scaling or expanding the image in equal proportion to enable the side length of the shortest side of the scaled or expanded image to be the maximum side length of the size of the input image of the preset hidden danger image classification model; performing center cutting on the zoomed or enlarged image to enable the size of the cut image to be in accordance with the size of an input image of a preset hidden danger image classification model; and determining the cut image as an image to be identified.
In this implementation, the image may be scaled up or down by prior art means, such as bilinear interpolation, convolutional neural networks, etc., where bilinear interpolation is preferred in view of data processing speed while preserving more image information. The following description will take the size 224 × 224 of the input image of the preset hidden danger image classification model as an example.
Firstly, the pixels of the image are automatically adjusted and are interpolated according to bilinear
The size of the pixel. Wherein HoriginalRepresents the height of the image; wOriginalRepresents the width of the image;
wherein Q is the original image, and f (x, y) is the interpolated image.
Then, center clipping is performed again to (224 × 224) pixel size:
In this step, considering that the difference between different potential safety hazards of the base station room is large, the difference between the images corresponding to the different captured potential safety hazard categories is large, in order to accurately judge whether the potential safety hazard target contained in the image to be recognized has a potential safety hazard, the image to be recognized needs to be classified first, subsequently, the preset potential safety hazard recognition model corresponding to the potential safety hazard category of the potential safety hazard target contained in the image to be recognized is selected to judge the potential safety hazard, and whether the potential safety hazard target contained in the image to be recognized has a potential safety hazard is determined. It should be understood that the potential safety hazard category in the embodiment of the present application is only a prejudgment on a potential safety hazard that may occur in a potential safety hazard target included in an image to be recognized, and cannot indicate that a potential safety hazard exists in a potential safety hazard target included in an image to be recognized.
In an example, the first sample set includes a plurality of sample images corresponding to a plurality of preset potential safety hazard categories, where the sample images may include a plurality of images with potential safety hazards corresponding to the plurality of preset potential safety hazard categories, and further may include a plurality of images without potential safety hazards corresponding to the plurality of preset potential safety hazard categories, so as to ensure accuracy of model training based on the first sample set. The model training process is prior art and will not be described herein in any greater detail. Correspondingly, a preset hidden danger image classification model is used for learning global features. Specifically, the preset potential safety hazard categories include any two or more of plugging of a through-wall hole, water leakage of a wall surface, damage of a door or a door lock and sanitation disorder, and correspondingly, the potential safety hazards can be that the through-wall hole is not plugged, the door or the door lock is damaged, and the sanitation disorder is caused. Specifically, the potential safety hazard category of the image to be identified is any one of wall through hole plugging, wall surface water leakage, door or door lock damage and sanitation mess.
In one example, the first predetermined neural network is a network modified from the fully connected layer in the ResNet. An overview of the ResNet structure is given in FIG. 3. The forward reasoning in the ResNet network dictates the flow direction of the network data:
(1) after entering the network, the data firstly passes through an input part (conv1, BN1, ReLu, Max Pool);
(2) then enter the intermediate convolution section (conv2, conv3, conv4, conv 5);
(3) and finally, outputting the data through an Average pooling (Average Pool) and a full connection layer (fc) to obtain a result.
The number of potential safety hazard categories in the embodiment of the application is small, and meanwhile, considering that the training sample of the model is small, the fc layer in the ResNet network is reconstructed, and the reconstruction points are as follows:
(1) dropout is added in the fc layer. Dropout is added to the fc layer because effective samples which can be collected at the early stage are limited and overfitting is easily caused in model training. Dropout may be selected as a type of buck for training the convolutional neural network, and over-fitting may be significantly reduced by ignoring half of the feature detectors (leaving half of the hidden layer nodes in the hidden layer with 0 values) in each training batch. It should be noted that the fc layer in the trained network does not need to be complete by Dropout, in other words, the fc layer of the preset hidden danger image classification model is complete.
(2) The activation function is added between the linear transformations of the fc layer. Since the expression ability of the linear model is insufficient, the activation function is increased to add a non-linear factor. Alternatively, the activation function selects the ReLu function, which is formulated as follows:
according to the formula, the ReLu function is a piecewise function, the effect is similar to that of dividing and folding space, and the problem that the gradient disappears in the training process can be solved to a certain degree.
It should be understood that the linear transformation refers to multiplication of an input vector of the fc layer by a weight matrix, or multiplication of an output of a hidden layer in the fc layer by a weight matrix.
