CN112270207A - Smoke and fire detection method in community monitoring scene - Google Patents
Smoke and fire detection method in community monitoring scene Download PDFInfo
- Publication number
- CN112270207A CN112270207A CN202011030101.5A CN202011030101A CN112270207A CN 112270207 A CN112270207 A CN 112270207A CN 202011030101 A CN202011030101 A CN 202011030101A CN 112270207 A CN112270207 A CN 112270207A
- Authority
- CN
- China
- Prior art keywords
- smoke
- fire
- firework
- image
- detection
- 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.)
- Withdrawn
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 89
- 239000000779 smoke Substances 0.000 title claims abstract description 84
- 238000012544 monitoring process Methods 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 40
- 230000002087 whitening effect Effects 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 238000002372 labelling Methods 0.000 claims abstract description 11
- 238000004519 manufacturing process Methods 0.000 claims abstract description 6
- 238000000034 method Methods 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims 6
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000013527 convolutional neural network Methods 0.000 description 13
- 238000005516 engineering process Methods 0.000 description 11
- 238000000605 extraction Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/955—Hardware or software architectures specially adapted for image or video understanding using specific electronic processors
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Fire-Detection Mechanisms (AREA)
Abstract
The invention relates to the technical field of image processing and deep learning, and particularly discloses a smoke and fire detection method in a community monitoring scene, which comprises the following steps: acquiring firework video streams in a community monitoring scene, and decoding to acquire firework images; inputting part of the firework images into a generation countermeasure network GAN to generate a large number of firework images in the community environment; labeling smoke and fire images, adding labels to the smoke and fire images to manufacture a data set, and providing a data basis for supervision training of a subsequent detection model; PCA whitening preprocessing is carried out on the data set, so that redundant information of the image is reduced, and the correlation strength among image pixels is reduced; inputting the processed data set into a smoke and fire detection network model SFD-CNN for training and learning until a final parameter is obtained; and detecting whether the graph to be recognized has smoke and fire signs and the like by using the trained SFD-CNN model. The invention can effectively improve the detection rate and efficiency of smoke and fire in the community environment.
Description
Technical Field
The invention relates to the technical field of image processing and deep learning, in particular to a smoke and fire detection method in a community monitoring scene.
Background
With the rapid development of scientific technology and information technology, the wide application prospect of the intelligent video monitoring technology in the fields of civilian use, business, national defense and the like arouses the attention of many domestic and foreign experts, and a large number of research and development personnel and resource conditions are invested in a dispute to promote the application development of the intelligent video monitoring technology. In addition, the development of scientific technology drives social progress, smart communities and smart cities are no longer conceptual topics, and meanwhile, a powerful research trial platform is provided for the research of video monitoring technology. The development of deep learning in recent years has further led to the innovation of video monitoring technology. The convolutional neural network is a typical representation of deep learning and has remarkable effects in the aspects of image processing, image recognition and the like. For feature extraction, the convolutional neural network has the advantage of automatic learning, replaces a complicated traditional manual extraction method, reduces manual intervention and improves the accuracy of feature extraction. The convolution nerve not only has remarkable effect on the aspect of feature extraction, but also plays an irreplaceable role in other aspects such as image recognition and the like.
Target detection and identification technology based on convolutional neural network is widely applied at present. The Faster R-CNN is one of the most effective methods in the field of target identification, and has the advantages that the lifting part of the candidate frame is put on a GPU for operation, the extraction part of the area candidate frame is embedded into the network from the network, and the feature map after convolution can be used for obtaining the area candidate frame. Similar target recognition networks also include Mask-RCNN, YOLO, SSD and the like, although the accuracy of many technologies in experimental effect reaches more than 96%, the applicability of the network depends heavily on the detection environment, and target detection is influenced by objective reasons such as weather and illumination, and external factors such as uncertainty of the target and the like, so that the accuracy of many detection models is reduced. The algorithms cannot meet the requirements of various complex environments, and target detection and identification still have no complete system, so how to design an algorithm for improving the accuracy of target detection and identification aiming at a specific environment is still the focus of current research.
The community is a place where people live together, and smoke and fire are used as objects for target detection and identification in a community monitoring scene, so that the community has great research significance and application value. The smoke and fire detection has strict requirements on detection precision and efficiency, the background of a community environment is complex, the difficulty of target detection and identification is increased, and the smoke and fire detection precision and efficiency based on the convolutional neural network are key problems to be solved at present.
