CN117831131B - Compression method of typical violation intelligent recognition algorithm based on convolutional neural network - Google Patents
Compression method of typical violation intelligent recognition algorithm based on convolutional neural network Download PDFInfo
- Publication number
- CN117831131B CN117831131B CN202410026886.0A CN202410026886A CN117831131B CN 117831131 B CN117831131 B CN 117831131B CN 202410026886 A CN202410026886 A CN 202410026886A CN 117831131 B CN117831131 B CN 117831131B
- Authority
- CN
- China
- Prior art keywords
- images
- image
- size
- layer
- module
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 27
- 238000007906 compression Methods 0.000 title claims abstract description 11
- 230000006835 compression Effects 0.000 title claims abstract description 11
- 230000006399 behavior Effects 0.000 claims abstract description 22
- 230000006870 function Effects 0.000 claims description 23
- 238000011176 pooling Methods 0.000 claims description 22
- 230000004913 activation Effects 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 238000007639 printing Methods 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 230000009471 action Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 238000009472 formulation Methods 0.000 claims 1
- 239000000203 mixture Substances 0.000 claims 1
- 238000007781 pre-processing Methods 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 8
- 230000009286 beneficial effect Effects 0.000 abstract description 6
- 230000000694 effects Effects 0.000 abstract description 4
- 238000012806 monitoring device Methods 0.000 abstract description 3
- 230000001502 supplementing effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of safety recognition of power systems, and discloses a compression method of an intelligent recognition algorithm for typical illegal behaviors based on a convolutional neural network, which is characterized by comprising the following steps: based on the convolutional neural network model, the corresponding category of the input image is identified through the characteristics of the input image, so that the illegal behaviors of the staff during the power work are identified. The invention has the following beneficial technical effects: the method has the advantages that whether the operation personnel have illegal behaviors during the near-electricity operation of the transformer substation is identified, the method is beneficial to supplementing the existing identification method, the reasonable accuracy and the high popularization are realized, the size of the model is further reduced under the condition that the accuracy and the speed of intelligent identification are ensured, and the analysis efficiency of the intelligent identification monitoring device is improved.
Description
Technical Field
The invention relates to the technical field of safety identification of power systems, in particular to a compression method of a typical violation intelligent identification algorithm based on a convolutional neural network.
Background
With the rapid development of the economic society in China, the scale of the power grid is rapidly increased, the number of power grid operation equipment is rapidly increased, the power grid operation mode is gradually complicated, and higher requirements are provided for personnel, power grid and equipment safety level and working efficiency. Production safety is a basic guarantee for operation and development of an electric power system.
CN117115916A discloses a power plant illegal behavior recognition method, a system and a network server based on a 5G technology, wherein the method adopts a deep convolutional neural network to construct a plurality of behavior recognition modules, acquires video information through a wireless camera, and acquires position information of the wireless camera; transmitting the video information to a corresponding behavior recognition module according to the acquired position of the wireless camera and a preset rule; the behavior recognition module performs frame extraction recognition and intelligent video analysis on the received video information to obtain violation information; the violation information is remotely sent to an event storage processing module; and the event storage processing module is used for processing the event. The intelligent analysis and recognition system can perform intelligent analysis and recognition on unsafe personnel and illegal behaviors, timely find out the violations, improve anti-violation work efficiency, reduce manpower operation cost and reduce the higher missing report recognition rate of monitoring relying on manpower.
CN116524240a discloses a model, method, device and storage medium for identifying electric power operation scene violation, the model comprises: the convolution layer is used for extracting the characteristics of the input electric power operation scene pictures to obtain a plurality of characteristic pictures; the saliency maximization layer is used for carrying out maximization operation on the plurality of feature graphs; the saliency component positioning learning layer is used for carrying out weighted summation on the pooling result to obtain a summation result; and the classification layer is used for outputting the identification result according to the summation result. The method adopts the maximum saliency pooling layer to replace the full-connection layer in the model, solves the problems of overlarge parameter quantity and easy overfitting in the convolutional neural network, adopts the maximum saliency pooling layer to connect the convolutional layer, directly endows each channel with practical category meaning, eliminates the characteristics of black boxes in the full-connection layer, and ensures that the network structure has interpretability. And a significance component learning layer is arranged, so that the recognition of the illegal behaviors in the power operation scene becomes traceable.
