CN111145222A - Fire detection method combining smoke movement trend and textural features - Google Patents
Fire detection method combining smoke movement trend and textural features Download PDFInfo
- Publication number
- CN111145222A CN111145222A CN201911405413.7A CN201911405413A CN111145222A CN 111145222 A CN111145222 A CN 111145222A CN 201911405413 A CN201911405413 A CN 201911405413A CN 111145222 A CN111145222 A CN 111145222A
- Authority
- CN
- China
- Prior art keywords
- smoke
- fire
- video
- frame
- network
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Fire-Detection Mechanisms (AREA)
Abstract
The invention discloses a fire detection method combining smoke movement trend and textural features, which comprises the steps of firstly, acquiring a fire smoke video, and constructing a training video set; and then constructing a fire video smoke detection model consisting of a smoke texture feature deep network, a full-connection network and a Softmax layer, and calculating the motion intensity and the overall trend of the fire smoke in the training video set in each direction by a method for extracting HOG features. And then, splicing the characteristic vector output by the smoke texture characteristic deep network and the HOG characteristic, inputting the spliced characteristic vector and the HOG characteristic into a fully-connected network together, and outputting the confidence coefficient for judging the fire smoke probability. And inputting the video to be detected into a fire video smoke detection model frame by frame, and judging whether smoke generated by fire exists in the current frame or not according to whether the confidence coefficient of the output result reaches a set threshold value or not. The method can well distinguish the natural cloud and the fire smoke, accurately identify the fire smoke and improve the accuracy of identifying the fire smoke.
Description
Technical Field
The invention belongs to the field of machine vision application, and particularly relates to a fire detection method combining smoke motion tendency and textural features.
Background
The research of fire smoke detection algorithm has gained great attention all the time, and with the rapid development of deep learning technology, it has become one of the research hotspots therein. The fire smoke detection technology based on deep learning has the advantages of low cost, quick response, strong real-time performance, wide coverage area, high accuracy and the like, and has wide application prospect. Due to the environmental complexity of fire smoke, the variability of light weather, the complexity of algorithm models, computational requirements on hardware, etc., fire smoke detection is currently in the initial landing stage. Sometimes, even if smoke exists, the detection effect is limited by the angle of a camera, object shielding and the like, and the detection effect also has a great space, so that a fire smoke detection algorithm with high precision, low complexity and shielding resistance is yet to be developed.
The current fire smoke detection methods have only two main modes: one method is based on deep learning, but the algorithm complexity is high, the requirement on hardware computing power is high, the deployment is difficult, and the design requirement on a neural network is high. The other method is based on the traditional method, and utilizes manually designed characteristics, such as motion characteristics, optical characteristics, diffusion characteristics and the like of fire smoke to identify the fire smoke, but the method has low accuracy, is slow in identification and easy to fail, and is far less accurate than the former method.
Disclosure of Invention
The invention aims to provide a fire detection method combining smoke movement tendency and textural features aiming at the defects of the prior art. The fire smoke detection is carried out by combining deep learning and motion characteristics, and high-precision real-time fire video smoke detection is realized.
The purpose of the invention is realized by the following technical scheme: a fire detection method combining smoke motion trend and textural features comprises the following steps:
(1) collecting a fire smoke video, extracting video frames with fixed length and frame number in the fire smoke video as training data, marking the fire smoke on the video frames by using rectangular frames, forming a marking file according to a marking result, and taking the marking file and corresponding video frames thereof as a training video set;
(2) constructing a fire video smoke detection model, wherein the model consists of a smoke texture feature depth network, a full-connection network and a Softmax layer, and the smoke texture feature depth network is a network with a LeNet removing the full-connection layer and is used for positioning smoke positions in a video frame by frame; and (3) inputting the training video set in the step (1) into a smoke texture feature depth network, and outputting a feature vector coded by a neural network.
(3) The fire video smoke detection model calculates the motion intensity and the overall trend of fire smoke in the training video set in each direction by a method of extracting HOG characteristics.