Disclosure of Invention
The invention provides a smoke and fire detection method in a community monitoring scene, aiming at solving the problems of low smoke and fire detection accuracy and efficiency in a community environment. The method designs a smoke and Fire Detection model SFD-CNN (smoke and Fire Detection-restriction neutral Network) based on a convolutional neural Network, collects smoke and Fire images in a community, generates a large number of smoke and Fire image sets under the community environment by using a generation countermeasure Network GAN, and performs PCA whitening preprocessing operation on the data sets, so that the applicability and the practicability of the Detection model are improved, and the Detection precision of the model is further improved by preprocessing. The GPU service condition in the GPU processor is monitored in real time and scheduling is carried out according to the scheduling strategy, so that the model running speed is increased, and the model detection efficiency is further improved.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a smoke and fire detection method under a community monitoring scene comprises the steps of collecting smoke and fire images under a community environment, marking smoke and fire areas, manufacturing training data sets, preprocessing the images, training a smoke and fire detection network model, detecting images to be identified, transmitting results to a terminal and scheduling GPU resources, and further comprises the following steps:
step 1: acquiring firework video streams in a community monitoring scene, and decoding to acquire firework images;
step 2: partial firework images are input into the generation countermeasure network GAN, and a large number of firework images in the community environment are generated. Labeling smoke and fire images, adding labels to the smoke and fire images to manufacture a data set, and providing a data basis for supervision training of a subsequent detection model;
and step 3: whitening preprocessing is carried out on the data set, redundant information of the image is reduced, and the correlation strength between image pixels is reduced;
and 4, step 4: inputting the processed data set into a smoke and fire detection network model SFD-CNN designed by the invention for training and learning until a final parameter is obtained;
and 5: and detecting whether the graph to be recognized has smoke and fire signs or not by using the trained SFD-CNN model. And transmitting the result to a security terminal for further processing.
Step 6: adopting a GPU scheduling strategy to perform GPU scheduling;
and 7: an infrared smoke and fire detector is arranged to detect smoke and fire signs.
Preferably, the step 1 includes: high-definition cameras or video acquisition devices are installed at various places of a community, an area needing to be monitored is selected, all firework video streams in the area are obtained, and a firework image set is obtained by decoding.
Preferably, the step 2 includes: a smoke and fire detection network model belongs to supervised training, a smoke and fire image set in a community environment is generated by using a generation countermeasure network GAN according to rare smoke and fire images collected in the step 1, and is subjected to region labeling, a labeling tool uses LabelImg, smog and fire labels are added, and a data basis is provided for the supervised training of a subsequent detection model. Wherein LabelImg is a visual image calibration tool. Wherein the generating of the countermeasure network GAN specifically includes: the generator is trained using the acquired partial smoke and fire images, producing more and more simulated smoke and fire images. And training the discriminator with the image as input and predicting whether the image is from a training set or created by a generator network until the discriminator no longer distinguishes the image produced by the generator. At this time, a large number of smoke and fire images are generated by the generator for training and learning of the smoke and fire detection model.
Preferably, step 3 includes: in order to further improve the detection accuracy of the detection model, whitening preprocessing operation is carried out on the input image, the redundancy of the input image is reduced, and the strong correlation among image pixels is weakened. And whitening is divided into PCA whitening and ZCA whitening, and PCA whitening is adopted to carry out preprocessing operation on the image according to the characteristics and the detection requirement of the image. PCA whitening is divided into two steps: and pca performs data dimension reduction, whitening and data variance processing. The method comprises the following specific operations: and performing projection operation on the high latitude characteristic vector of the image through the covariance matrix to obtain a new low-dimensional vector. And then, carrying out standard deviation normalization processing on the feature vectors of each dimension.
Preferably, the step 4 includes: in order to solve the problems that the monitoring environment background in the community is complex and other models are not applicable, the method is combined with the characteristic of strong learning capacity of a convolutional neural Network, and a smoke and Fire Detection Network model SFD-CNN (Smog and Fire Detection-restriction neural Network) based on the convolutional neural Network is designed, wherein the Network structure comprises the following components in percentage by weight: the 1 st and 2 nd layers all contain conv, pool and norm, the 3 rd and 4 th layers use the same conv, then are connected with conv5 and pool5, the 6 th layer is fc, and the results are subjected to classification processing by softmax. Training process: firstly, training by using default parameters, and continuously adjusting the initial weight, the training rate and the iteration times according to a training intermediate result until the network achieves the optimal detection effect with preset efficiency.
Preferably, the step 5 includes: and detecting whether the graph to be recognized has smoke and fire signs or not by using the trained SFD-CNN model. And transmitting the result to a security terminal for further processing.
Preferably, the step 6 includes: and monitoring the GPU use condition in the GPU processing cluster in real time, and adopting a proper scheduling strategy to schedule the GPU in real time.