CN116453212a discloses a method for detecting illegal behaviors in a power construction scene, which comprises the following steps: s1, a camera is arranged on a bracket, and communication is established between the camera and a computer terminal through a wireless transmission technology; s2, before construction, an electric worker installs a bracket provided with a camera, adjusts the position of the camera, enables the camera to shoot the position to be monitored, adopts deep learning, transfer learning, convolutional neural network and other technologies to realize accurate recognition of some complex illegal behaviors, improves recognition accuracy, deploys an intelligent recognition algorithm to one end of a working site through marginalization of a computing platform, transmits recognition results to a rear-end server when the illegal operation behaviors are recognized, greatly reduces the network bandwidth and the pressure of the rear-end server, and improves recognition instantaneity.
With the technology of mobile terminals becoming more and more different, a variety of mobile security management and control devices including security management and control balls, law enforcement instruments, video monitoring and smart helmets are increasingly used in the field. However, the current safety control devices mainly play a role in video monitoring and acquisition, and the device basically has no intelligent analysis capability or has intelligent analysis capability, but the acquired video data is seriously dependent on manual checking and needs to be further improved.
Disclosure of Invention
In order to solve the problems, the invention discloses a compression method of a typical violation intelligent recognition algorithm based on a convolutional neural network, which is realized by adopting the following technical scheme.
A compression method of a typical violation intelligent recognition algorithm based on a convolutional neural network is characterized by comprising the following steps: based on the convolutional neural network model, the corresponding category of the input image is identified through the characteristics of the input image, so that the illegal behaviors of the staff during the power work are identified.
The invention has the following beneficial technical effects: the method has the advantages that whether the operation personnel have illegal behaviors during the near-electricity operation of the transformer substation is identified, the method is beneficial to supplementing the existing identification method, the reasonable accuracy and the high popularization are realized, the size of the model is further reduced under the condition that the accuracy and the speed of intelligent identification are ensured, and the analysis efficiency of the intelligent identification monitoring device is improved.
Drawings
FIG. 1 is a first part of a schematic diagram of a convolutional neural network-based recognition method of the present invention.
Fig. 2 is a second part of a schematic diagram of a framework of the convolutional neural network-based recognition method of the present invention.
Fig. 3 is a third part of a schematic diagram of a framework of the convolutional neural network-based recognition method of the present invention.
Fig. 4 is a first portion of a model specific parameter table of the present invention.
FIG. 5 is a second portion of the model specific parameter table of the present invention.
Detailed Description
So that those skilled in the art can better understand and practice the present invention, the present invention will now be described in detail with reference to the accompanying drawings.
Referring to fig. 1 to 5, a compression method of a typical violation intelligent recognition algorithm based on a convolutional neural network is used for recognizing the violation of staff during power operation, and has reasonable accuracy and high popularization, and the specific method comprises the following steps:
the first step: the compression method is characterized in that based on a convolutional neural network model, corresponding categories of an input image are identified through characteristics of the input image, so that the illegal behaviors of staff in power work are identified, and the method is formulated as follows:
For n input images, defined as: i total＝{I1,I2,...,In }, taking the n images as a training set, wherein the images without the violations are defined as I r, the images with the violations are defined as I df, and for p data sources, the images without the violations are represented as I r:
Wherein: representing a set of x non-offending images in the p-th dataset,
Each data source consists of x images of non-violations, denoted N k, and thus I r is further represented as follows:
For the image I df with violations, for q data sources, the image I df with violations is expressed as:
Wherein: representing a set of z images of the presence of violations in the qth dataset,
Each data source consists of z images of presence violations, denoted N j, and thus I df is further represented as follows:
Based on And/>I total is further represented as: /(I)
The labels of the corresponding classes may be defined as: y= [ Y 1,y2,...,ym ]
Where m represents the total number of images in I total, the proposed architecture belongs to the binary classification problem, identified as I r when y=0; when y=1, identified as I df,
And a second step of: for the binary classification model proposed in the first step, deep features are extracted from an input image by using a convolutional neural network, and the algorithm is as follows:
Taking the image as a directory address of the input and stored image, also called as a position of the stored image, outputting a recognition result and printing the processed image, setting the image 'Height' and 'Width' as 160 in the algorithm, then creating ImageDataGenerator objects containing parameters of the required image processing, wherein the created objects are called DataGen, using the DataGen objects, streaming the image one by one according to the parameters of flow_from_direction (), accepting the target address Height and Width of the stored image as parameters, adjusting the input image to a given size, and returning an object named Generator;