The HOG feature extraction specifically comprises the following steps: calculating a moving image of a video frame input with a training video set by a frame difference method, graying the moving image, calculating gradients pixel by pixel after normalization, dividing the input frame into 6 x 6 cells, counting a gradient histogram of each Cell, connecting the gradient histograms in each 3 x 3 cells in series and scanning a full image to obtain an HOG feature, wherein the HOG feature comprises the movement intensity and the overall trend in each direction of fire and smoke;
(4) and splicing the characteristic vector output by the smoke texture characteristic deep network and the HOG characteristic, then inputting the characteristic vector and the HOG characteristic into the full-connection network to judge whether smoke exists in the training video set, and outputting the confidence coefficient of the fire smoke probability in the training video set through a Softmax layer.
(5) And inputting the video to be detected into a fire video smoke detection model frame by frame, judging whether smoke generated by fire exists in the current frame or not by judging whether the confidence coefficient of the output result reaches a set threshold value or not, and warning through a loudspeaker according to the detection result.
Furthermore, the collected fire smoke video is used as a positive sample, the natural cloud and fog video without the fire smoke is collected as a negative sample, a training video set is constructed together to train the fire video smoke detection model, and the anti-interference capability of the fire video smoke detection model is improved.
Further, in the step (5), the smoke detection is performed frame by frame, and an alarm is triggered only when smoke is detected for a plurality of times, and if the number of the continuous frames does not reach the threshold value of the number of triggered alarm frames, the smoke detection is regarded as noise.
The invention has the beneficial effects that: according to the invention, deep learning is combined with a traditional method, a real-time fire smoke detection model is established, and mixed decision is carried out through a post-stage full-connection network by utilizing the accuracy of the deep learning fire smoke on single-frame content and texture detection and the effectiveness of motion characteristic detection. The invention combines the advantages of deep learning and motion characteristics to detect the smoke of the fire simultaneously, and realizes high-precision real-time video smoke detection.
Drawings
FIG. 1 is a flow chart of a fire smoke real-time detection algorithm;
fig. 2 is a schematic diagram of fire smoke labeling.
Detailed Description
The invention is described in further detail below with reference to the figures and specific examples.
According to the invention, a smoke detection model is built and trained, deep learning and a traditional motion detection method are combined to realize a high-precision fire smoke detection method, and the key point is to provide a method for fusing the characteristics of the smoke detection model and the traditional motion detection model and carrying out collaborative judgment. First, a real dataset with fire smoke and natural cloud is generated for model training. Secondly, the smoke motion states obtained by the method based on the lightweight deep neural network LeNet and the HOG feature extraction are fused, and a reasonable integration method is provided. And then, designing a rear-stage full-connection network to carry out overall cooperative detection on the smoke, and designing a network evaluation index to reflect the overall accuracy and real-time performance of the model. As shown in fig. 1, the technical scheme adopted by the invention mainly comprises the following steps:
(1) acquiring a fire smoke video, extracting a video frame with a fixed length frame number in the fire smoke video as a positive sample, acquiring a natural cloud and fog video without fire smoke as a negative sample, constructing a training video set together, improving the anti-interference capability of a fire video smoke detection model, marking the fire smoke on the video frame by using a rectangular frame, forming a marking file according to a marking result, and taking the marking file and a corresponding video frame thereof as the training video set, wherein the number of samples of the training video set is generally more than 1000 as shown in FIG. 2; the video frame data in the training video set is generally subjected to preprocessing operations including cropping, scaling, and the like before subsequent operations.
(2) Constructing a fire video smoke detection model, wherein the model consists of a smoke texture feature depth network, a full-connection network and a Softmax layer, and the smoke texture feature depth network is a network with a LeNet removing the full-connection layer and is used for positioning smoke positions in a video frame by frame; and (3) inputting the training video set in the step (1) into a smoke texture feature depth network, and outputting a feature vector coded by a neural network.
(3) The fire video smoke detection model calculates the motion intensity and the overall trend of fire smoke in the training video set in each direction by a method of extracting HOG characteristics.
And (3) HOG feature extraction: calculating a moving image of a video frame input with a training video set by a frame difference method, graying the moving image, calculating gradients pixel by pixel after normalization, dividing the input frame into 6 cells, counting a gradient histogram of each Cell, connecting the gradient histograms in each 3 cells in series and scanning a full image to obtain an HOG (hot object) feature, wherein the HOG feature comprises the movement intensity and the overall trend in each direction of fire and smoke;
(4) and splicing the characteristic vector output by the smoke texture characteristic deep network and the HOG characteristic, inputting the spliced characteristic vector and the HOG characteristic into a fully-connected network together, judging whether smoke exists in the video through the fully-connected network, and outputting a confidence coefficient for judging the probability of the fire smoke through a Softmax layer. And in the training process, inputting the output confidence coefficient into a back propagation algorithm to update the model parameters.
