CN113139500B - Smoke detection method, system, medium and equipment - Google Patents
Smoke detection method, system, medium and equipment Download PDFInfo
- Publication number
- CN113139500B CN113139500B CN202110506235.8A CN202110506235A CN113139500B CN 113139500 B CN113139500 B CN 113139500B CN 202110506235 A CN202110506235 A CN 202110506235A CN 113139500 B CN113139500 B CN 113139500B
- Authority
- CN
- China
- Prior art keywords
- smoke
- image
- detected
- key points
- diffusion
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Abstract
The invention belongs to the technical field of visual identification, and particularly provides a smoke detection method. The invention aims to solve the problems of high smoke detection labeling difficulty and large occupied computing resources in the prior art. For this purpose, the smoke detection method of the invention marks small batches of smoke image samples and trains a first detection model, and detects large batches of smoke image samples by applying the first detection model to generate pseudo tags, the first detection model applies the marked small batches of smoke image samples and large batches of smoke image samples with the pseudo tags, and a knowledge distillation algorithm is applied to guide the second detection model to train to obtain the smoke detection model. According to the invention, through semi-supervised learning and knowledge distillation algorithms, the workload of labeling of smoke image samples is reduced, and meanwhile, a teacher network with high accuracy is applied to train student networks, so that the occupation of computing resources can be reduced under the condition of ensuring the accuracy.
Description
Technical Field
The invention belongs to the technical field of visual identification, and particularly relates to a smoke detection method, a system, a medium and equipment.
Background
Fire is one of the disasters easy to happen in the production and life of people, smoke is one of the important early features of fire occurrence, and the occurrence of the fire can be timely predicted by detecting the smoke, so that measures can be timely taken, and the loss of lives and properties of people is reduced. Meanwhile, the smoke is also one of important indexes of environment-friendly detection, and the detection of the smoke can assist in evaluating the air quality.
The traditional smoke detection method is mainly realized based on smoke alarm devices such as smoke sensors and the like. However, such a smoke sensor needs to be able to detect smoke after it has arrived, and in a large space, smoke diffusion may be affected by height, airflow, etc. to make it difficult for smoke to reach the smoke sensor, so that the smoke sensor cannot detect smoke. Meanwhile, the concentration of the smoke in a large space is also an important factor for restricting the operation of the smoke sensor, and when the smoke concentration reaches the trigger of the smoke sensor, the fire is difficult to control, the optimal rescue time is missed, and great loss is caused.
The smoke detection algorithm based on deep learning can solve some problems of traditional smoke detection, and has the remarkable characteristics of wide monitoring range, long monitoring distance, quick response and the like. However, because the shape of the smoke is not fixed, the change is large, the marking is difficult, and the accuracy of the smoke detection algorithm based on deep learning is not high; meanwhile, the method is limited by computational resources, and the instantaneity of the smoke detection algorithm based on deep learning is one of technical problems for restricting the corresponding algorithm to fall to the ground.
Disclosure of Invention
In order to solve the problems in the prior art, namely, the problems that the difficulty in labeling a large number of smoke images is high when a deep learning technology is adopted to detect smoke and the consumption of computing resources is high when corresponding algorithms are executed by using equipment such as a computer and the like, the invention provides a smoke detection method, which comprises the following steps:
training a first type of smoke image sample with tag data to obtain a first detection model, wherein the tag data comprises a smoke area;
performing smoke detection on second-type smoke image samples without label data by adopting the first detection model to obtain smoke areas in each second-type smoke image sample and respectively generating label data of each second-type smoke image sample according to detection results;
constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data;
a knowledge distillation algorithm is adopted, so that the first detection model guides the second detection model to conduct model training by using the smoke image training set, and a smoke detection model is obtained;
and detecting smoke of the image to be detected by adopting the smoke detection model.
In an optional aspect of the above smoke detection method, after the step of performing smoke detection on the image to be detected using the smoke detection model, the method further includes:
respectively acquiring a smoke region detected on each frame of to-be-detected image in the adjacent multi-frame to-be-detected images and corresponding confidence coefficient according to a smoke detection result obtained by carrying out smoke detection on the to-be-detected image by the smoke detection model;
carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value;
and analyzing whether smoke appears in each frame of the image to be detected according to the smoke diffusion value and the confidence coefficient of the smoke area corresponding to each frame of the image to be detected, and outputting a final smoke detection result according to the corresponding smoke area.
In an optional technical scheme of the above smoke detection method, the step of performing smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion score specifically includes:
respectively acquiring image key points of the images to be detected of two adjacent frames, and performing image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
Respectively calculating included angles between line segments formed by two mutually matched image key points in each group of image key points under an image coordinate system of a preset image to be detected and preset coordinate axes in the image coordinate system;
performing angle value distribution analysis according to the respective corresponding included angles of each group of image key points to determine the balance degree of the included angle distribution;
obtaining the smoke diffusion score according to the included angle distribution balance degree;
wherein, the smoke diffusion score and the included angle distribution equilibrium degree form a positive correlation;
or, the step of performing smoke diffusion analysis on the image to be detected in each frame to obtain a smoke diffusion score is specifically as follows:
respectively acquiring image key points of the images to be detected of two adjacent frames, and performing image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
calculating to obtain the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected;
determining the smoke rising degree according to the displacement of the key points corresponding to each group of the image key points;
Obtaining the smoke diffusion value according to the smoke ascending degree;
wherein the smoke spread score is in positive correlation with the smoke rise;
or, the step of performing smoke diffusion analysis on the image to be detected in each frame to obtain a smoke diffusion score is specifically as follows:
determining smoke concentration information of the images to be detected of two adjacent frames;
comparing smoke concentration information of the images to be detected of two adjacent frames;
and determining the smoke diffusion score according to the comparison result of the smoke concentration information.
In an optional technical solution of the above smoke detection method, the step of analyzing whether smoke appears in the image to be detected in each frame and outputting a final smoke detection result according to the corresponding smoke region according to the smoke diffusion score and the confidence of the smoke region corresponding to each image to be detected in each frame specifically includes:
respectively carrying out weighted calculation or multiplication calculation on the confidence coefficient of the smoke region corresponding to each image to be detected and the smoke diffusion value to obtain the smoke score corresponding to each image to be detected;
judging whether the smoke fraction is greater than or equal to a preset threshold value; if yes, judging that smoke appears in the image to be detected corresponding to the smoke fraction.