(3) Selecting LogSoftmax function by classification regressor in fc layer
The expression of the Softmax function is as follows:
the expression of the LogSoftmax function is as follows:
because the values obtained by Softmax are all in the interval of 0, 1, the value range after log evaluation is negative infinity to 1. LogSoftmax changes the complex exponential operation of Softmax into addition and subtraction operation, efficiency is improved, the probability obtained by exp operation calculation is more than a decimal point, data effectiveness is affected, invalid 0 place occupation is effectively removed by adopting a log mode, and digital effectiveness is improved.
It should be noted that, in the embodiment of the present application, the potential safety hazard categories are mutually exclusive attribute relationships, so softmax is selected as a classification regressor. It takes a vector of real numbers and returns a probability distribution, defining x as a vector of real numbers (neither positive nor negative is so-called, without limitation). Each softmax value output by the classification regressor is a probability distribution, all softmax values are non-negative, and the sum of all softmax values is 1.
In addition, in order to ensure model accuracy in consideration of the limited samples, the first predetermined neural network is a network obtained by performing the above-described improvement on the full connection layer (fc) in ResNet18 or ResNet 34.
In one example, an image to be recognized is input into a preset hidden danger image classification model to obtain the probability that an image to be identified output by the preset hidden danger image classification model corresponds to each preset hidden danger category in a plurality of preset hidden danger categories; and determining the potential safety hazard class corresponding to the maximum probability as the potential safety hazard class to which the potential safety hazard target contained in the image to be identified belongs.
Specifically, the calculation result of the preset hidden danger image classification model is an N-dimensional vector, and each dimension corresponds to the probability of one potential safety hazard category. Illustratively, if the network structure of the preset hidden danger image classification model is obtained by training the first preset neural network, exp operation is performed on the calculation result, and the probability is mapped back to the interval of [0, 1] to represent the confidence of each potential safety hazard category. Considering that the preset hidden danger image classification model can identify a plurality of hidden danger categories and the number of the first sample sets is limited, the preset hidden danger image classification model has a large error, and the hidden danger category corresponding to the maximum confidence degree is used as the hidden danger category to which the hidden danger target contained in the image to be identified belongs, so that the accuracy of image classification is ensured.
103, carrying out potential safety hazard judgment on the potential safety hazard target contained in the image to be recognized through a preset potential safety hazard recognition model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard target contained in the image to be recognized has a potential safety hazard or not, wherein the preset potential safety hazard recognition model is obtained by training a second preset neural network based on a second sample set corresponding to the potential safety hazard category.
In the step, the difference between the shot images corresponding to different potential safety hazard types is considered to be large, and in order to accurately judge whether the potential safety hazard target contained in the image to be recognized has the potential safety hazard, the potential safety hazard judgment is carried out on the image to be recognized based on the preset potential safety hazard image classification model corresponding to the potential safety hazard type of the image to be recognized so as to determine whether the potential safety hazard target contained in the image to be recognized has the potential safety hazard. It should be understood that the preset potential safety hazard identification models corresponding to different potential safety hazard categories are different, in other words, the potential safety hazard identification models are used for learning local features of potential safety hazard targets in the images, and the potential safety hazard targets in the images are represented in more detail.
In order to improve the model performance on the basis of ensuring the precision, a preset potential safety hazard identification model corresponding to the potential safety hazard class trains a second preset neural network on the basis of a second sample set corresponding to the potential safety hazard class; the second preset neural network is a network formed by replacing the residual block in the first preset neural network (the network obtained by improving the full connection layer in ResNet) with a lightweight residual network, so that the calculation amount is reduced, and the calculation efficiency is improved. More specifically, each 3 × 3 convolutional neural network in the first preset neural network is replaced with a lightweight residual network, for example, each 3 × 3 convolutional neural network in the first preset neural network is replaced with a network shown in fig. 5a or 5 b. The description of the network shown in fig. 5a or fig. 5b refers to the description below.
In one example, the lightweight residual network can be a ShuffleNet-v2 network, see FIG. 4 for an overview of the structure of ShuffleNet-v 2. The core of ShuffleNet-v2 is the following operation: channel Convolution (DWConv), point Convolution (1 × 1Conv) and channel shuffle (channel split) and channel split, which greatly reduces the calculation amount of the model while maintaining the precision. The ResNet network uses the residual block to improve the computational efficiency, and under the same computational expense, the ShuffleNet-v2 has more characteristic diagram channel quantity and can encode more information, which is particularly important for the performance of a very small network.