(5) And inputting the video to be detected into a fire video smoke detection model frame by frame, and judging whether smoke generated by fire exists in the current frame or not according to whether the confidence coefficient of the output result reaches a set threshold value or not. The confidence threshold set varies from data set to data set and is typically greater than 0.7. And determining whether to warn through a loudspeaker according to the detection result, triggering alarm only when smoke is detected for a plurality of times, and considering as noise if the number of continuous frames does not reach the threshold value of the number of triggered alarm frames. According to the statistical accuracy of the model experiment result, the probability of four times of false detections is 0.4%, so the threshold value of the number of triggering alarm frames is set to be 4.
Taking fire video detection in a natural scene in the field as an example, acquiring a field fire video (such as forest fire and the like) and a confusing non-fire video (such as cloud sea, cloud fog and the like) to construct a training video set in the step (1), obtaining a trained model after the processing of the steps (2), (3) and (4), and outputting a detection result through the step (5).
And (5) regarding the fire confidence coefficient result obtained in the step (5), taking the Accuracy (Accuracy) of whether the confidence coefficient is the same as the label and the frame number (FPS) of the video frame to be detected which can be processed by the algorithm per second as the standard of the evaluation algorithm.
The training stage adopts the data of the first half data set, and the testing stage adopts the second half data set as a testing object. The mAP of the experimental result of the model reaches 0.75, the accuracy requirement of fire smoke detection is met, the speed can reach 30FPS, and the real-time requirement is met.
The above-described embodiments are intended to illustrate rather than to limit the invention, and any modifications and variations of the present invention are within the spirit of the invention and the scope of the appended claims.
Claims (3)
1. A fire detection method combining smoke motion trend and textural features is characterized by comprising the following steps:
(1) collecting a fire smoke video, extracting video frames with fixed length and frame number in the fire smoke video as training data, marking the fire smoke on the video frames by using rectangular frames, forming a marking file according to a marking result, and taking the marking file and corresponding video frames thereof as a training video set;
(2) constructing a fire video smoke detection model, wherein the model consists of a smoke texture feature depth network, a full-connection network and a Softmax layer, and the smoke texture feature depth network is a network with a LeNet removing the full-connection layer and is used for positioning smoke positions in a video frame by frame; and (3) inputting the training video set in the step (1) into a smoke texture feature depth network, and outputting a feature vector coded by a neural network.
(3) The fire video smoke detection model calculates the motion intensity and the overall trend of fire smoke in the training video set in each direction by a method of extracting HOG characteristics.
The HOG feature extraction specifically comprises the following steps: calculating a moving image of a video frame input with a training video set by a frame difference method, graying the moving image, calculating gradients pixel by pixel after normalization, dividing the input frame into 6 x 6 cells, counting a gradient histogram of each Cell, connecting the gradient histograms in each 3 x 3 cells in series and scanning a full image to obtain an HOG feature, wherein the HOG feature comprises the movement intensity and the overall trend in each direction of fire and smoke;
(4) and splicing the characteristic vector output by the smoke texture characteristic deep network and the HOG characteristic, then inputting the characteristic vector and the HOG characteristic into the full-connection network to judge whether smoke exists in the training video set, and outputting the confidence coefficient of the fire smoke probability in the training video set through a Softmax layer.
(5) And inputting the video to be detected into a fire video smoke detection model frame by frame, judging whether smoke generated by fire exists in the current frame or not by judging whether the confidence coefficient of the output result reaches a set threshold value or not, and warning through a loudspeaker according to the detection result.
2. The fire detection method combining the smoke motion trend and the textural features according to claim 1, wherein a training video set is constructed to train the fire video smoke detection model by taking a collected fire smoke video as a positive sample and a natural cloud and fog video without fire smoke as a negative sample, so that the anti-interference capability of the fire video smoke detection model is improved.