The invention also provides a smoke detection system comprising:
the first training module is used for training a first detection model by using a first type of smoke image sample with tag data, wherein the tag data comprises a smoke area;
the second training module is used for carrying out smoke detection on second-class smoke image samples without label data by adopting the first detection model so as to acquire a smoke area in each second-class smoke image sample and respectively generating label data of each second-class smoke image sample according to a detection result;
the training set module is used for constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data;
the knowledge distillation module is used for enabling the first detection model to guide the second detection model to conduct model training by using the smoke image training set by adopting a knowledge distillation algorithm, so as to obtain a smoke detection model;
and the smoke detection module is used for detecting smoke of the image to be detected by adopting the smoke detection model.
In an alternative aspect of the above smoke detection system, the system further comprises:
The confidence coefficient acquisition module is used for respectively acquiring a smoke region detected on each frame of to-be-detected image in the adjacent multi-frame to-be-detected images and corresponding confidence coefficient according to a smoke detection result obtained by carrying out smoke detection on the to-be-detected image by the smoke detection model;
the diffusion analysis module is used for carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value;
and the detection result module is used for analyzing whether smoke appears in the image to be detected in each frame according to the smoke diffusion value and the confidence degree of the smoke area corresponding to each image to be detected in each frame and outputting a final smoke detection result according to the corresponding smoke area.
In an optional aspect of the above smoke detection system, the diffusion analysis module includes:
the first key point unit is used for respectively acquiring image key points of the images to be detected of two adjacent frames and carrying out image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
The included angle calculation unit is used for calculating included angles between line segments formed by two mutually matched image key points in each group of image key points under an image coordinate system of a preset image to be detected and preset coordinate axes in the image coordinate system;
the balance degree analysis unit is used for carrying out angle value distribution analysis according to the included angles corresponding to each group of image key points so as to determine the balance degree of the included angle distribution;
the first diffusion score analysis unit is used for obtaining a smoke diffusion score according to the included angle distribution balance degree; wherein, the smoke diffusion score and the included angle distribution equilibrium degree form a positive correlation;
and/or, the diffusion analysis module comprises:
the second key point unit is used for respectively acquiring the image key points of the images to be detected of two adjacent frames and carrying out image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
the displacement calculation unit is used for calculating and obtaining the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected;
The ascending degree analysis unit is used for determining the ascending degree of smoke according to the displacement of the key points corresponding to each group of the image key points;
a second diffusion score analysis unit for obtaining a smoke diffusion score according to the smoke rising degree; wherein the smoke spread score is in positive correlation with the smoke rise;
and/or, the diffusion analysis module comprises:
a concentration acquisition unit for determining smoke concentration information of the images to be detected of two adjacent frames;
the concentration comparison unit is used for comparing the smoke concentration information of the images to be detected of two adjacent frames;
and the third diffusion score analysis unit is used for determining the smoke diffusion score according to the comparison result of the smoke concentration information.
In an optional aspect of the above smoke detection system, the detection result module includes:
the smoke score obtaining unit is used for respectively carrying out weighted calculation or multiplication calculation on the confidence coefficient of the smoke region corresponding to each image to be detected and the smoke diffusion score to obtain the smoke score corresponding to each image to be detected;
a smoke score comparison unit for judging whether the smoke score is greater than or equal to a preset threshold; if yes, judging that smoke appears in the image to be detected corresponding to the smoke fraction.
The invention also provides a computer readable storage medium having stored therein a plurality of program codes adapted to be loaded and run by a processor to perform a smoke detection method as defined in any one of the above-mentioned smoke detection methods.
The invention also provides a computer device comprising a processor and a memory adapted to store a plurality of program codes adapted to be loaded and run by the processor to perform the smoke detection method of any one of the solutions described above.
As can be appreciated by those skilled in the art, in the technical solution for implementing the present invention, the smoke detection method includes: training a first type of smoke image sample with tag data to obtain a first detection model, wherein the tag data comprises a smoke area; performing smoke detection on second-class smoke image samples without tag data by adopting a first detection model to obtain smoke areas in each second-class smoke image sample and respectively generating tag data of each second-class smoke image sample according to detection results; constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data; a knowledge distillation algorithm is adopted, so that the first detection model guides the second detection model to conduct model training by using a smoke image training set, and a smoke detection model is obtained; and detecting smoke of the image to be detected by adopting a smoke detection model.
According to the embodiment of the invention, the method for detecting the smoke can be used for training a large-scale detection model based on deep learning by using the first smoke image sample with the label data, obtaining the first detection model, and labeling more unlabeled data samples (second smoke image samples) by using the first detection model to generate the label data of the second smoke image sample, so that the defect that a large number of smoke images cannot be labeled quickly and accurately in the prior art is overcome. Further, in the smoke detection method according to the embodiment of the present invention, the first detection model may be used as a teacher network, the first smoke image sample, the second smoke image sample and the tag data corresponding thereto are applied, and a knowledge distillation algorithm is used to train a small student network (second detection model) to obtain the smoke detection model. In the process of obtaining the smoke detection model, the workload of labeling the smoke image samples is reduced, the smoke detection model is a small student network obtained by using a knowledge distillation technology, the requirement on calculation resources is lower, and meanwhile, the process of training the second detection model is realized based on a large number of smoke image samples labeled with correct label data, so that the smoke detection model has higher accuracy of smoke detection.
In addition, in the smoke detection method according to the embodiment of the invention, besides the smoke detection model can be used for detecting the smoke of the image to be detected to obtain the smoke area and the confidence coefficient of the smoke area in the image to be detected, the smoke diffusion analysis can be performed on each frame of the image to be detected to obtain the smoke diffusion score, then the smoke score in the corresponding image is obtained based on the smoke diffusion score and the confidence coefficient, and whether the smoke appears in the corresponding image to be detected is judged based on the smoke score. Based on the above embodiments, according to the smoke detection method provided by the embodiment of the invention, the smoke diffusion characteristics are considered, and the smoke diffusion score and the confidence are combined to comprehensively judge whether the smoke appears in the image, so that the accuracy of smoke detection is further improved, the anti-interference capability of smoke detection is enhanced, and the false alarm rate is reduced.