Referring to FIG. 5a, ShuffleNet-v2 introduces a new operation: the input feature map is initially divided into two branches in the channel dimension: the number of channels is c and c-c ', and c' is c/2 in practical implementation. The left branch is mapped identically, the right branch contains 3 consecutive convolutions, and the input and output channels are identical. Moreover, the two 1 × 1 convolutions are no longer group convolution, and the other two branches are equivalent to having been divided into two groups. The outputs concat of the two branches are taken together, followed by a channle shuffle of the results of the two branches to guarantee that the two branches are in communication. In fact, concat and channle shuffle can be combined with channle split of the next module unit to form an element-level operation. For the downsampling module, there is no longer channle split, but each branch is directly copy-input, each branch has a stride of 2 downsampling, and after last concat is connected together, the size of the feature map space is reduced by half, but the number of channels is doubled. In addition, the structure shown in FIG. 5b can also be adopted for ShuffleNet-v 2. Bn (batch normalization) shown in fig. 5a and 5b represents batch normalization, and ReLu represents an activation function ReLu function.
It should be understood that the above-mentioned model training process is prior art and will not be described in detail herein.
In one example, the second sample set is an image in the first sample set described above. On the one hand, considering the limited number of samples, on the other hand, considering the model training by the same samples can improve the accuracy.
According to the technical scheme, the beneficial effects of the embodiment are as follows:
the images are classified through the preset hidden danger image classification model to obtain the potential safety hazard types corresponding to the images, the potential safety hazards are judged based on the preset potential safety hazard identification model corresponding to the potential safety hazard types, and the characteristic difference between the images corresponding to different potential safety hazard types is fully considered, so that whether the potential safety hazard targets contained in the images have potential safety hazards or not is judged accurately. Meanwhile, the model adopts a unified rule to judge the potential safety hazard, and can feed back the potential safety hazard condition of the base station machine room in time, so that the potential safety hazard can be managed and tracked more conveniently.
Based on the same concept as the method embodiment of the present application, please refer to fig. 6, an embodiment of the present application further provides a device for identifying a potential safety hazard of a base station room, including:
the image acquisition module 601 is used for acquiring an image to be identified, which is obtained by shooting the interior of the base station room by the shooting device;
the hidden danger classification module 602 is configured to classify the image to be recognized through a preset hidden danger image classification model to determine a potential safety hazard class of a potential safety hazard target included in the image to be recognized, where the preset hidden danger image classification model is obtained by training a first preset neural network based on a first sample set corresponding to a plurality of preset potential safety hazard classes including the potential safety hazard class;
the hidden danger identification module 603 is configured to perform hidden danger judgment on a hidden danger target contained in the image to be identified through a preset hidden danger identification model corresponding to the hidden danger category, so as to determine whether a hidden danger exists in the hidden danger target contained in the image to be identified, and the preset hidden danger identification model is obtained by training a second preset neural network based on a second sample set corresponding to the hidden danger category.
In a possible implementation manner, the first sample set includes a plurality of sample images corresponding to a plurality of preset potential safety hazard categories, and the sample images include potential safety hazard targets corresponding to any one of the plurality of preset potential safety hazard categories; the first set of samples comprises the second set of samples.
In a possible implementation manner, the first preset neural network is a network obtained by improving a full connection layer in ResNet; the improvement in the fully-connected layer comprises: the full link layer is added with dropout in the training process, an activation function is added between linear transformation of the full link layer, and/or a classification regressor of the full link layer is a LogSoftmax function.
In one possible implementation, the activation function added between the linear transformations of the full connection layer is a ReLu function;
in one possible implementation, the number of convolution layers of the first preset neural network is 15 or 33.
In a possible implementation manner, the second preset neural network is a network formed by replacing the residual block in the first preset neural network with a lightweight residual network.
In one possible implementation, the lightweight residual network is ShuffleNet-v 2.
In one possible implementation, the first set of samples includes the second set of samples.
In a possible implementation manner, the hidden danger classification module 602 includes: a probability obtaining unit and a category determining unit; wherein the content of the first and second substances,
the probability obtaining unit is configured to input the image to be identified into the preset hidden danger image classification model to obtain a probability that the image to be identified output by the preset hidden danger image classification model corresponds to each preset hidden danger category in the plurality of preset hidden danger categories;
the category determining unit is configured to determine the potential safety hazard category corresponding to the maximum probability as the potential safety hazard category to which the potential safety hazard target included in the image to be identified belongs.