Loading a pre-trained model into an object named model, reading images from an input catalog one by a user, taking the images as the input of the model, carrying out image recognition through a convolutional neural network, wherein the recognition range is from 0 to 1, rounding the recognition result to obtain a recognition label, ir is represented by 0, I df is represented by 1, and finally printing the recognition result to the user;
and a third step of: for the neural network model of the second step, it is specifically expressed as follows:
Establishing a CNN structure consisting of a convolution layer and a pooling layer, wherein the convolution layer extracts deep features from an input image, the pooling layer is used for reducing the dimension of an input feature map, after the convolution layer, a flat layer is used for preparing all the feature maps into a one-dimensional array and taking the one-dimensional array as the input of a full-connection layer, after the full-connection layer, an output layer predicts subsequent classes based on the input image, finally, a sigmoid function is used for predicting an output result, and after each layer, the pooling layer is added for reducing the dimension of the feature map on the subsequent layer;
fourth step: for the neural network model of the third step, the details are expressed as follows:
First, an image having a height and width of 160 pixels and a batch size of 64 is input, and then the image is preprocessed, and the image is rescaled into an array, an image position is adjusted, and a scaling process is performed, the image position is adjusted including: rotating, horizontally and vertically turning; obtaining an input image of size (160,160,3);
In the first module, the smaller filter can maintain the advanced features of the image, so that 8 (3, 3) filters are used for 2D convolution operation, and the Leaky ReLU is used as an activation function to extract an initial feature map from the input image, but the distribution of the input batch can be different according to the image types contained in different batches, so that problems are caused to the convergence of an optimizer algorithm, the training process is unstable, and for this reason, the feature map is subjected to batch normalization processing, so that the convergence of training and the dependence of weight initialization are accelerated;
Then, the feature images with the size (160,160,8) are subjected to batch normalization and then transferred to a second module, the module comprises two convolution layers, the convolution layers perform convolution operation by using 16 filters with the size (3, 3), the module can extract image features with deeper layers by using the Leaky ReLU as an activation function, the feature images are subjected to batch normalization again by using the extracted feature images, and the size of the feature images is increased continuously along with the extraction of the features with the deeper layers, so that the calculation amount is increased; to this end, the present invention uses a pooling layer to reduce the dimensionality of the extracted feature map. An average pooling layer of size (2, 2) is used, thereby reducing the dimension of the feature map by half, and the second module obtains a feature map of size (80, 80, 16);
Passing a feature map of size (80, 80, 16) to a third module that takes the same architecture as the second module, except that three consecutive convolution layers of 32 (3, 3) size filters are used, using a leak ReLU as an activation function, and then batch normalization and average pooling layer operations are performed, the feature map size becoming (40, 40, 32);
Passing the feature map of size (40, 40, 32) to a fourth module having 4 layers of 64 filters of size (3, 3) in succession, using the leak ReLU as an activation function, then performing batch normalization and average pooling layer operations, the feature map becoming of size (20, 20, 64);
passing the feature map of size (20, 20, 64) to a fifth module, based on which the extracted depth image features can be used to classify the image as a depth-containing typical offending image, the fifth module performing a convolution operation using the (5, 5) sized filters using 128 (5, 5) sized filters, using the Leaky ReLU as an activation function, and then performing batch normalization and averaging pooling layer operations, the feature map size becoming (10,10,128);
Passing the feature map of size (10,10,128) to a sixth module, the sixth module performing a convolution operation using 256 (5, 5) size filters, using the leak ReLU as an activation function, and then performing batch normalization and average pooling layer operations, the feature map size becoming (5,5,256);
Transferring the feature map with the size of (5,5,256) to a seventh module, converting the output of the sixth module into a one-dimensional array by using a flat layer in the seventh module, and randomly setting half of input units to zero by using a discarding layer with the value of 0.5 after the flat layer, so as to avoid overfitting of training data;
The eighth module uses the full connection layer of 32 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
the ninth module uses the full connection layer of 16 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
the tenth module uses the full connection layer of 16 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
The eleventh module uses the output layer of the single neuron and the S-shaped activation function to identify whether the input image contains typical violations, and if the value is less than 0.5, the output result is identified to have no violations; otherwise, the output result is identified to contain the illegal action.