3. A method as claimed in claim 1, wherein the smoke detection in step (5) is performed on a frame-by-frame basis, and the smoke detection triggers an alarm only when smoke is detected for a number of consecutive times, and the smoke detection is considered noise if the number of consecutive frames does not reach a threshold value for the number of triggered alarm frames.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911405413.7A CN111145222A (en) | 2019-12-30 | 2019-12-30 | Fire detection method combining smoke movement trend and textural features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911405413.7A CN111145222A (en) | 2019-12-30 | 2019-12-30 | Fire detection method combining smoke movement trend and textural features |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111145222A true CN111145222A (en) | 2020-05-12 |
Family
ID=70522327
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911405413.7A Pending CN111145222A (en) | 2019-12-30 | 2019-12-30 | Fire detection method combining smoke movement trend and textural features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111145222A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112071016A (en) * | 2020-09-14 | 2020-12-11 | 广州市几米物联科技有限公司 | Fire monitoring method, device, equipment and storage medium |
CN112766179A (en) * | 2021-01-22 | 2021-05-07 | 郑州轻工业大学 | Fire smoke detection method based on motion characteristic hybrid depth network |
CN112949453A (en) * | 2021-02-26 | 2021-06-11 | 南京恩博科技有限公司 | Training method of smoke and fire detection model, smoke and fire detection method and smoke and fire detection equipment |
CN114120171A (en) * | 2021-10-28 | 2022-03-01 | 华能盐城大丰新能源发电有限责任公司 | Fire smoke detection method, device and equipment based on video frame and storage medium |
CN114283569A (en) * | 2022-01-24 | 2022-04-05 | 北京鼎好鑫源科技有限公司 | Big data characteristic parameter early warning device and method |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103116746A (en) * | 2013-03-08 | 2013-05-22 | 中国科学技术大学 | Video flame detecting method based on multi-feature fusion technology |
CN105787458A (en) * | 2016-03-11 | 2016-07-20 | 重庆邮电大学 | Infrared behavior identification method based on adaptive fusion of artificial design feature and depth learning feature |
CN106203334A (en) * | 2016-07-11 | 2016-12-07 | 电子科技大学 | A kind of indoor scene flare detection method |
CN106599907A (en) * | 2016-11-29 | 2017-04-26 | 北京航空航天大学 | Multi-feature fusion-based dynamic scene classification method and apparatus |
CN107463968A (en) * | 2017-08-25 | 2017-12-12 | 太原理工大学 | Smog judges the detection method of the production method of code book, generation system and smog |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
CN108985192A (en) * | 2018-06-29 | 2018-12-11 | 东南大学 | A kind of video smoke recognition methods based on multitask depth convolutional neural networks |
CN109165577A (en) * | 2018-08-07 | 2019-01-08 | 东北大学 | A kind of early stage forest fire detection method based on video image |
CN109522965A (en) * | 2018-11-27 | 2019-03-26 | 天津工业大学 | A kind of smog image classification method of the binary channels convolutional neural networks based on transfer learning |
CN109784259A (en) * | 2019-01-08 | 2019-05-21 | 江河瑞通(北京)技术有限公司 | Water transparency intelligent identification Method and Sai Shi disk component based on image recognition |
CN109859199A (en) * | 2019-02-14 | 2019-06-07 | 浙江科技学院 | A kind of method of the fresh water pipless pearl quality testing of SD-OCT image |
CN110503021A (en) * | 2019-08-19 | 2019-11-26 | 温州大学 | Fire hazard smoke detecting method based on time compression track characteristic identification |
-
2019
- 2019-12-30 CN CN201911405413.7A patent/CN111145222A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103116746A (en) * | 2013-03-08 | 2013-05-22 | 中国科学技术大学 | Video flame detecting method based on multi-feature fusion technology |
CN105787458A (en) * | 2016-03-11 | 2016-07-20 | 重庆邮电大学 | Infrared behavior identification method based on adaptive fusion of artificial design feature and depth learning feature |
CN106203334A (en) * | 2016-07-11 | 2016-12-07 | 电子科技大学 | A kind of indoor scene flare detection method |
CN106599907A (en) * | 2016-11-29 | 2017-04-26 | 北京航空航天大学 | Multi-feature fusion-based dynamic scene classification method and apparatus |
CN107463968A (en) * | 2017-08-25 | 2017-12-12 | 太原理工大学 | Smog judges the detection method of the production method of code book, generation system and smog |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
CN108985192A (en) * | 2018-06-29 | 2018-12-11 | 东南大学 | A kind of video smoke recognition methods based on multitask depth convolutional neural networks |
CN109165577A (en) * | 2018-08-07 | 2019-01-08 | 东北大学 | A kind of early stage forest fire detection method based on video image |
CN109522965A (en) * | 2018-11-27 | 2019-03-26 | 天津工业大学 | A kind of smog image classification method of the binary channels convolutional neural networks based on transfer learning |
CN109784259A (en) * | 2019-01-08 | 2019-05-21 | 江河瑞通(北京)技术有限公司 | Water transparency intelligent identification Method and Sai Shi disk component based on image recognition |
CN109859199A (en) * | 2019-02-14 | 2019-06-07 | 浙江科技学院 | A kind of method of the fresh water pipless pearl quality testing of SD-OCT image |
CN110503021A (en) * | 2019-08-19 | 2019-11-26 | 温州大学 | Fire hazard smoke detecting method based on time compression track characteristic identification |
Non-Patent Citations (2)
Title |
---|
殷梦霞: "面向复杂场景的烟雾检测研究与实现", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》, no. 6, pages 1 - 74 * |
陈俊周 等: "基于级联卷积神经网络的视频动态烟雾检测", 《电子科技大学学报》, vol. 45, no. 6, pages 992 - 996 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112071016A (en) * | 2020-09-14 | 2020-12-11 | 广州市几米物联科技有限公司 | Fire monitoring method, device, equipment and storage medium |
CN112766179A (en) * | 2021-01-22 | 2021-05-07 | 郑州轻工业大学 | Fire smoke detection method based on motion characteristic hybrid depth network |
CN112949453A (en) * | 2021-02-26 | 2021-06-11 | 南京恩博科技有限公司 | Training method of smoke and fire detection model, smoke and fire detection method and smoke and fire detection equipment |
CN112949453B (en) * | 2021-02-26 | 2023-12-26 | 南京恩博科技有限公司 | Training method of smoke and fire detection model, smoke and fire detection method and equipment |
CN114120171A (en) * | 2021-10-28 | 2022-03-01 | 华能盐城大丰新能源发电有限责任公司 | Fire smoke detection method, device and equipment based on video frame and storage medium |
CN114283569A (en) * | 2022-01-24 | 2022-04-05 | 北京鼎好鑫源科技有限公司 | Big data characteristic parameter early warning device and method |
CN114283569B (en) * | 2022-01-24 | 2023-09-15 | 北京鼎好鑫源科技有限公司 | Big data characteristic parameter early warning device and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
Zhang et al. | Deep convolutional neural networks for forest fire detection | |
CN111460968B (en) | Unmanned aerial vehicle identification and tracking method and device based on video | |
CN109389185B (en) | Video smoke identification method using three-dimensional convolutional neural network | |
CN115205264A (en) | High-resolution remote sensing ship detection method based on improved YOLOv4 | |
CN109165602B (en) | Black smoke vehicle detection method based on video analysis | |
CN113469050A (en) | Flame detection method based on image subdivision classification | |
CN112464893A (en) | Congestion degree classification method in complex environment | |
Zhang et al. | Application research of YOLO v2 combined with color identification | |
CN115131747A (en) | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system | |
CN115661932A (en) | Fishing behavior detection method | |
CN109446938B (en) | Black smoke vehicle detection method based on multi-sequence double projection | |
Chen et al. | Real-time garbage object detection with data augmentation and feature fusion using SUAV low-altitude remote sensing images | |
CN117576632B (en) | Multi-mode AI large model-based power grid monitoring fire early warning system and method | |
He et al. | A novel image recognition algorithm of target identification for unmanned surface vehicles based on deep learning | |
CN113378668A (en) | Method, device and equipment for determining accumulated water category and storage medium | |
CN116205905B (en) | Power distribution network construction safety and quality image detection method and system based on mobile terminal | |
CN114821486B (en) | Personnel identification method in power operation scene | |
CN115100546A (en) | Mobile-based small target defect identification method and system for power equipment | |
Wang et al. | YOLOv5-light: efficient convolutional neural networks for flame detection | |
CN114550032A (en) | Video smoke detection method of end-to-end three-dimensional convolution target detection network | |
CN114898290A (en) | Real-time detection method and system for marine ship | |
CN113743304A (en) | Moving object detection and identification method for video monitoring | |
Jiang et al. | Fast Traffic Accident Identification Method Based on SSD Model | |
CN113379603A (en) | Ship target detection method based on deep learning |
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 |