In a possible implementation manner, the image obtaining module 601 includes: the device comprises an image acquisition unit, a zooming and expanding unit and a cutting unit; wherein the content of the first and second substances,
the image acquisition unit is used for acquiring an image obtained by shooting the interior of the base station room by the shooting device;
the scaling and expanding unit is used for scaling or expanding the image in an equal proportion so that the side length of the shortest side of the scaled or expanded image is the maximum side length of the size of the input image of the preset hidden danger image classification model;
the cutting unit is used for cutting the zoomed or enlarged image to make the size of the cut image accord with the size of the input image of the preset hidden danger image classification model; and determining the cut image as an image to be identified.
In one possible implementation manner, the preset potential safety hazards are any of plugging through-wall holes, water leakage on wall surfaces, damage to doors or door locks, sanitation and mess.
Fig. 7 shows a schematic structural diagram of an electronic device provided in an embodiment of the present application. The potential safety hazard identification device 400 of the base station room can be used for implementing the potential safety hazard identification method of the base station room described in the above method embodiment. The electronic device may be a chip, a terminal, a server, or a device capable of data processing, etc.
The electronic device includes one or more processors 701, and the one or more processors 701 may implement the method for identifying a security risk of a base station room described in this embodiment of the application, for example, the method shown in fig. 1. The processor 701 executes an execution instruction stored in the memory, so as to implement the method for identifying the potential safety hazard of the base station room provided in any embodiment of the present application through the executed execution instruction.
Optionally, the electronic device may include one or more memories 702, on which programs (which may also be instructions or codes) are stored, and the programs may be executed by the processor 701, so that the processor 701 executes the methods described in the above method embodiments. Optionally, data may also be stored in the memory 702. Alternatively, the processor 701 may also read data (for example, a plurality of preset potential safety hazard categories, a preset potential safety hazard image classification model, and a preset potential safety hazard identification model corresponding to each of the plurality of preset potential safety hazard categories) stored in the memory 702, where the data may be stored at the same storage address as the program, and the data may also be stored at a different storage address from the program. The processor 701 and the memory 702 may be provided separately or integrated together, for example, on a single board or a System On Chip (SOC). The Memory 702 may include a Memory 7021, such as a Random-Access Memory (RAM), and may also include a non-volatile Memory 7022 (e.g., at least 1 disk Memory). When the processor 701 executes the execution instructions stored in the memory 702, the processor 701 executes the method in any of the embodiments of the present application, and at least is configured to execute the method as shown in fig. 1.
Optionally, the electronic device further comprises an internal bus 703 and a network interface 704. The processor 701, the network interface 704, and the memory 702 may be connected to each other through an internal bus 703, where the internal bus 703 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like; the internal bus 703 may be divided into an address bus, a data bus, a control bus, etc., which is indicated by a double-headed arrow in fig. 7 for convenience of illustration, but does not indicate only one bus or one type of bus. Of course, the electronic device may also include hardware required for other services.
It should be understood that the steps of the above-described method embodiments may be performed by logic circuits in the form of hardware or instructions in the form of software in the processor 701. The processor 701 may be a CPU, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic device, such as a discrete gate, a transistor logic device, or a discrete hardware component.
The embodiment of the present application further provides a computer-readable storage medium, which includes an execution instruction, and when a processor of an electronic device executes the execution instruction, the processor executes the method provided in any embodiment of the present application. The electronic device may specifically be the electronic device shown in fig. 7; the execution instruction is a computer program corresponding to the potential safety hazard identification device of the base station room.
It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects.
The embodiments in the present application are described in a progressive manner, and the same and similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, as for the apparatus embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. A potential safety hazard identification method for a base station room is characterized by comprising the following steps:
acquiring an image to be identified, which is obtained by shooting the interior of a base station room by a shooting device, wherein the image to be identified comprises a potential safety hazard target;
classifying the images to be recognized through a preset hidden danger image classification model to determine the potential safety hazard classes of potential safety hazard targets contained in the images to be recognized, wherein the preset hidden danger image classification model is obtained by training a first preset neural network based on a first sample set corresponding to a plurality of preset potential safety hazard classes containing the potential safety hazard classes;
and carrying out potential safety hazard judgment on the potential safety hazard target contained in the image to be recognized through a preset potential safety hazard recognition model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard target contained in the image to be recognized has a potential safety hazard or not, and training a second preset neural network based on a second sample set corresponding to the potential safety hazard category by the preset potential safety hazard recognition model.