The loss function used in the training process is binary cross entropy, the used optimizer is Adam, the learning rate is 0.01, and FIGS. 1-3 are frame diagrams of the recognition method based on the convolutional neural network. Fig. 4 and 5 describe the output dimensions of each layer and the number of parameters.
The experimental environment configuration is shown in table 1. The trained batch-size was 64, momentum was 0.9, and decay rate was 0.0005.
Table 1 experimental environment
The size of the prior box is obtained by K-means clustering, and the prior box distribution is shown in Table2 when the input is 160×160. The full convolution network with the step length of 2 is used for realizing downsampling, the gradient negative effect caused by pooling is reduced, and the downsampling is sequentially carried out at 8 times, 16 times and 32 times for detection. In order to use deep features, up-sampling with the step length of 2 is carried out on the feature graphs which are respectively sampled at 32 times and 16 times, so that a new feature graph with the size increased by one time is obtained, and the feature graphs become the dimensions of 16 times and 8 times of down-sampling, thereby the deep features can be used for detection.
Table 2 bounding box distribution
Table 2 Distribution of bounding boxes
To verify the superiority of the methods herein, the accuracy pairs for the same parameters are shown in table 3.
Table 3 algorithm comparison
Table 3 Algorithm Comparison
As can be seen from Table 3, compared with the fast-RCNN algorithm and the PCA-SVM algorithm, the method provided by the invention has higher accuracy, can accurately identify the illegal behaviors of field operators, and reduces the accident occurrence probability.
The invention has the following beneficial technical effects: the method has the advantages that whether the operation personnel have illegal behaviors during the near-electricity operation of the transformer substation is identified, the method is beneficial to supplementing the existing identification method, the reasonable accuracy and the high popularization are realized, the size of the model is further reduced under the condition that the accuracy and the speed of intelligent identification are ensured, and the analysis efficiency of the intelligent identification monitoring device is improved.
The above-described embodiments are only preferred embodiments of the present invention, and should not be construed as limiting the present invention. The protection scope of the present invention is defined by the claims, and the protection scope includes equivalent alternatives to the technical features of the claims. I.e., equivalent replacement modifications within the scope of this invention are also within the scope of the invention.