Preferably, the step 7 includes: generally, fireworks radiate in an infrared mode, are rarely perceived in a visible range, and the detection of fireworks can be reliably guaranteed by installing an infrared firework detector in a community environment, so that the requirement of community all-day detection is met.
By adopting the technical scheme, the smoke and fire detection method under the community monitoring scene has the following beneficial effects:
(1) the problem of low smoke and fire data in a community environment is solved by using the generation countermeasure network GAN.
(2) The PCA technology is adopted to reduce the dimension of the image data, and the whitening technology is used for reducing the correlation strength among the pixels of the input image, so that the detection precision of the model is improved;
(3) in the design of the SFD-CNN network model structure, the operation of pooling firstly and then normalizing is adopted, so that the normalized calculated amount is reduced, and the learning speed of the model is improved;
(4) the infrared smoke and fire detector is arranged in the community environment to meet the requirement of detection all day long.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a general flowchart of a smoke and fire detection method under a community monitoring scenario in the present invention;
FIG. 2 is a structural diagram of a smoke and fire detection model SFD-CNN based on a convolutional neural network in the invention;
FIG. 3 is a schematic diagram of the present invention relating to generation of a countermeasure network GAN;
FIG. 4 is a diagram of a GPU resource scheduling strategy in a GPU processor cluster according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, 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 invention.
As shown in FIG. 1, the smoke and fire detection method under the community monitoring scene comprises the following basic steps: acquiring smoke and fire images in a community environment, marking smoke and fire areas to manufacture training data sets, preprocessing the images, training smoke and fire detection network models, detecting the images to be recognized, transmitting results to a terminal and scheduling GPU resources.
The invention is implemented as follows:
in the specific implementation, high-definition cameras or video acquisition devices are installed at various places of a community, an area needing to be monitored is selected, all firework video streams in the area are obtained, and a firework image set is obtained by decoding. The firework detection network model belongs to supervised training, so that firework region labeling is carried out on the collected firework image set by using a LabelImg labeling tool, smog labels and fire labels are added, and a data basis is provided for the supervised training of a subsequent detection model.
In order to further improve the detection accuracy of the detection model, whitening preprocessing operation is carried out on the input image, the redundancy of the input image is reduced, and the strong correlation among image pixels is weakened. And whitening is divided into PCA whitening and ZCA whitening, and PCA whitening is adopted to carry out preprocessing operation on the image according to the characteristics and the detection requirement of the image. PCA whitening is divided into two steps: and pca performs data dimension reduction, whitening and data variance processing. The method comprises the following specific operations: and performing projection operation on the high latitude characteristic vector of the image through the covariance matrix to obtain a new low-dimensional vector. And then, carrying out standard deviation normalization processing on the feature vectors of each dimension.
In order to solve the problems that the background of the monitored environment in the community is complex and other models are not applicable, the method is combined with the characteristic of strong learning capacity of a convolutional neural Network, a smoke and Fire Detection Network model SFD-CNN (smoke and Fire Detection-restriction neural Network) based on the convolutional neural Network is designed, the model is used for detecting whether smoke and Fire signs exist in the monitored environment of the community and transmitting the Detection result to a terminal for further processing. The detection process specifically comprises the following steps: firstly, a smoke and fire data set is obtained, the problem of rare smoke and fire data in a community environment is solved, a generation countermeasure network GAN is introduced, a generator is utilized to generate a large number of smoke and fire image sets in the community environment to manufacture the data set, and then an SFD-CNN network model is trained and learned. In order to further improve the detection precision of the target, a PCA whitening preprocessing method is adopted, so that redundant information of the acquired image is reduced. And inputting the image to be recognized into the trained smoke and fire detection model to obtain a recognition result. The network model is trained by adopting the data set of the community environment, the practicability of the model is improved, and the detection precision of the detection model is improved by the preprocessing operation. And the GPU service condition in the GPU processor is monitored in real time in the experiment and is scheduled according to a scheduling strategy, so that the running speed of the model is increased. In addition, in order to meet the requirement of detection all day long, the smoke and fire detection network model is not applicable at night, an infrared smoke and fire detector is installed in a community to meet the requirement of detection of smoke and fire at night in the community environment, and experiments show that the smoke and fire detection rate and efficiency in the community environment can be effectively improved.
A smoke and fire detection network model belongs to supervised training, a smoke and fire image set in a community environment is generated by using a generation countermeasure network GAN according to rare smoke and fire images collected in the step 1, and is subjected to region labeling, a labeling tool uses LabelImg, smog and fire labels are added, and a data basis is provided for the supervised training of a subsequent detection model. Wherein LabelImg is a visual image calibration tool.
Wherein the generating of the countermeasure network GAN specifically includes: the generator is trained using the acquired partial smoke and fire images, producing more and more simulated smoke and fire images. And training the discriminator with the image as input and predicting whether the image is from a training set or created by a generator network until the discriminator no longer distinguishes the image produced by the generator. At this time, a large number of smoke and fire images are generated by the generator for training and learning of the smoke and fire detection model. The implementation principle is shown in fig. 3.
In specific implementation, a smoke and Fire Detection Network model SFD-CNN (smoke and Fire Detection-restriction neural Network) based on a convolutional neural Network is designed, and a Network structure is shown in fig. 2 and is: the 1 st and 2 nd layers all contain conv, pool and norm, the 3 rd and 4 th layers use the same conv, then are connected with conv5 and pool5, the 6 th layer is fc, and the results are subjected to classification processing by softmax. The specific training process comprises the following steps: firstly, training by using default parameters, and continuously adjusting the initial weight, the training rate and the iteration times according to a training intermediate result until the network achieves the optimal detection effect with preset efficiency.
In specific implementation, a trained detection model is used for detecting the area under the monitoring environment. Whether there is evidence of smoke and fire. And transmitting the result to a security terminal for further processing.
In specific implementation, the GPU resource scheduling layer monitors the current GPU resource usage in real time according to the scheduling policy as shown in fig. 2, before allocating tasks to a cluster of GPU processors, first checks whether the current GPU consumption is too large, and if the current GPU consumption is too large, checks the GPU usage list and the GPU computing power list, and reselects a GPU receiving task.
According to the smoke and fire detection method under the community monitoring scene, the smoke and fire detection model based on the convolutional neural network is designed, the detection method is more suitable for detecting the community environment, preprocessing operation is performed on the image during recognition, and the detection precision is improved. And the operation efficiency is further improved by combining GPU scheduling.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. A firework detection method under a community monitoring scene is characterized in that: the method comprises the following steps:
step 1: acquiring firework video streams in a community monitoring scene, and decoding to acquire firework images;
step 2: inputting part of the firework images into a generation countermeasure network GAN to generate a large number of firework images in the community environment; labeling smoke and fire images, adding labels to the smoke and fire images to manufacture a data set, and providing a data basis for supervision training of a subsequent detection model;
and step 3: PCA whitening preprocessing is carried out on the data set, so that redundant information of the image is reduced, and the correlation strength among image pixels is reduced;
and 4, step 4: inputting the processed data set into a smoke and fire detection network model SFD-CNN for training and learning until a final parameter is obtained;
and 5: detecting the graph to be recognized by using the trained SFD-CNN model to determine whether the graph has smoke and fire signs; and transmitting the result to the terminal for further processing;
step 6: adopting a GPU scheduling strategy to perform GPU scheduling;
and 7: an infrared smoke and fire detector is arranged to detect smoke and fire signs.
2. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in the step 1, high-definition cameras or video acquisition devices are installed at various places of a community, an area needing to be monitored is selected, all firework video streams in the area are obtained, and a firework image set is obtained by decoding.
3. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in the step 2, a smoke and fire detection network model belongs to supervised training, a smoke and fire image set in a community environment is generated by using a generation countermeasure network GAN according to the rare smoke and fire images collected in the step 1, and is subjected to region labeling, a labeling tool uses LabelImg, smog and fire labels are added, and a data basis is provided for the supervised training of a subsequent detection model; wherein LabelImg is a visual image calibration tool; wherein the generating of the countermeasure network GAN specifically includes: training the generator using the collected partial smoke and fire images to generate more and more simulated smoke and fire images; training the discriminator by taking the image as input, and predicting whether the image comes from a training set or is created by a generator network until the discriminator does not distinguish the image generated by the generator; at this time, a large number of smoke and fire images are generated by the generator for training and learning of the smoke and fire detection model.
4. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in the step 3, in order to further improve the detection accuracy of the detection model, a whitening preprocessing operation is performed on the input image, so that the redundancy of the input image is reduced, and the strong correlation among the image pixels is weakened; whitening is divided into PCA whitening and ZCA whitening, and PCA whitening is adopted to carry out preprocessing operation on the image according to the characteristics and the detection requirement of the image; PCA whitening is divided into two steps: PCA (principal component analysis) is used for data dimension reduction, and whitening is used for data variance processing; the method comprises the following specific operations: performing projection operation on the high latitude characteristic vector of the image through a covariance matrix to obtain a new low-dimensional vector; and then, carrying out standard deviation normalization processing on the feature vectors of each dimension.
5. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in the step 4, the network structure of the smoke detection network model SFD-CNN is: the 1 st and 2 nd layers contain conv, pool and norm, the 3 rd and 4 th layers use the same conv, then are connected with conv5 and pool5, the 6 th layer is fc, and the results are subjected to binary classification by using softmax; training process: firstly, training by using default parameters, and continuously adjusting the initial weight, the training rate and the iteration times according to a training intermediate result until the network achieves the optimal detection effect with preset efficiency.
6. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in the step 5, detecting the graph to be recognized by using the trained SFD-CNN model to determine whether the graph has smoke and fire signs; and transmitting the result to a security terminal for further processing.
7. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in step 6, the method further includes monitoring the usage of the GPUs in the GPU processing cluster in real time, and adopting an appropriate scheduling policy to schedule the GPUs in real time.
8. The firework detection method under the community monitoring scene as claimed in claim 1, wherein: in step 7, the firework is radiated in an infrared mode.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011030101.5A CN112270207A (en) | 2020-09-27 | 2020-09-27 | Smoke and fire detection method in community monitoring scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011030101.5A CN112270207A (en) | 2020-09-27 | 2020-09-27 | Smoke and fire detection method in community monitoring scene |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112270207A true CN112270207A (en) | 2021-01-26 |
Family
ID=74349843
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011030101.5A Withdrawn CN112270207A (en) | 2020-09-27 | 2020-09-27 | Smoke and fire detection method in community monitoring scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112270207A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113128422A (en) * | 2021-04-23 | 2021-07-16 | 重庆市海普软件产业有限公司 | Image smoke and fire detection method and system of deep neural network |
CN113361345A (en) * | 2021-05-24 | 2021-09-07 | 上海可深信息科技有限公司 | Intelligent firework identification method |
-
2020
- 2020-09-27 CN CN202011030101.5A patent/CN112270207A/en not_active Withdrawn
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113128422A (en) * | 2021-04-23 | 2021-07-16 | 重庆市海普软件产业有限公司 | Image smoke and fire detection method and system of deep neural network |
CN113128422B (en) * | 2021-04-23 | 2024-03-29 | 重庆市海普软件产业有限公司 | Image smoke and fire detection method and system for deep neural network |
CN113361345A (en) * | 2021-05-24 | 2021-09-07 | 上海可深信息科技有限公司 | Intelligent firework identification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110070033B (en) | Method for detecting wearing state of safety helmet in dangerous working area in power field | |
Jiang et al. | Insulator fault detection in aerial images based on ensemble learning with multi-level perception | |
CN105426820B (en) | More people's anomaly detection methods based on safety monitoring video data | |
US20200097769A1 (en) | Region proposal with tracker feedback | |
CN109359697A (en) | Graph image recognition methods and inspection system used in a kind of power equipment inspection | |
CN113901911B (en) | Image recognition method, image recognition device, model training method, model training device, electronic equipment and storage medium | |
CN112270207A (en) | Smoke and fire detection method in community monitoring scene | |
Chen et al. | Distributed real-time object detection based on edge-cloud collaboration for smart video surveillance applications | |
Wang et al. | Automatic detection of indoor occupancy based on improved YOLOv5 model | |
Yun et al. | Intelligent control of urban lighting system based on video image processing technology | |
CN111126155A (en) | Pedestrian re-identification method for generating confrontation network based on semantic constraint | |
CN112907138B (en) | Power grid scene early warning classification method and system from local to whole perception | |
CN112258552A (en) | Pedestrian multi-target tracking method under community monitoring scene | |
CN112200008A (en) | Face attribute recognition method in community monitoring scene | |
CN116152699B (en) | Real-time moving target detection method for hydropower plant video monitoring system | |
CN112200007A (en) | License plate detection and identification method under community monitoring scene | |
CN114782675B (en) | Dynamic item pricing method and system in safety technical service field | |
CN108549882A (en) | A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning | |
CN112307896A (en) | Method for detecting lewd behavior abnormity of elevator under community monitoring scene | |
CN113822240B (en) | Method and device for extracting abnormal behaviors from power field operation video data | |
Liu et al. | Research on design of intelligent background differential model for training target monitoring | |
Jiang et al. | Deep learning of qinling forest fire anomaly detection based on genetic algorithm optimization | |
Qiao et al. | Design of abnormal behavior detection system in the state grid business office | |
CN115410116B (en) | Multitasking video anomaly detection method, device, equipment and medium | |
Huang et al. | Real-time Video Fire Detection via Convolutional Neural Networks |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210126 |
|
WW01 | Invention patent application withdrawn after publication |