CN113449588A - Smoke and fire detection method - Google Patents
Smoke and fire detection method Download PDFInfo
- Publication number
- CN113449588A CN113449588A CN202110525231.4A CN202110525231A CN113449588A CN 113449588 A CN113449588 A CN 113449588A CN 202110525231 A CN202110525231 A CN 202110525231A CN 113449588 A CN113449588 A CN 113449588A
- Authority
- CN
- China
- Prior art keywords
- smoke
- model
- detection method
- firework
- image
- 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
- 239000000779 smoke Substances 0.000 title claims abstract description 81
- 238000001514 detection method Methods 0.000 title claims abstract description 36
- 238000003709 image segmentation Methods 0.000 claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 34
- 238000013145 classification model Methods 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 23
- 230000008859 change Effects 0.000 claims abstract description 6
- 238000002372 labelling Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 2
- 230000002411 adverse Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012795 verification 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Fire-Detection Mechanisms (AREA)
Abstract
The invention provides a smoke and fire detection method, which comprises the steps of firstly establishing a smoke and fire identification model, including image collection and image annotation, training an initial basic image segmentation model by using an annotated image, and iterating according to the requirement; training an image classification model for judging whether the picture contains smoke or flame, and introducing the output of the image segmentation model into the image classification model; obtaining a video stream from a video source, inputting the video stream into a firework identification model, and reserving pictures of a latest complex frame; if the output result of the firework identification model is that no firework exists, the result is directly output; and if the output result of the firework identification model is firework, judging whether the model false alarm occurs according to the gray level change condition of the plurality of recent frames of pictures. The firework identification method has the advantages of high accuracy, low false alarm rate and capability of providing pixel-level precision.
Description
Technical Field
The invention relates to the field of computer vision, in particular to a smoke and fire detection method.
Background
The fire disaster is a disaster which seriously threatens the safety of human life and property, and once the fire disaster happens, the loss is large. Therefore, the prevention and control of fire has been an important issue in human society. The early detection of the fireworks can certainly greatly reduce the loss caused by the fire. With the progress made by humans in artificial intelligence, and in particular in computer vision, it has become possible to identify particular objects in images by computer. However, unlike general objects, the visual characteristics of flames and smoke, such as shape and color, change, which makes recognition of fireworks more difficult than general object recognition.
Patent CN 109165575 a, "a firework identification algorithm based on SSD framework", discloses a firework identification algorithm based on SSD. The detection model used in the smoke and fire recognition algorithm is obtained by training a redesigned model network, the model network reconstructs a classic deep learning model VGG16 network, the model network reduces one full connection layer on the basis of VGG16, retains two full connection layers, and increases 6 convolutional layers and 1 pooling layer. This patent uses an object recognition based network to detect smoke and fire, however due to the dispersivity of the smoke and fire morphology, it is difficult to label smoke and fire with a labeling box in the object recognition mode. If the firework is to be completely framed, a large number of images of irrelevant areas are easily framed, which may adversely affect the training of the model, resulting in a reduction in recognition rate.
Patent CN 111814635 a, "firework identification model building method and firework identification method based on deep learning" discloses a firework identification model building method and firework identification method based on deep learning, in which a generative countermeasure network (gan network) is used to synthesize firework pictures and normal pictures of an area to be detected, and then the firework identification model to be built is obtained through labeling, training and verification by yolov3(You Only Look Once v 3). In the image labeling process, whether the ratio of smoke and fire in the labeling box exceeds 1/2 is compared, if yes, smoke and fire re-labeling is carried out, the smoke and fire parts are divided, and the ratio of smoke and fire in each part is not lower than 1/2 of the labeling box. This patent recognizes the problem of difficulty in framing fireworks with the aforementioned labeling boxes and attempts to frame fireworks itself as much as possible and exclude extraneous image content from the box by breaking up a large box into several smaller boxes. However, this leads to a large number of adjacent labeling boxes in the labeling, which have no distinct boundaries in the image, and this also has an adverse effect on the training of the model.
Neither patent CN 109165575 a nor patent CN 111814635 a jumped out of a nest for firework detection using an object recognition model, and both highly depended on the performance of the object recognition model, and once the detection model was recognized by mistake (which could hardly be completely avoided), false alarm resulted. Therefore, a reliable scheme is needed to solve the problems of the prior art that the recognition rate is not high enough, the false alarm rate is not low enough, the applicable scene is narrow, and the like
Disclosure of Invention
The invention aims to provide a firework identification method based on image segmentation, classification and attention mechanism, which is used for solving the problems of low identification rate, low false alarm rate, narrow applicable scene and the like in the conventional method.
To achieve this object, the invention provides a method of smoke detection comprising the steps of: (1) and establishing a firework identification model. The establishment of the firework identification model comprises image collection, image annotation, training of an initial basic image segmentation model by using the annotated image, and iteration as required. Training an image classification model for judging whether the picture contains smoke or flame, and introducing the output of the image segmentation model into the image classification model; and obtaining video stream input from a video source, sequentially inputting pictures in the video stream into the firework identification model, and reserving the pictures of the latest complex frame. If the output result of the firework identification model is that no firework exists, the result is directly output; and if the output result of the firework identification model is firework, judging whether the model false alarm occurs according to the gray level change condition of the plurality of recent frames of pictures. If the false alarm is found, the alarm is adjusted to be smoke-free fire, otherwise, the alarm is sent out.
The images used in the process of building the smoke and fire identification model comprise scenes containing flame or smoke and scenes not containing flame or smoke; the image forms used include pictures and videos; the image acquisition means comprises downloading from a network and constructing an environment for acquiring; the image annotation mode is pixel-level image segmentation annotation; the image labeling mode is mainly to label a single picture and is assisted by a small number of picture frames intercepted from a video.
In the training method of the initial basic image segmentation model in the process of building the smoke and fire recognition model, dozens of pictures are marked from each collected video, and for the videos of different scenes, the marked pictures are used for carrying out fine-grained adjustment on the initial basic image segmentation model; then, predicting the video by using the obtained new model, and selecting a correct prediction result for marking; making the obtained new label and the original label into a data set for new training, and training a segmentation model with more enhanced performance by using the combined data set; this step can be iterated as necessary.
In the process of building the smoke and fire recognition model, an image classification model is trained and used for judging whether the picture contains smoke or flame, and the output of the image segmentation model is introduced into the image classification model. The introduction of the output of the image segmentation model into the image classification model is introduced by an attention mechanism. And in the process of building the firework recognition model, the image classification model is trained by using a new training data set. The image segmentation model of the smoke and fire recognition model is a panoramic feature pyramid network, a 34-layer residual error neural network, a full convolution neural network, a semantic segmentation network, an L.nkNet, a pyramid scene analysis network, a personal area network or a campus area network.
The video source can be a monitoring camera, an inspection robot and a vehicle-mounted camera; the false model alarm is judged according to the fact that the ratio of the sum of gray changes of all pixels of continuous frames in a smoke region output by the image segmentation model to the arithmetic square root of the number of the pixels is smaller than a threshold value (n-1) × 0.45, and n is the number of frames for calculating the gray difference of the continuous frames. Generally, n is 3, that is, three frames are obtained for the current frame and one frame before and after the current frame, and in this case, the threshold is (3-1) × 0.45 is 0.9. The threshold value may be adjusted experimentally. The larger the threshold, the more stringent the determination of whether a given zone is pyrotechnic. Whether an alarm is given or not finally, the corresponding picture is saved for rapidly knowing the scene situation when the fireworks occur, and the saved picture is re-marked, so that the model performance is further improved.
The image classification model of the smoke and fire recognition model is 34 layers of residual neural network, ResNeXt, ResNeSt or dense connection network.
The firework identification method based on the image segmentation, classification and attention mechanism has the advantages that the problem that fireworks are difficult to label by using an object identification labeling frame due to form dispersion and the problem that the identification rate is not high enough due to limitation of an object identification model are solved.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
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, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for building a smoke and fire identification model according to the present invention.
FIG. 2 is a flow chart of a smoke and fire detection method of the present invention.
Detailed Description
The technical solution 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. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The present invention is described in further detail below to enable those skilled in the art to practice the invention with reference to the description.
Referring to fig. 1, fig. 1 is a flow chart of a firework identification model building method according to the present invention. As shown in fig. 1:
image collection is performed first. The collected image content comprises scenes containing flame or smoke and scenes not containing flame or smoke, the form comprises pictures and videos, and the means comprises downloading from a network and acquiring from the self-constructed environment. The image source can be a monitoring camera, a patrol robot, a vehicle-mounted camera and the like.
And then carrying out image annotation. Unlike the conventional method, the labeling method of the present embodiment is a pixel-level image segmentation labeling. The labeling at this stage is mainly to label a single picture, and is assisted by a small number of picture frames captured from the video.
And then training a basic image segmentation model by using the labeled data. Here, an initial segmentation model is trained, followed by iterative optimization.
For each different scene, fine-grained adjustment (fine-tune) is performed on the initial underlying image segmentation model using the annotated small number of video frames. And then, predicting the video by using a Fine-tune image segmentation model obtained by Fine-tune, and selecting a correct prediction result as a label to manufacture a new training data set. According to experiments, the method can generate a large amount of high-quality labels through a model by a very small amount of manual labels. The method is used for solving the problem that pixel level segmentation labeling is time-consuming and labor-consuming. Since the generalization performance of the fine-tune derived model on the initial segmentation model is limited, different fine-tune models need to be adopted for videos of different scenes.
And then combining a large number of obtained new labels generated based on the video and previous labels into a new data set, and training an advanced image segmentation model with more enhanced performance by using the combined data set. The above process can be iterated as needed.
And training an image classification model by using the combined new data set, judging whether the picture contains smoke or flame, and introducing the output of the advanced image segmentation model into the image classification model through an attention mechanism. And outputting a final detection result by the image classification model.
In this embodiment, the image segmentation model is a panoramic Feature Pyramid Network (hereinafter referred to as FPN), and the image classification model is a 34-layer residual neural Network (hereinafter referred to as ResNet 34).
In some other embodiments, the image segmentation model may use, in addition to FPN, techniques including, but not limited to: FCN, U-Net and its variants, LinkNet, PSPNet, PAN, DAN and other segmentation models; the image classification model may use classification models including, but not limited to, ResNeXt, ResNeSt, DenseNet, etc., in addition to Resnet.
Referring again to FIG. 2, FIG. 2 is a flow chart of a smoke and fire detection method of the present invention. As can be seen from the figure 2 of the drawings,
the invention relates to a smoke and fire detection method, which comprises the following steps:
(1) the process begins by obtaining video stream input from a video source, which may be a surveillance camera, a patrol robot, a vehicle-mounted camera, etc.
(2) Reading video streams
(3) Inputting continuous frames
(4) Model output
(5) Whether fireworks are detected or not
(6) Without detection of smoke or fire, without alarm
(7) And under the condition of smoke and fire detection, storing the picture, calculating the gray change conditions of all pixels of continuous frames in a smoke and fire area, if the ratio of the sum of the gray changes to the arithmetic square root of the number of the pixels is less than a certain threshold, judging that the model is false alarm, adjusting the output to be smoke and fire free, not giving an alarm, and otherwise, giving a smoke and fire alarm. The threshold in this context is typically (n-1) × 0.45, n being the number of frames used to calculate the difference in gray levels for successive frames. Generally, n is 3, that is, three frames are obtained for the current frame and one frame before and after the current frame, and in this case, the threshold is (3-1) × 0.45 is 0.9. The threshold value may be adjusted experimentally. The larger the threshold, the more stringent the determination of whether a given zone is pyrotechnic.
(8) And manually confirming the stored picture and judging whether the picture is misinformed. And if the picture is false, marking the stored picture and retraining the model.
In another embodiment, a video stream input is obtained from a video source, pictures in the video stream are sequentially input into the smoke recognition model, and the pictures of the last several frames, for example 5 frames, are retained. If: (i) if the output result of the firework identification model is that no firework exists, the result is directly output without giving an alarm; (ii) if the output result of the firework identification model is firework, analyzing the latest 5 frames of pictures, calculating the gray change conditions of all pixels of continuous frames in a firework area, if the ratio of the sum of the gray changes to the arithmetic square root of the number of the pixels is less than a certain threshold value, judging that the model is false alarm, adjusting the output to be firework-free, not giving an alarm, otherwise, giving a firework alarm.
Whether an alarm is given or not is finally carried out, the corresponding picture is saved and is used for rapidly knowing the scene situation when fireworks occur. And manually confirming the stored pictures, judging whether the pictures are misinformation or not, and using the pictures for firework identification model training to further improve the model performance.
The firework identification method solves the problems that the identification rate is not high enough, the false alarm rate is not low enough, the applicable scene is narrow and the like in the existing method.
While embodiments of the invention have been disclosed above, it is not intended to be limited to the uses set forth in the specification and examples. It can be applied to all kinds of fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. Therefore, the invention is not to be limited to the specific details and illustrations shown and described herein, without departing from the general concept defined by the claims and their equivalents.
Claims (22)
1. A method of smoke detection comprising the steps of:
(1) establishing a smoke and fire recognition model, wherein the smoke and fire recognition model comprises the steps of collecting images, labeling the images, training an initial basic image segmentation model by using the labeled images, and iterating according to the requirement;
training an image classification model, judging whether the picture contains smoke or flame, and introducing the output of the segmentation model into the classification model;
(2) obtaining a video stream input from a video source;
(3) inputting pictures in a video stream obtained from a video source into the firework identification model, and reserving pictures of a most recent complex frame; if the output result of the firework identification model is that no firework exists, the result is directly output; if the output result of the firework identification model is firework, judging whether the model false alarm occurs according to the gray level change condition of the plurality of recent frames of pictures; if the false alarm is found, the alarm is adjusted to be smoke-free fire, otherwise, the alarm is sent out.
2. The smoke detection method of claim 1, wherein the images used in the process of building the smoke recognition model comprise scenes containing flames or smoke and scenes not containing flames or smoke.
3. The smoke detection method of claim 1, wherein the image format used in the process of building the smoke recognition model comprises pictures and videos.
4. The smoke and fire detection method according to claim 1, wherein the means for acquiring images used in the process of building the smoke and fire identification model comprises downloading from a network and acquiring from a self-constructed environment.
5. The smoke and fire detection method according to claim 1, wherein the image annotation during the smoke and fire identification model building process is a pixel-level image segmentation annotation.
6. The firework detection method according to claim 1, wherein in the firework identification model building process, the image labeling mode is mainly to label a single picture and is assisted by a small number of picture frames captured from a video.
7. The firework detection method according to claim 1, wherein the training method of the initial basic image segmentation model in the firework identification model building process is to label dozens of pictures from each collected video, and for each video of different scenes, fine-grained adjustment is performed on the initial basic image segmentation model by using the labeled pictures; then, predicting the video by using the obtained new model, and selecting a correct prediction result for marking; making the obtained new label and the original label into a new data set for training, and training an image segmentation model with stronger performance by using the combined data set; this step can be iterated as necessary.
8. The smoke detection method of claim 1, wherein said introducing the output of the image segmentation model into the classification model during the smoke recognition model building process is introduced by an attention mechanism.
9. The smoke detection method of claim 1, wherein the image classification model is trained using a combined training dataset during the smoke recognition model building process.
10. The smoke detection method of claim 1, wherein the image segmentation model of the smoke recognition model is a panoramic feature pyramid network.
11. The smoke detection method according to claim 1, wherein the image classification model of the smoke recognition model is a 34-layer residual neural network.
12. The smoke detection method of claim 1, wherein the model false positive is determined based on a ratio of a sum of gray level changes of all pixels of successive frames within a smoke region output by the image segmentation model to an arithmetic square root of a number of pixels being less than a threshold value (n-1) × 0.45, where n is a number of frames used to calculate a gray level difference of successive frames. The number of frames n is generally equal to 3, that is, three frames are obtained in the current frame and in the previous and subsequent frames, and the threshold value is (3-1) × 0.45 equal to 0.9. The threshold value may be adjusted experimentally.
13. The firework detecting method according to claim 1, wherein whether an alarm is finally issued or not, the corresponding picture is saved for rapidly knowing the scene situation when the firework is generated, and the saved picture is re-labeled, so that the model performance is further improved.
14. The smoke detection method of claim 1, wherein the image segmentation model of the smoke recognition model is a full convolution neural network.
15. The smoke detection method according to claim 1, wherein the image segmentation model of the smoke recognition model is a semantic segmentation network.
16. The smoke detection method according to claim 1, wherein the image segmentation model of the smoke recognition model is LinkNet.
17. The smoke detection method of claim 1, wherein the image segmentation model of the smoke recognition model is a pyramid scene analysis network.
18. The smoke detection method of claim 1, wherein the image segmentation model of the smoke recognition model is a personal area network.
19. The smoke detection method of claim 1, wherein the image segmentation model of the smoke recognition model is a campus area network.
20. The smoke detection method according to claim 1, wherein the image classification model of the smoke recognition model is ResNeXt.
21. The smoke detection method according to claim 1, wherein the image classification model of the smoke recognition model is ResNeSt.
22. The smoke detection method according to claim 1, wherein the image classification model of the smoke recognition model is a dense connection network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110525231.4A CN113449588A (en) | 2021-05-13 | 2021-05-13 | Smoke and fire detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110525231.4A CN113449588A (en) | 2021-05-13 | 2021-05-13 | Smoke and fire detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113449588A true CN113449588A (en) | 2021-09-28 |
Family
ID=77809760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110525231.4A Pending CN113449588A (en) | 2021-05-13 | 2021-05-13 | Smoke and fire detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113449588A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114140732A (en) * | 2021-12-09 | 2022-03-04 | 安徽大学 | Fire-fighting fire source detection method and device and storage medium under small sample condition |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018128741A1 (en) * | 2017-01-06 | 2018-07-12 | Board Of Regents, The University Of Texas System | Segmenting generic foreground objects in images and videos |
| CN110827505A (en) * | 2019-10-29 | 2020-02-21 | 天津大学 | Smoke segmentation method based on deep learning |
| CN111445488A (en) * | 2020-04-22 | 2020-07-24 | 南京大学 | Method for automatically identifying and segmenting salt body through weak supervised learning |
| CN111598174A (en) * | 2020-05-19 | 2020-08-28 | 中国科学院空天信息创新研究院 | Training method of image ground feature element classification model, image analysis method and system |
| CN112633231A (en) * | 2020-12-30 | 2021-04-09 | 珠海大横琴科技发展有限公司 | Fire disaster identification method and device |
| CN112633061A (en) * | 2020-11-18 | 2021-04-09 | 淮阴工学院 | Lightweight FIRE-DET flame detection method and system |
| CN112699801A (en) * | 2020-12-30 | 2021-04-23 | 上海船舶电子设备研究所(中国船舶重工集团公司第七二六研究所) | Fire identification method and system based on video image |
-
2021
- 2021-05-13 CN CN202110525231.4A patent/CN113449588A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018128741A1 (en) * | 2017-01-06 | 2018-07-12 | Board Of Regents, The University Of Texas System | Segmenting generic foreground objects in images and videos |
| CN110827505A (en) * | 2019-10-29 | 2020-02-21 | 天津大学 | Smoke segmentation method based on deep learning |
| CN111445488A (en) * | 2020-04-22 | 2020-07-24 | 南京大学 | Method for automatically identifying and segmenting salt body through weak supervised learning |
| CN111598174A (en) * | 2020-05-19 | 2020-08-28 | 中国科学院空天信息创新研究院 | Training method of image ground feature element classification model, image analysis method and system |
| CN112633061A (en) * | 2020-11-18 | 2021-04-09 | 淮阴工学院 | Lightweight FIRE-DET flame detection method and system |
| CN112633231A (en) * | 2020-12-30 | 2021-04-09 | 珠海大横琴科技发展有限公司 | Fire disaster identification method and device |
| CN112699801A (en) * | 2020-12-30 | 2021-04-23 | 上海船舶电子设备研究所(中国船舶重工集团公司第七二六研究所) | Fire identification method and system based on video image |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114140732A (en) * | 2021-12-09 | 2022-03-04 | 安徽大学 | Fire-fighting fire source detection method and device and storage medium under small sample condition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6867153B2 (en) | Abnormality monitoring system | |
| CN109145759B (en) | Vehicle attribute identification method, device, server and storage medium | |
| EP1844444B1 (en) | Identifying spurious regions in a video frame | |
| DK2377044T3 (en) | DETECTING ANORMAL EVENTS USING A LONG TIME MEMORY IN A VIDEO ANALYSIS SYSTEM | |
| CN113158738B (en) | Port environment target detection method, system, terminal and readable storage medium based on attention mechanism | |
| EP2549759B1 (en) | Method and system for facilitating color balance synchronization between a plurality of video cameras as well as method and system for obtaining object tracking between two or more video cameras | |
| CN111881853B (en) | Method and device for identifying abnormal behaviors in oversized bridge and tunnel | |
| CN102348128A (en) | Surveillance camera system having camera malfunction detection function | |
| WO2006082429A1 (en) | Classifying an object in a video frame | |
| CN111738054B (en) | Behavior anomaly detection method based on space-time self-encoder network and space-time CNN | |
| Bedruz et al. | Real-time vehicle detection and tracking using a mean-shift based blob analysis and tracking approach | |
| CN110827505A (en) | Smoke segmentation method based on deep learning | |
| CN110942456B (en) | Tamper image detection method, device, equipment and storage medium | |
| KR102391853B1 (en) | System and Method for Processing Image Informaion | |
| CN111582074A (en) | Monitoring video leaf occlusion detection method based on scene depth information perception | |
| CN112597928A (en) | Event detection method and related device | |
| CN116310922A (en) | Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium | |
| CN113449588A (en) | Smoke and fire detection method | |
| CN114373162A (en) | Dangerous area personnel intrusion detection method and system for transformer substation video monitoring | |
| CN112633179A (en) | Farmer market aisle object occupying channel detection method based on video analysis | |
| CN113095160A (en) | Power system personnel safety behavior identification method and system based on artificial intelligence and 5G | |
| CN117197746A (en) | Safety monitoring system and method based on deep learning | |
| CN116419059A (en) | Automatic monitoring method, device, equipment and medium based on behavior label | |
| CN116543333A (en) | Target recognition method, training method, device, equipment and medium of power system | |
| Bharathi et al. | A custom yolov5-based real-time fire detection system: a deep learning approach |
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 |