Claims (3)
1. A compression method of a typical violation intelligent recognition algorithm based on a convolutional neural network is characterized by comprising the following steps: based on the convolutional neural network model, identifying the corresponding category of the input image through the characteristics of the input image, so as to identify the illegal behaviors of the staff during the power work; the formulation is as follows:
For n input images, defined as: i total＝{I1,I2,...,In }, taking the n images as a training set, wherein the images without the violations are defined as I r, the images with the violations are defined as I df, and for p data sources, the images without the violations are represented as I r:
Wherein: representing a set of x non-offending images in a p-th dataset;
Each data source consists of x images of non-offending behavior, denoted as N k,Ir, further denoted as follows:
For the image I df with violations, for q data sources, the image I df with violations is expressed as:
Wherein: representing a set of z images of offending behaviors in the q-th data set;
Each data source consists of z images of the existence of violations, denoted as N j,Idf, further denoted as follows:
Based on And/>
I total is further represented as:
the labels of the corresponding classes may be defined as: y= [ Y 1,y2,...,ym ]
Where m represents the total number of images in I total, and when y=0, is identified as I r; when y=1, it is identified as I df; establishing a CNN structure consisting of a convolution layer and a pooling layer; the method comprises the steps that a convolution layer extracts deep features from an input image, a pooling layer is used for reducing the dimension of the input feature map, a flat layer is used for manufacturing all the feature maps into a one-dimensional array after the convolution layer and is used as the input of a full-connection layer, an output layer predicts subsequent classes based on the input image after the full-connection layer, finally, a sigmoid function is used for predicting an output result, and the pooling layer is added after each layer to reduce the dimension of the feature map on the subsequent layer;
firstly, inputting an image with a height and a width of 160 pixels and a batch size of 64 pixels, preprocessing the image, rescaling the image into an array, adjusting the position of the image and performing scaling processing to obtain an input image with a size of (160,160,3);
In the first module, 8 (3, 3) filters are used for carrying out 2D convolution operation, a Leaky ReLU is used as an activation function, an initial feature map is extracted from an input image, but the distribution of input batches is different according to image types contained in different batches, so that problems are brought to the convergence of an optimizer algorithm, the training process is unstable, the feature map is subjected to batch normalization processing, and therefore the training convergence is accelerated and the dependence of weight initialization is reduced;
Then, the feature images with the size of (160,160,8) are subjected to batch normalization and then transferred to a second module, the module comprises two convolution layers, the convolution layers perform convolution operation by using 16 filters with the size of (3, 3), the module can extract image features with a deeper level, the feature images are subjected to batch normalization again by using the extracted feature images, the size of the feature images is continuously increased along with the extraction of the features with the deeper level, so that the calculation amount is increased, the dimension of the extracted feature images is reduced by using a pooling layer, the dimension of the feature images is reduced by half by using an average pooling layer with the size of (2, 2), and the second module obtains the feature images with the size of (80, 80, 16);
Passing a feature map of size (80, 80, 16) to a third module that takes the same architecture as the second module, except that three consecutive convolution layers of 32 (3, 3) size filters are used, using a leak ReLU as an activation function, and then batch normalization and average pooling layer operations are performed, the feature map size becoming (40, 40, 32);
Passing the feature map of size (40, 40, 32) to a fourth module having 4 layers of 64 filters of size (3, 3) in succession, using the leak ReLU as an activation function, then performing batch normalization and average pooling layer operations, the feature map becoming of size (20, 20, 64);
Passing the feature map of size (20, 20, 64) to a fifth module, based on which the extracted depth image features can be used to classify the image as a depth-containing typical offending image, the fifth module performing a convolution operation using the (5, 5) sized filters using 128 (5, 5) sized filters, using the leak ReLU as an activation function, and then performing batch normalization and average pooling layer operations, the feature map size becoming (10,10,128);
Passing the feature map of size (10,10,128) to a sixth module, the sixth module performing a convolution operation using 256 (5, 5) size filters, using the leak ReLU as an activation function, and then performing batch normalization and average pooling layer operations, the feature map size becoming (5,5,256);
Transferring the feature map with the size of (5,5,256) to a seventh module, converting the output of the sixth module into a one-dimensional array by using a flat layer in the seventh module, and randomly setting half of input units to zero by using a discarding layer with the value of 0.5 after the flat layer, so as to avoid overfitting of training data;
The eighth module uses the full connection layer of 32 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
the ninth module uses the full connection layer of 16 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
the tenth module uses the full connection layer of 16 neurons, uses the leak ReLU as an activation function, and uses the discard layer with a value of 0.5 to randomly set half of the input units to zero;
The eleventh module uses the output layer of the single neuron and the S-shaped activation function to identify whether the input image contains typical violations, and if the value is less than 0.5, the output result is identified to have no violations; otherwise, the output result is identified to contain the illegal action.