2. The method according to claim 1, wherein the first sample set comprises a plurality of sample images corresponding to a plurality of preset potential safety hazard categories, and the sample images comprise potential safety hazard targets corresponding to any one of the plurality of preset potential safety hazard categories;
the first set of samples comprises the second set of samples.
3. The method of claim 1, wherein the first predetermined neural network is a network modified for a full connectivity layer in ResNet;
the improvement in the fully-connected layer comprises: the full link layer is added with dropout in the training process, an activation function is added between linear transformation of the full link layer, and/or a classification regressor of the full link layer is a LogSoftmax function.
4. The method of claim 3, wherein the activation function added between the linear transformations of the fully-connected layer is a ReLu function;
the number of convolution layers of the first preset neural network is 15 or 33.
5. The method of claim 3, wherein the second predetermined neural network is a network formed by replacing the residual blocks in the first predetermined neural network with a lightweight residual network.
6. The method of claim 5, wherein the lightweight residual network is ShuffleNet-v 2.
7. The method according to claim 1, wherein the classifying the image to be recognized through a preset hidden danger image classification model to determine a security hidden danger category of a security hidden danger target included in the image to be recognized comprises:
inputting the image to be identified into the preset hidden danger image classification model to obtain the probability that the image to be identified output by the preset hidden danger image classification model corresponds to each preset hidden danger category in the plurality of preset hidden danger categories;
and determining the potential safety hazard class corresponding to the maximum probability as the potential safety hazard class to which the potential safety hazard target contained in the image to be identified belongs.
8. The method according to claim 1, wherein the acquiring the image to be identified, which is obtained by shooting the inside of the base station room by the shooting device, comprises:
acquiring an image obtained by shooting the interior of a base station room by a shooting device;
scaling or expanding the image in an equal proportion so that the side length of the shortest side of the scaled or expanded image is the maximum side length of the size of the input image of the preset hidden danger image classification model;
performing center cutting on the zoomed or enlarged image to enable the size of the cut image to be in accordance with the size of the input image of the preset hidden danger image classification model;
and determining the cut image as an image to be identified.
9. The method of claims 1-8, wherein the plurality of predetermined potential safety hazard categories are any plurality of through-wall hole plugging, wall surface water leakage, door or door lock damage, sanitation mess.
10. The utility model provides a potential safety hazard recognition device of base station computer lab which characterized in that includes:
the image acquisition module is used for acquiring an image to be identified, which is obtained by shooting the interior of the base station machine room by the shooting device;
the hidden danger classification module is used for classifying the images to be recognized through a preset hidden danger image classification model so as to determine the potential safety hazard class of a potential safety hazard target contained in the images to be recognized, and the preset hidden danger image classification model is obtained by training a first preset neural network based on a first sample set corresponding to a plurality of preset potential safety hazard classes containing the potential safety hazard class;
and the hidden danger identification module is used for judging the potential safety hazard of the potential safety hazard target contained in the image to be identified through a preset potential safety hazard identification model corresponding to the potential safety hazard category so as to determine whether the potential safety hazard target contained in the image to be identified has the potential safety hazard, and the preset potential safety hazard identification model is obtained by training a second preset neural network based on a second sample set corresponding to the potential safety hazard category.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110743899.6A CN113361637A (en) | 2021-06-30 | 2021-06-30 | Potential safety hazard identification method and device for base station room |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110743899.6A CN113361637A (en) | 2021-06-30 | 2021-06-30 | Potential safety hazard identification method and device for base station room |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113361637A true CN113361637A (en) | 2021-09-07 |
Family
ID=77537803
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110743899.6A Pending CN113361637A (en) | 2021-06-30 | 2021-06-30 | Potential safety hazard identification method and device for base station room |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113361637A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024087275A1 (en) * | 2022-10-28 | 2024-05-02 | 清华大学 | Method and apparatus for checking potential safety hazard in industrial scene, and device and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929847A (en) * | 2019-11-15 | 2020-03-27 | 国网浙江省电力有限公司电力科学研究院 | Converter transformer fault diagnosis method based on deep convolutional neural network |
| CN110956086A (en) * | 2019-10-24 | 2020-04-03 | 国家电网有限公司 | Machine room equipment online identification method based on image identification technology |
| CN111507381A (en) * | 2020-03-31 | 2020-08-07 | 上海商汤智能科技有限公司 | Image recognition method and related device and equipment |
| CN111709404A (en) * | 2020-08-05 | 2020-09-25 | 广东电网有限责任公司 | Method, system and equipment for identifying machine room remnants |
| US20200394442A1 (en) * | 2019-06-14 | 2020-12-17 | International Business Machines Corporation | Screening technique for prohibited objects at security checkpoints |
| CN112418214A (en) * | 2020-11-06 | 2021-02-26 | 北京五八信息技术有限公司 | Vehicle identification code identification method and device, electronic equipment and storage medium |
-
2021
- 2021-06-30 CN CN202110743899.6A patent/CN113361637A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200394442A1 (en) * | 2019-06-14 | 2020-12-17 | International Business Machines Corporation | Screening technique for prohibited objects at security checkpoints |
| CN110956086A (en) * | 2019-10-24 | 2020-04-03 | 国家电网有限公司 | Machine room equipment online identification method based on image identification technology |
| CN110929847A (en) * | 2019-11-15 | 2020-03-27 | 国网浙江省电力有限公司电力科学研究院 | Converter transformer fault diagnosis method based on deep convolutional neural network |
| CN111507381A (en) * | 2020-03-31 | 2020-08-07 | 上海商汤智能科技有限公司 | Image recognition method and related device and equipment |
| CN111709404A (en) * | 2020-08-05 | 2020-09-25 | 广东电网有限责任公司 | Method, system and equipment for identifying machine room remnants |
| CN112418214A (en) * | 2020-11-06 | 2021-02-26 | 北京五八信息技术有限公司 | Vehicle identification code identification method and device, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 董洪义: "《深度学习之PyTorch物体检测实战》", 31 January 2020, 北京：机械工业出版社 * |
| 陈云霁: "《智能计算***》", 29 February 2020, 北京：机械工业出版社 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024087275A1 (en) * | 2022-10-28 | 2024-05-02 | 清华大学 | Method and apparatus for checking potential safety hazard in industrial scene, and device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543627B (en) | Method and device for judging driving behavior category and computer equipment | |
| CN111539370A (en) | Image pedestrian re-identification method and system based on multi-attention joint learning | |
| CN111311475A (en) | Detection model training method and device, storage medium and computer equipment | |
| CN111274987B (en) | Facial expression recognition method and facial expression recognition device | |
| CN114821390B (en) | Method and system for tracking twin network target based on attention and relation detection | |
| CN110837786B (en) | Density map generation method and device based on spatial channel, electronic terminal and medium | |
| CN110942456B (en) | Tamper image detection method, device, equipment and storage medium | |
| Jiang et al. | A self-attention network for smoke detection | |
| CN108229495B (en) | Target object detection method and device, electronic equipment and storage medium | |
| CN110910360A (en) | Power grid image positioning method and image positioning model training method | |
| CN111369557B (en) | Image processing method, device, computing equipment and storage medium | |
| CN104751164A (en) | Method and system for capturing movement trajectory of object | |
| CN116129353B (en) | Method and system for intelligent monitoring based on image recognition | |
| Wang | A survey on IQA | |
| Hsu et al. | Industrial smoke detection and visualization | |
| CN112949459A (en) | Smoking image recognition method and device, storage medium and electronic equipment | |
| CN114898416A (en) | Face recognition method and device, electronic equipment and readable storage medium | |
| CN115578616A (en) | Training method, segmentation method and device of multi-scale object instance segmentation model | |
| CN110866473B (en) | Target object tracking detection method and device, storage medium and electronic device | |
| CN113361637A (en) | Potential safety hazard identification method and device for base station room | |
| CN114331837A (en) | Method for processing and storing panoramic monitoring image of protection system of extra-high voltage converter station | |
| CN116681687B (en) | Wire detection method and device based on computer vision and computer equipment | |
| CN114170425A (en) | Model training method, image classification method, server and storage medium | |
| CN112348011B (en) | Vehicle damage assessment method and device and storage medium | |
| CN112884866A (en) | Coloring method, device, equipment and storage medium for black and white video |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210907 |
|
| RJ01 | Rejection of invention patent application after publication |