Drawings
The present disclosure will become more readily understood with reference to the accompanying drawings. As will be readily appreciated by those skilled in the art: the drawings are for illustrative purposes only and are not intended to limit the scope of the present invention. Wherein:
FIG. 1 is a flow chart illustrating the main steps of a smoke detection method according to an embodiment of the present invention;
FIG. 2 is a flow chart illustrating overall steps of a smoke detection method according to another embodiment of the present invention;
fig. 3 is a detailed flowchart of step S700 in a smoke detection method according to one embodiment of the present invention;
fig. 4 is a detailed flowchart of step S700 in a smoke detection method according to another embodiment of the present invention;
fig. 5 is a detailed flowchart of step S700 in a smoke detection method according to still another embodiment of the present invention;
fig. 6 is a detailed flowchart of step S800 in a smoke detection method according to one embodiment of the present invention;
FIG. 7 is a block diagram of the main structure of a smoke detection system according to one embodiment of the invention;
FIG. 8 is an exemplary diagram of smoke diffusion analysis according to one embodiment of the invention;
fig. 9 is an exemplary view of a movement angle distribution of a rigid object according to one embodiment of the present invention.
Detailed Description
Some embodiments of the invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are merely for explaining the technical principles of the present invention, and are not intended to limit the scope of the present invention.
In the description of the present invention, a "module," "processor" may include hardware, software, or a combination of both. A module may comprise hardware circuitry, various suitable sensors, communication ports, memory, or software components, such as program code, or a combination of software and hardware. The processor may be a central processor, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and/or signal processing functions. The processor may be implemented in software, hardware, or a combination of both. Non-transitory computer readable storage media include any suitable medium that can store program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, and the like. The term "a and/or B" means all possible combinations of a and B, such as a alone, B alone or a and B. The term "at least one A or B" or "at least one of A and B" has a meaning similar to "A and/or B" and may include A alone, B alone or A and B. The singular forms "a", "an" and "the" include plural referents.
A smoke detection method according to an embodiment of the invention, which may include the steps of:
step S100, training to obtain a first detection model by using first smoke image samples with label data.
In this embodiment, first, a rectangular frame labeling manner is applied to the obtained smoke image samples to label the smoke area in each image sample, then, scale adjustment is performed on the labeled image areas to adjust the size of the labeled image areas to be required by the first detection model, and then normalization processing is performed to adjust the image areas to be in a format supported by the target detection neural network. The first detection model is trained using the first type of smoke image samples that have been labeled.
In a possible implementation manner, since the number of the first smoke image samples marked is not large, a data enhancement technology can be adopted to change parameters such as position, visual angle and illumination of each image sample so as to obtain more smoke image samples with different parameters, and the first detection model is trained to obtain a model with stronger generalization capability (capability of accurately detecting smoke in images with different conditions such as position, visual angle and illumination), so that the model is better suitable for more application scenes. It should be noted that, the normalization processing and the data enhancement technology are conventional technologies in the image processing technology field, and are not described herein for brevity.
Step S200, carrying out smoke detection on second-class smoke image samples without label data by adopting a first detection model so as to acquire smoke areas in each second-class smoke image sample and respectively generating label data of each second-class smoke image sample according to detection results.
In this embodiment, the first detection model which has been trained may be applied to detect the large-scale second-type smoke image samples, and generate pseudo tags of the second-type smoke image samples. The first trained detection model is applied to carry out smoke detection on the second type of smoke image samples, and label data marking is carried out on the second type of smoke image samples according to the smoke detection result. Because the first detection model is obtained by training according to the first type of smoke image samples marked with correct tag data, the first detection model has higher smoke detection capability, and when the first detection model is used for detecting smoke of the second type of smoke image samples, more accurate smoke detection results can be obtained, and further tag data generated according to the more accurate smoke detection results are also more accurate tag data. In one implementation manner of the embodiment of the invention, in order to ensure the accuracy of the label data of the first type smoke image samples, a manual labeling mode can be adopted to label a small amount of the first type smoke image samples, then a first detection model trained by the first type smoke image samples is used for carrying out smoke detection on a large amount of the second type smoke image samples, label data is generated according to a smoke detection result, namely, the first detection model is used for automatically carrying out quick and accurate label data labeling on the large amount of the second type smoke image samples, so that the efficiency and accuracy of the label labeling work of the image samples are greatly improved.
Further, as can be seen from the foregoing step S100, before the first detection model is acquired, operations such as resizing and normalization may be performed on the image area in the tag data to form new tag data, and then the first detection model may be trained using the new tag data. In this regard, in one embodiment of step S200, after generating the label data of each second type of smoke image sample, the operations such as resizing and normalization may also be performed on the image area in the label data of the second type of smoke image sample to form new label data in a similar manner to the previous step S100, that is, the image areas in the label data of the first type of smoke image sample and the second type of smoke image sample are provided with the same format through the above operations, so that the subsequent step S300 can use these label data to better train the second detection model at the same time.
Step S300, constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data.
In this embodiment, a new smoke image training set may be generated from the first type of smoke image sample and the second type of smoke image sample that have been labeled with the label data.
And step S400, adopting a knowledge distillation algorithm to enable the first detection model to guide the second detection model to conduct model training by using the smoke image training set, so as to obtain a smoke detection model.
The knowledge distillation algorithm (Knowledge distillation algorithm) is a conventional machine learning algorithm in the technical field of machine learning, and the algorithm is used for guiding training of a student model by a teacher model by constructing a teacher model-student model framework, distilling out the 'knowledge' about characteristic representation learned by the teacher model with complex model structure and large parameter quantity, and transferring the 'knowledge' into the student model with simple model structure, small parameter quantity and weak learning capacity.
In this embodiment, the above-mentioned smoke image training set may be applied, where a first detection model with a higher accuracy is used as a teacher network, and a second detection model with a light weight is used as a student network, and a knowledge distillation algorithm is used to enable the first detection model to guide the second detection model to perform model training by using the above-mentioned smoke image training set, where the second detection model after model training is the above-mentioned smoke detection model to be obtained.
It should be noted that, in the embodiment of the present invention, a first detection model and a second detection model may be constructed by using a model structure that is conventional in the machine learning technical field, where the first detection model and the second detection model may be a detection model constructed based on YOLO, SSD and various derivative models, and a person skilled in the art may flexibly select a model structure to be used according to actual needs to construct the first detection model and the second detection model. In one possible implementation, the first detection model may be a detection model constructed based on a YOLOV5s network, and the second detection model may be a detection model constructed based on a YOLOV4-tiny network.
And S500, detecting smoke of the image to be detected by adopting a smoke detection model.
In this embodiment, the smoke detection model obtained in step S100-step S400 may be applied to detect the image to be detected, and determine whether there is smoke on the image to be detected.
Based on the embodiment of the smoke detection method described in the above steps S100-S500, it is known that the first type smoke image sample with tag data is applied to train a large-scale detection model based on deep learning, a first detection model is obtained, the first detection model is applied to label more unlabeled data samples (the second type smoke image sample), tag data of the second type smoke image sample is generated, and the defect that the tag data cannot be labeled quickly and accurately on a large number of smoke images in the prior art is overcome. Further, in the smoke detection method according to the embodiment of the present invention, the first detection model may be used as a teacher network, the first smoke image sample, the second smoke image sample and the tag data corresponding thereto are applied, and a knowledge distillation algorithm is used to train a small student network (second detection model) to obtain the smoke detection model. In the process of obtaining the smoke detection model, the workload of labeling the smoke image samples is reduced, the smoke detection model is a small student network obtained by using a knowledge distillation algorithm, the requirement on calculation resources is lower, and meanwhile, the process of training the second detection model is realized based on a large number of smoke image samples labeled with correct label data, so that the smoke detection model has higher accuracy of smoke detection.
Further, in another embodiment of the smoke detection method according to the present invention as shown in fig. 2, the smoke detection method may include the following steps S600 to S800 in addition to the steps S100 to S500 described in the embodiment of the smoke detection method shown in fig. 1.
Step S600, according to a smoke detection result obtained by carrying out smoke detection on the images to be detected by the smoke detection model, respectively acquiring a smoke region detected on each frame of images to be detected in the adjacent multi-frame images to be detected and corresponding confidence coefficient. The smoke detection model is obtained through the steps S100-S500.
The smoke region refers to a region where smoke appears in an image to be detected.
The confidence of the smoke area refers to the confidence level of the detected smoke occurrence area in the image to be detected, namely the probability that the detected area is a real smoke area.
In this embodiment, a smoke detection result, that is, a position and a smoke confidence of smoke in an image to be detected, may be obtained according to the smoke detection model in step S500.
And step S700, performing smoke diffusion analysis on each frame of image to be detected to obtain a smoke diffusion value.
As shown in fig. 8 to 9, in this embodiment, the image to be detected may be subjected to smoke diffusion analysis according to the diffusion characteristics of the smoke to obtain a smoke diffusion score. Wherein fig. 8 is an exemplary graph of smoke diffusion analysis according to one embodiment of the present invention, and fig. 9 is an exemplary graph of motion angle distribution of a rigid object. The smoke diffusion score represents the smoke diffusion degree of two adjacent frames of images, and the obtained score. The smoke diffusion degree can be represented by parameters such as smoke diffusion angle, smoke rising displacement, smoke concentration variation and the like. The smoke spread score and the smoke spread degree are in positive correlation, the smoke spread degree is higher if the smoke spread score is higher, and the smoke spread degree is lower if the smoke spread score is lower.
Further, as shown in fig. 3, in one possible implementation manner of the present embodiment, step S700 may further include:
step S711, respectively acquiring image key points of the images to be detected of two adjacent frames and performing image key point matching to obtain one or more groups of image key points.
In this embodiment, as shown in fig. 8 to 9, key point information in smoke images of two adjacent frames may be obtained, and a matching relationship between key points of images to be detected of two adjacent frames may be calculated. Wherein the key points may refer to feature points of the image to be detected.
In one possible implementation, the OpenCV model is applied to extract the keypoints of the two adjacent frames of images, and the keypoints of the two adjacent frames of images are matched, i.e. the corresponding keypoints of the two adjacent frames are connected. The keypoint extraction algorithm includes, but is not limited to, ORB and SIFT algorithms, and the OpenCV model is a conventional keypoint detection model, and will not be described herein.
Step S712, calculating the included angles between the line segments formed by the two mutually matched image key points in each group of image key points under the image coordinate system of the preset image to be detected and the preset coordinate axes in the image coordinate system.
In this embodiment, an included angle between a line segment formed by a matching relationship between key points of two adjacent frames of images to be detected and a coordinate axis in an image coordinate system may be calculated.
In one possible implementation, the included angle between the key point matched with two adjacent frames and the X axis is calculated and used as the movement angle of the key point.
In step S713, an angle value distribution analysis is performed according to the angles corresponding to each group of image key points, so as to determine the degree of angle distribution equilibrium.
In this embodiment, the degree of balance of the distribution of the clip angles may be analyzed according to the included angles corresponding to the matching key points of each group in each group of images.
Step S714, obtaining the smoke diffusion value according to the included angle distribution balance degree.
In this embodiment, the spreading score of the smoke, that is, the spreading condition of the smoke between two frames, may be determined according to the degree of the angle balance corresponding to each group of key points.
In one possible implementation, the smoke diffusion score calculation method includes that a minimum included angle and a maximum included angle are obtained first, and an angle interval formed by the minimum included angle and the maximum included angle is divided into a plurality of angle subintervals in equal proportion. And then counting the included angle number of the included angles in each angle subinterval respectively and obtaining the maximum included angle number Xmax. And finally, calculating a diffusion score according to a calculation formula of the diffusion score shown in the following formula:
diffusion score = 1-Xmax/Total. Wherein Total is the Total number of included angles.
Because of the diffusion characteristic of the smoke, the movement angle (included angle) distribution of the key points of the smoke is relatively balanced, and the movement angle distribution of the key points of the rigid object (the included angle between the line segment formed by processing the two adjacent frames of images of the rigid object by adopting the method similar to the steps S711-S714 and the preset coordinate axis in the image coordinate system) is relatively concentrated. So when the movement angle (included angle) distribution of the key points is more uniform, the diffusion score of the corresponding smoke image is higher, whereas when the variation angle distribution of the key points is more concentrated, the diffusion score of the corresponding smoke image is lower. In other words, in the embodiment of the invention, the smoke diffusion value and the included angle distribution balance degree are in positive correlation.
Further, as shown in fig. 4, in another possible implementation manner of this embodiment, step S700 may further include:
step S721, respectively obtaining image key points of images to be detected of two adjacent frames and performing image key point matching to obtain one or more groups of image key points;
step S722, calculating to obtain the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected;
step S723, determining the smoke rising degree according to the displacement of the key points corresponding to each group of image key points;
and step S724, obtaining a smoke diffusion value according to the smoke ascending degree.
Specifically, similar to the distribution of angle values, image key points in images of two adjacent frames are obtained and matched, the displacement of each group of matching key points in the vertical direction is further calculated, and the rising degree of smoke is determined according to the displacement of each group of matching key points in the vertical direction. Generally, according to the smoke diffusion characteristic, the motion process has a rising motion component, so when the displacement of the matching key point in the vertical direction is positive and is greater than a set threshold value, the object in the image can be determined to be in a rising state. And calculating the proportion of the key points in the ascending state of the matching key points in each group of images as smoke diffusion scores according to the ascending degree of the smoke.
Further, as shown in fig. 5, in still another possible implementation manner of the present embodiment, step S700 may further include:
step S731, determining smoke concentration information of the images to be detected of two adjacent frames;
step S732, comparing smoke concentration information of the images to be detected of two adjacent frames;
and step S733, determining a smoke diffusion score according to the comparison result of the smoke concentration information.
Specifically, according to the diffusion characteristics of the smoke, the image color in the image to be detected can be identified to determine the smoke concentration information, and the smoke diffusion value is determined according to the smoke concentration of the two adjacent frames.
In one possible embodiment, the smoke concentration is identified by applying the smoke concentration quantification method disclosed in the patent application CN201911127988.7, and smoke concentration information is obtained, and the concentration of the smoke gradually decreases in the diffusion process according to the diffusion characteristic of the smoke. In two adjacent frames of images, the smoke concentration of the previous frame of smoke image is X1, the smoke concentration of the next frame of smoke image is X2, and the calculation formula of the smoke diffusion value is as follows:
diffusion score = 1-X 2 /X 1 。
Step S800, analyzing whether smoke appears in each frame of image to be detected according to the smoke diffusion value and the confidence coefficient of the smoke area corresponding to each frame of image to be detected, and outputting a final smoke detection result according to the corresponding smoke area.
In this embodiment, by comprehensively determining whether smoke appears in the image to be detected by combining the smoke diffusion value and the confidence of the smoke, a more accurate smoke detection result can be obtained.
In one implementation of the present embodiment, as shown in fig. 5, step S800 may further include:
step S810, respectively carrying out weighted calculation or multiplication calculation on the confidence coefficient of the smoke region corresponding to each frame of image to be detected and the smoke diffusion value to obtain the smoke score corresponding to each frame of image to be detected;
step S820, judging whether the smoke fraction is larger than or equal to a preset threshold value; if yes, judging that smoke appears in the image to be detected corresponding to the smoke fraction.
In this embodiment, the smoke confidence and the smoke diffusion score may be fused to obtain a smoke score corresponding to each smoke region. And setting a corresponding threshold value, comparing the smoke score with the threshold value, and judging that smoke appears in the image to be detected when the smoke score is larger than the threshold value.
When no smoke exists in the image to be detected, outputting a smoke-free result; when the smoke exists in the image to be detected, the result of the smoke and the region where the smoke exists are adjusted to be output in a fixed format. The fixed format may be selected according to the actual application scenario.
In one possible implementation, the smoke confidence and the smoke diffusion score are weighted to obtain the corresponding smoke score, wherein the weights of the smoke confidence and the diffusion score can be set according to the needs in the practical application process. In another possible embodiment, the smoke confidence and the smoke spread score are multiplied to obtain the corresponding smoke score.
As can be seen from the implementation of the smoke detection method in the foregoing steps S100 to S800, the smoke detection method according to the embodiment of the present invention can label a small number of smoke image samples, train a first detection model, label a large-scale unlabeled smoke image sample by using the first detection model to generate a corresponding label, further train a second detection model by using the first detection model to generate a smoke detection model by using the small number of labeled smoke image samples, the large-scale smoke image sample and the corresponding label data. In the generation process of the smoke detection model, the workload of image sample labeling is reduced, meanwhile, a large model (a first detection model) with higher accuracy is used by applying a knowledge distillation algorithm, a small student network (a second detection model) is trained based on a large amount of data samples, and the smoke detection model is obtained, so that the smoke detection model can realize good trade-off of accuracy and real-time. In the smoke detection process, diffusion motion analysis is integrated, the false alarm rate is reduced, the accuracy of smoke detection can be further improved, and a smoke detection model is more real and reliable.
Further, the invention also provides a smoke detection system.
Referring to fig. 7, fig. 7 is a main block diagram of a smoke detection system according to an embodiment of the present invention. As shown in fig. 7, the smoke detection system in this embodiment may include a first training module for training to obtain a first detection model using a first type of smoke image sample with tag data, where the tag data includes a smoke region; the second training module is used for carrying out smoke detection on the second-type smoke image samples without the label data by adopting the first detection model so as to acquire a smoke area in each second-type smoke image sample and respectively generating label data of each second-type smoke image sample according to a detection result; the training set module is used for constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data; the knowledge distillation module is used for enabling the first detection model to guide the second detection model to conduct model training by using the smoke image training set by adopting a knowledge distillation algorithm to obtain a smoke detection model; the smoke detection module is used for detecting smoke of the image to be detected by adopting a smoke detection model; the confidence coefficient acquisition module is used for respectively acquiring a smoke region detected on each frame of to-be-detected image in the adjacent multi-frame to-be-detected images and corresponding confidence coefficient according to a smoke detection result obtained by carrying out smoke detection on the to-be-detected image according to the smoke detection model; the diffusion analysis module is used for carrying out smoke diffusion analysis on each frame of image to be detected to obtain a smoke diffusion value; the detection result module is used for analyzing whether smoke appears in each frame of image to be detected according to the smoke diffusion value and the confidence degree of the smoke area corresponding to each frame of image to be detected, and outputting a final smoke detection result according to the corresponding smoke area.
In one embodiment, the diffusion analysis module may include a first keypoint unit configured to acquire image keypoints of images to be detected of two adjacent frames respectively and perform image keypoint matching to obtain one or more groups of image keypoints; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of two adjacent frames; the included angle calculation unit is used for calculating included angles between line segments formed by two mutually matched image key points in each group of image key points under an image coordinate system of a preset image to be detected and preset coordinate axes in the image coordinate system; the balance degree analysis unit is used for carrying out angle value distribution analysis according to the included angles corresponding to each group of image key points so as to determine the balance degree of the included angle distribution; the first diffusion score analysis unit is used for obtaining a smoke diffusion score according to the included angle distribution equilibrium degree; wherein, the smoke diffusion score and the included angle distribution equilibrium degree form a positive correlation.
In one embodiment, the diffusion analysis module may include a second keypoint unit configured to acquire image keypoints of the images to be detected of two adjacent frames and perform image keypoint matching, to obtain one or more groups of image keypoints; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of two adjacent frames; the displacement calculation unit is used for calculating and obtaining the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected; the ascending degree analysis unit is used for determining the ascending degree of smoke according to the displacement of the key points corresponding to each group of image key points; a second diffusion score analysis unit for obtaining a smoke diffusion score according to the smoke rising degree; wherein the smoke spread score is positively correlated with the degree of smoke rise.
In one embodiment, the diffusion analysis module may include a concentration acquisition unit for determining smoke concentration information of images to be detected of two adjacent frames; the concentration comparison unit is used for comparing the smoke concentration information of the images to be detected of two adjacent frames; and the third diffusion score analysis unit is used for determining a smoke diffusion score according to the comparison result of the smoke concentration information.
In one embodiment, the detection result module includes a smoke score obtaining unit, configured to perform weighted calculation or multiplication calculation on the confidence coefficient and the smoke diffusion score of the smoke area corresponding to each frame of the image to be detected, so as to obtain the smoke score corresponding to each frame of the image to be detected; the smoke score comparison unit is used for judging whether the smoke score is larger than or equal to a preset threshold value; if so, judging that smoke appears in the image to be detected corresponding to the smoke score.
The above-mentioned smoke detection system is used for executing the embodiment of the smoke detection method shown in fig. 1, and the technical principles of both, the technical problems to be solved and the technical effects to be produced are similar, and those skilled in the art can clearly understand that, for convenience and brevity of description, the specific working process and the related description of the smoke detection system can refer to the description of the embodiment of the smoke detection method, and the description is omitted herein.
In particular, the smoke detection method is mainly based on a deep learning target detection algorithm to realize model training and smoke detection, and can also be applied to a semantic segmentation algorithm to realize model training and smoke detection.
The smoke detection method can timely early warn fire occurrence, thereby being beneficial to timely rescue of firefighters. Meanwhile, the smoke detection method can also detect the occurrence of smoke in the environment so as to assist in evaluating the air quality.
It will be appreciated by those skilled in the art that the present invention may implement all or part of the flow of the method of the above-described embodiment, or may be implemented by a computer program for instructing relevant hardware, the computer program being stored in a computer readable storage medium, the computer program realizing the steps of the above-described embodiment of the smoke detection method when being executed by a processor. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device, medium, usb disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier wave signals, telecommunications signals, software distribution media, and the like capable of carrying the computer program code. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
Further, the invention also provides a computer readable storage medium. In one computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium may be configured to store a program that performs the above-described smoke detection method embodiment, the program being loadable and executable by a processor to implement the above-described smoke detection method. For convenience of explanation, only those portions of the embodiments of the present invention that are relevant to the embodiments of the present invention are shown, and specific technical details are not disclosed, please refer to the method portions of the embodiments of the present invention. The computer readable storage medium may be a storage device including various electronic devices, and optionally, the computer readable storage medium in the embodiments of the present invention is a non-transitory computer readable storage medium.
Further, the invention also provides computer equipment. In one embodiment of a computer device according to the present invention, the computer device may comprise a processor and a storage means, the storage means may be configured to store a program for executing the above-described embodiment of the smoke detection method, and the processor may be configured to execute the program in the storage means, including but not limited to the program for executing the above-described embodiment of the smoke detection method. For convenience of explanation, only those portions of the embodiments of the present invention that are relevant to the embodiments of the present invention are shown, and specific technical details are not disclosed, please refer to the method portions of the embodiments of the present invention. The computer device may be a device formed of various electronic devices.
Further, it should be understood that, since the respective modules are merely set to illustrate the functional units of the apparatus of the present invention, the physical devices corresponding to the modules may be the processor itself, or a part of software in the processor, a part of hardware, or a part of a combination of software and hardware. Accordingly, the number of individual modules in the figures is merely illustrative.
Those skilled in the art will appreciate that the various modules in the apparatus may be adaptively split or combined. Such splitting or combining of specific modules does not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solution after splitting or combining falls within the protection scope of the present invention.
Thus far, the technical solution of the present invention has been described in connection with one embodiment shown in the drawings, but it is easily understood by those skilled in the art that the scope of protection of the present invention is not limited to these specific embodiments. Equivalent modifications and substitutions for related technical features may be made by those skilled in the art without departing from the principles of the present invention, and such modifications and substitutions will fall within the scope of the present invention.
Claims (8)
1. A smoke detection method comprising:
training a first type of smoke image sample with tag data to obtain a first detection model, wherein the tag data comprises a smoke area;
performing smoke detection on second-type smoke image samples without label data by adopting the first detection model to obtain smoke areas in each second-type smoke image sample and respectively generating label data of each second-type smoke image sample according to detection results;
constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data;
a knowledge distillation algorithm is adopted, so that the first detection model guides the second detection model to conduct model training by using the smoke image training set, and a smoke detection model is obtained;
adopting the smoke detection model to detect smoke of the image to be detected;
after the step of "smoke detection of an image to be detected using the smoke detection model", the method further comprises:
respectively acquiring a smoke region detected on each frame of to-be-detected image in the adjacent multi-frame to-be-detected images and corresponding confidence coefficient according to a smoke detection result obtained by carrying out smoke detection on the to-be-detected image by the smoke detection model;
Carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value;
analyzing whether smoke appears in each frame of the image to be detected according to the smoke diffusion value and the confidence coefficient of the smoke area corresponding to each frame of the image to be detected, and outputting a final smoke detection result according to the corresponding smoke area;
the step of carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value comprises the following steps:
respectively acquiring image key points of the images to be detected of two adjacent frames, and performing image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
obtaining the smoke diffusion score according to the image key points; or alternatively, the first and second heat exchangers may be,
determining the smoke diffusion value according to the smoke concentration information of the image to be detected;
the step of determining the smoke diffusion score according to the smoke concentration information of the image to be detected comprises the following steps:
determining smoke concentration information of the images to be detected of two adjacent frames;
comparing smoke concentration information of the images to be detected of two adjacent frames;
Determining the smoke diffusion score according to the comparison result of the smoke concentration information;
the smoke concentration information is determined according to the image color in the image to be detected.
2. The smoke detection method according to claim 1, wherein the step of obtaining the smoke diffusion score from the image keypoints is specifically:
respectively calculating included angles between line segments formed by two mutually matched image key points in each group of image key points under an image coordinate system of a preset image to be detected and preset coordinate axes in the image coordinate system;
performing angle value distribution analysis according to the respective corresponding included angles of each group of image key points to determine the balance degree of the included angle distribution;
obtaining the smoke diffusion score according to the included angle distribution balance degree;
wherein, the smoke diffusion score and the included angle distribution equilibrium degree form a positive correlation;
or alternatively, the process may be performed,
the step of obtaining the smoke diffusion score according to the image key points is specifically as follows:
calculating to obtain the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected;
Determining the smoke rising degree according to the displacement of the key points corresponding to each group of the image key points;
obtaining the smoke diffusion value according to the smoke ascending degree;
wherein the smoke spread score is positively correlated with the degree of smoke rise.
3. The smoke detection method according to claim 1 or 2, wherein the step of analyzing whether smoke appears in each frame of the image to be detected and outputting a final smoke detection result according to the corresponding smoke region according to the smoke diffusion score and the confidence level of the smoke region corresponding to each frame of the image to be detected is specifically:
respectively carrying out weighted calculation or multiplication calculation on the confidence coefficient of the smoke region corresponding to each image to be detected and the smoke diffusion value to obtain the smoke score corresponding to each image to be detected;
judging whether the smoke fraction is greater than or equal to a preset threshold value; if yes, judging that smoke appears in the image to be detected corresponding to the smoke fraction.
4. A smoke detection system comprising:
the first training module is used for training a first detection model by using a first type of smoke image sample with tag data, wherein the tag data comprises a smoke area;
The second training module is used for carrying out smoke detection on second-class smoke image samples without label data by adopting the first detection model so as to acquire a smoke area in each second-class smoke image sample and respectively generating label data of each second-class smoke image sample according to a detection result;
the training set module is used for constructing a smoke image training set according to the first smoke image sample, the second smoke image sample and the corresponding label data;
the knowledge distillation module is used for enabling the first detection model to guide the second detection model to conduct model training by using the smoke image training set by adopting a knowledge distillation algorithm, so as to obtain a smoke detection model;
the smoke detection module is used for detecting smoke of the image to be detected by adopting the smoke detection model;
the system further comprises:
the confidence coefficient acquisition module is used for respectively acquiring a smoke region detected on each frame of to-be-detected image in the adjacent multi-frame to-be-detected images and corresponding confidence coefficient according to a smoke detection result obtained by carrying out smoke detection on the to-be-detected image by the smoke detection model;
The diffusion analysis module is used for carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value;
the detection result module is used for analyzing whether smoke appears in each frame of the image to be detected according to the smoke diffusion value and the confidence coefficient of the smoke area corresponding to each frame of the image to be detected and outputting a final smoke detection result according to the corresponding smoke area;
the step of carrying out smoke diffusion analysis on the image to be detected of each frame to obtain a smoke diffusion value comprises the following steps:
respectively acquiring image key points of the images to be detected of two adjacent frames, and performing image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
obtaining the smoke diffusion score according to the image key points; or alternatively, the first and second heat exchangers may be,
determining the smoke diffusion value according to the smoke concentration information of the image to be detected;
the diffusion analysis module comprises:
a concentration acquisition unit for determining smoke concentration information of the images to be detected of two adjacent frames;
The concentration comparison unit is used for comparing the smoke concentration information of the images to be detected of two adjacent frames;
a third diffusion score analysis unit for determining the smoke diffusion score according to the comparison result of the smoke concentration information;
the smoke concentration information is determined according to the image color in the image to be detected.
5. The smoke detection system of claim 4 wherein said diffusion analysis module comprises:
the first key point unit is used for respectively acquiring image key points of the images to be detected of two adjacent frames and carrying out image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
the included angle calculation unit is used for calculating included angles between line segments formed by two mutually matched image key points in each group of image key points under an image coordinate system of a preset image to be detected and preset coordinate axes in the image coordinate system;
the balance degree analysis unit is used for carrying out angle value distribution analysis according to the included angles corresponding to each group of image key points so as to determine the balance degree of the included angle distribution;
The first diffusion score analysis unit is used for obtaining a smoke diffusion score according to the included angle distribution balance degree; wherein, the smoke diffusion score and the included angle distribution equilibrium degree form a positive correlation;
and/or the number of the groups of groups,
the diffusion analysis module comprises:
the second key point unit is used for respectively acquiring the image key points of the images to be detected of two adjacent frames and carrying out image key point matching to acquire one or more groups of image key points; each group of image key points respectively comprises two image key points matched with each other in the images to be detected of the two adjacent frames;
the displacement calculation unit is used for calculating and obtaining the displacement of the key points according to the coordinate values of the two mutually matched image key points in the vertical direction in the image coordinate system of the preset image to be detected;
the ascending degree analysis unit is used for determining the ascending degree of smoke according to the displacement of the key points corresponding to each group of the image key points;
a second diffusion score analysis unit for obtaining a smoke diffusion score according to the smoke rising degree; wherein the smoke spread score is positively correlated with the degree of smoke rise.
6. The smoke detection system of claim 4 or 5 wherein said detection result module comprises:
the smoke score obtaining unit is used for respectively carrying out weighted calculation or multiplication calculation on the confidence coefficient of the smoke region corresponding to each image to be detected and the smoke diffusion score to obtain the smoke score corresponding to each image to be detected;
a smoke score comparison unit for judging whether the smoke score is greater than or equal to a preset threshold; if yes, judging that smoke appears in the image to be detected corresponding to the smoke fraction.
7. A computer readable storage medium having stored therein a plurality of program codes, characterized in that the program codes are adapted to be loaded and executed by a processor to perform the smoke detection method of any one of claims 1-3.
8. A computer device comprising a processor and a memory, the memory being adapted to store a plurality of program codes, characterized in that the program codes are adapted to be loaded and executed by the processor to perform the smoke detection method of any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110506235.8A CN113139500B (en) | 2021-05-10 | 2021-05-10 | Smoke detection method, system, medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110506235.8A CN113139500B (en) | 2021-05-10 | 2021-05-10 | Smoke detection method, system, medium and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113139500A CN113139500A (en) | 2021-07-20 |
| CN113139500B true CN113139500B (en) | 2023-10-20 |
Family
ID=76816712
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110506235.8A Active CN113139500B (en) | 2021-05-10 | 2021-05-10 | Smoke detection method, system, medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113139500B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023019444A1 (en) * | 2021-08-17 | 2023-02-23 | 华为技术有限公司 | Optimization method and apparatus for semantic segmentation model |
| CN113610069B (en) * | 2021-10-11 | 2022-02-08 | 北京文安智能技术股份有限公司 | Knowledge distillation-based target detection model training method |
| 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 |
| CN114882344A (en) * | 2022-05-23 | 2022-08-09 | 海南大学 | Small-sample underwater fish body tracking method based on semi-supervision and attention mechanism |
| CN115099988A (en) * | 2022-06-28 | 2022-09-23 | 腾讯科技（深圳）有限公司 | Model training method, data processing method, device and computer medium |
| CN117237343B (en) * | 2023-11-13 | 2024-01-30 | 安徽大学 | Semi-supervised RGB-D image mirror detection method, storage medium and computer equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101751744A (en) * | 2008-12-10 | 2010-06-23 | 中国科学院自动化研究所 | Detection and early warning method of smoke |
| CN109214392A (en) * | 2017-07-04 | 2019-01-15 | 北京航天长峰科技工业集团有限公司 | A method of Smoke Detection is carried out by video monitoring |
| CN110390674A (en) * | 2019-07-24 | 2019-10-29 | 腾讯医疗健康(深圳)有限公司 | Image processing method, device, storage medium, equipment and system |
| EP3699825A2 (en) * | 2019-02-22 | 2020-08-26 | Ubotica Technologies Ltd. | Systems and methods for deploying and updating neural networks at the edge of a network |
| CN112052744A (en) * | 2020-08-12 | 2020-12-08 | 成都佳华物链云科技有限公司 | Environment detection model training method, environment detection method and device |
| CN112149583A (en) * | 2020-09-27 | 2020-12-29 | 山东产研鲲云人工智能研究院有限公司 | Smoke detection method, terminal device and storage medium |
| CN112257815A (en) * | 2020-12-03 | 2021-01-22 | 北京沃东天骏信息技术有限公司 | Model generation method, target detection method, device, electronic device, and medium |
| CN112766087A (en) * | 2021-01-04 | 2021-05-07 | 武汉大学 | Optical remote sensing image ship detection method based on knowledge distillation |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180268292A1 (en) * | 2017-03-17 | 2018-09-20 | Nec Laboratories America, Inc. | Learning efficient object detection models with knowledge distillation |
| CN109919251A (en) * | 2019-03-21 | 2019-06-21 | 腾讯科技（深圳）有限公司 | A kind of method and device of object detection method based on image, model training |
| US11080558B2 (en) * | 2019-03-21 | 2021-08-03 | International Business Machines Corporation | System and method of incremental learning for object detection |
-
2021
- 2021-05-10 CN CN202110506235.8A patent/CN113139500B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101751744A (en) * | 2008-12-10 | 2010-06-23 | 中国科学院自动化研究所 | Detection and early warning method of smoke |
| CN109214392A (en) * | 2017-07-04 | 2019-01-15 | 北京航天长峰科技工业集团有限公司 | A method of Smoke Detection is carried out by video monitoring |
| EP3699825A2 (en) * | 2019-02-22 | 2020-08-26 | Ubotica Technologies Ltd. | Systems and methods for deploying and updating neural networks at the edge of a network |
| CN110390674A (en) * | 2019-07-24 | 2019-10-29 | 腾讯医疗健康(深圳)有限公司 | Image processing method, device, storage medium, equipment and system |
| CN112052744A (en) * | 2020-08-12 | 2020-12-08 | 成都佳华物链云科技有限公司 | Environment detection model training method, environment detection method and device |
| CN112149583A (en) * | 2020-09-27 | 2020-12-29 | 山东产研鲲云人工智能研究院有限公司 | Smoke detection method, terminal device and storage medium |
| CN112257815A (en) * | 2020-12-03 | 2021-01-22 | 北京沃东天骏信息技术有限公司 | Model generation method, target detection method, device, electronic device, and medium |
| CN112766087A (en) * | 2021-01-04 | 2021-05-07 | 武汉大学 | Optical remote sensing image ship detection method based on knowledge distillation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113139500A (en) | 2021-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113139500B (en) | Smoke detection method, system, medium and equipment | |
| He et al. | Application of deep learning in integrated pest management: A real-time system for detection and diagnosis of oilseed rape pests | |
| CN113705478B (en) | Mangrove single wood target detection method based on improved YOLOv5 | |
| CN111178197B (en) | Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method | |
| CN108447080B (en) | Target tracking method, system and storage medium based on hierarchical data association and convolutional neural network | |
| CN110084165B (en) | Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation | |
| CN109978893A (en) | Training method, device, equipment and the storage medium of image, semantic segmentation network | |
| CN108960047B (en) | Face duplication removing method in video monitoring based on depth secondary tree | |
| CN112183166A (en) | Method and device for determining training sample and electronic equipment | |
| CN116229052B (en) | Method for detecting state change of substation equipment based on twin network | |
| CN115187772A (en) | Training method, device and equipment of target detection network and target detection method, device and equipment | |
| CN111368634B (en) | Human head detection method, system and storage medium based on neural network | |
| CN116492634A (en) | Standing long jump testing method based on image visual positioning | |
| Li et al. | Cross-modal object tracking: Modality-aware representations and a unified benchmark | |
| CN115659221A (en) | Teaching quality assessment method and device and computer readable storage medium | |
| CN115393616A (en) | Target tracking method, device, equipment and storage medium | |
| CN115311601A (en) | Fire detection analysis method based on video analysis technology | |
| CN111191575B (en) | Naked flame detection method and system based on flame jumping modeling | |
| CN114882344A (en) | Small-sample underwater fish body tracking method based on semi-supervision and attention mechanism | |
| CN114998985A (en) | Early warning control method of intelligent experiment table and intelligent experiment table | |
| CN113392927A (en) | Animal target detection method based on single-order deep neural network | |
| CN112329550A (en) | Weak supervision learning-based disaster-stricken building rapid positioning evaluation method and device | |
| Ji et al. | Influence of embedded microprocessor wireless communication and computer vision in Wushu competition referees’ decision support | |
| CN111626193A (en) | Face recognition method, face recognition device and readable storage medium | |
| CN115527083B (en) | Image annotation method and device and electronic equipment |
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 |