2. The compression method of the intelligent recognition algorithm for typical violation behaviors based on the convolutional neural network according to claim 1, which is characterized by comprising the following steps of: deep features are extracted from the input image using a convolutional neural network, with the algorithm as follows:
Taking the image as a catalog address of the input and stored image, outputting the identification result and printing the processed image, wherein in the algorithm, the image 'height' and 'width' are set as 160; then creating ImageDataGenerator objects, wherein the objects contain parameters of required image processing, the created objects are called DataGen, using the DataGen objects, carrying out stream processing on the images one by one according to the parameters of flow_from_direction (), receiving the target addresses Height and Width of the stored images as parameters, adjusting the input images to a given size, and returning an object named as a Generator;
The pre-trained model is loaded to an object named model, the user reads images one by one from the input catalog and takes the images as the input of the model, the images are identified through a convolutional neural network, the identification range is from 0 to 1, the identification result is rounded to obtain an identification label, ir is represented by 0, I df is represented by 1, and finally the identification result is printed to the user.
3. The compression method of the intelligent recognition algorithm for typical violation behaviors based on the convolutional neural network according to claim 1, which is characterized by comprising the following steps of: the loss function used in the training process is binary cross entropy, the optimizer used is Adam, and the learning rate is 0.01.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410026886.0A CN117831131B (en) | 2024-01-09 | 2024-01-09 | Compression method of typical violation intelligent recognition algorithm based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410026886.0A CN117831131B (en) | 2024-01-09 | 2024-01-09 | Compression method of typical violation intelligent recognition algorithm based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117831131A CN117831131A (en) | 2024-04-05 |
| CN117831131B true CN117831131B (en) | 2024-06-11 |
Family
ID=90520850
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410026886.0A Active CN117831131B (en) | 2024-01-09 | 2024-01-09 | Compression method of typical violation intelligent recognition algorithm based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117831131B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
| CN112183317A (en) * | 2020-09-27 | 2021-01-05 | 武汉大学 | Live working field violation behavior detection method based on space-time diagram convolutional neural network |
-
2024
- 2024-01-09 CN CN202410026886.0A patent/CN117831131B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
| CN112183317A (en) * | 2020-09-27 | 2021-01-05 | 武汉大学 | Live working field violation behavior detection method based on space-time diagram convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117831131A (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107808139B (en) | Real-time monitoring threat analysis method and system based on deep learning | |
| CN116168352B (en) | Power grid obstacle recognition processing method and system based on image processing | |
| CN113076215B (en) | Unsupervised anomaly detection method independent of data types | |
| CN102915453B (en) | Real-time feedback and update vehicle detection method | |
| CN111160424B (en) | NFC equipment fingerprint authentication method and system based on CNN image identification | |
| CN112580657B (en) | Self-learning character recognition method | |
| CN113436184B (en) | Power equipment image defect discriminating method and system based on improved twin network | |
| CN110390784A (en) | A kind of transmission line of electricity external force damage prevention monitoring system based on deep learning | |
| CN114915575B (en) | Network flow detection device based on artificial intelligence | |
| CN116150747A (en) | Intrusion detection method and device based on CNN and SLTM | |
| CN117831131B (en) | Compression method of typical violation intelligent recognition algorithm based on convolutional neural network | |
| CN113687610A (en) | Method for protecting terminal information of GAN-CNN power monitoring system | |
| CN116205905B (en) | Power distribution network construction safety and quality image detection method and system based on mobile terminal | |
| CN115879030A (en) | Network attack classification method and system for power distribution network | |
| CN112529513A (en) | Intelligent seal checking method and system | |
| CN110598650A (en) | License plate false alarm filtering method and device, electronic equipment and storage medium | |
| CN113673326A (en) | Unmanned aerial vehicle platform crowd counting method and system based on image deep learning | |
| CN111126405B (en) | Real-time scanning and updating pushing method for monitoring fault alarm information of converter station | |
| CN113114664A (en) | Abnormal flow detection system and method based on hybrid convolutional neural network | |
| CN113905405A (en) | Abnormal flow detection method for wireless power access private network | |
| CN112036472A (en) | Visual image classification method and system for power system | |
| CN112906811A (en) | Automatic classification method for images of engineering vehicle-mounted equipment based on Internet of things architecture | |
| CN112926368A (en) | Method and device for identifying obstacle | |
| CN110610152A (en) | Multispectral cloud detection method based on discriminative feature learning unsupervised network | |
| CN113221612A (en) | Visual intelligent pedestrian monitoring system and method based on Internet of things |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |