Fire early warning system based on AI image recognition

Abstract

The invention relates to the technical field of fire early warning, and particularly discloses a fire early warning system based on AI image recognition, which comprises an image acquisition unit, a fire danger alarm unit, an image recognition unit, a remote control unit and a fire monitoring analysis unit, the image acquisition unit is used for shooting a monitoring scene in real time by using a camera and acquiring real-time video streams or static image data; the image recognition unit is used for optimizing the acquired image data, analyzing the optimized image data by using a deep learning algorithm, and recognizing possible smoke signs; according to the system, image data are shot and acquired in real time through the camera, and the data are analyzed by using a deep learning algorithm, so that smoke signs are effectively identified, and particularly, possible fire risks can be found in time in the initial stage of a fire. Different from a traditional smoke sensor and a traditional temperature sensor, the system can still keep high recognition accuracy under variable environmental conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of early fire warning, and in particular relates to an early fire warning system based on AI image recognition. Background Art

[0002] Traditional fire detection methods mainly rely on sensors, such as using temperature sensors to detect high temperatures and infrared sensors to detect flames. However, such methods have strong limitations. Sensors need to be placed near the fire point, which is not only extremely strict in terms of installation location, but also easy to be damaged over time and difficult to detect. In addition, for forest fire monitoring, due to the large outdoor scene, the temperature, humidity, and smoke are unevenly distributed, resulting in insufficient stability and reliability.

[0003] Using cameras to monitor forest environments is more efficient and stable than traditional methods. The objects to be identified in the camera capture image mainly include flames and smoke. In the initial stage of a fire, the fire is often accompanied by light, heat, smoke and other phenomena. However, in a wild forest scene, the flames may be concentrated at the bottom of the forest and blocked by other trees, resulting in a serious fire when it is discovered. The smoke formed by the particles produced by the combustion has a large diffusion range and a high rising height, which is easier to identify than the flame. At present, the ultra-long-distance cameras deployed in the forest can detect a wide area, with an effective observation distance of up to 5km, and have the advantages of fast response, anti-interference, and high efficiency. However, the smoke features are not obvious in the ultra-long-distance camera imaging, and it is difficult to detect the target of the smoke in the image. Early smoke recognition was mainly based on manually designed features. Such methods extracted and recognized low-dimensional information of smoke, such as detecting the color, shape, and motion state of smoke, but these methods did not work well in actual scenes.

[0004] Therefore, it is necessary to propose a fire early warning system based on AI image recognition to solve the problem in the existing technology that traditional smoke sensors and temperature sensors have low efficiency in fire warning due to the vastness and complexity of the area.

[0005] The above information disclosed in this background technology is only used to increase the understanding of the background technology of the present invention and therefore, it may include information that does not constitute the prior art known to ordinary technicians in this field. Summary of the invention

[0006] The purpose of the present invention is to provide a fire early warning system based on AI image recognition to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A fire early warning system based on AI image recognition, comprising: an image acquisition unit, a fire danger alarm unit, an image recognition unit, a remote control unit, a fire monitoring and analysis unit, a fire range analysis unit, and a system background monitoring unit;

[0009] The image acquisition unit is used to use a camera to shoot the monitoring scene in real time and obtain real-time video stream or static image data;

[0010] The image recognition unit is used to optimize the collected image data, analyze the optimized image data using a deep learning algorithm, and identify possible signs of smoke;

[0011] The fire monitoring and analysis unit is used to obtain environmental information of the monitoring scene, infer whether there is a fire risk based on smoke signs, mark smoke signs that indicate a fire risk, and generate analysis results;

[0012] The fire alarm unit is used to send alarms to the control center and user terminals through the 5G network based on the analysis results, and to take and save real-time photos of the fire area;

[0013] The remote control unit is used to provide an interface based on the 5G network, remote monitoring, management and control system;

[0014] The fire range analysis unit is used to estimate the fire range by using the fire spread prediction model based on the real-time image data combined with the analysis results;

[0015] The system background monitoring unit is used to provide real-time fire monitoring functions, displaying image data, analysis results and warning information of each monitoring scene.

[0016] Preferably, the image acquisition unit is also used to power the camera with a mobile power supply and match it with a wind-solar complementary power generation module according to the environmental information of the monitoring scene; determine the scanning strategy of the camera based on the parameter indicators of the camera combined with the environmental information of the monitoring scene; after shooting is completed, save the collected image data as a local file, and upload it to the cloud for storage through the 5G network.

[0017] Preferably, the image recognition unit is further used to use a Mosaic data enhancement method to perform random cropping, random scaling and random distribution operations on the image data, and splice them into a new picture to obtain an input image; use a dark channel prior algorithm to optimize the input image, and adjust the size of the input image by adaptive image scaling according to a preset image size;

[0018] Use YOLOv5s to analyze the adjusted input image and identify possible signs of smoke; perform secondary detection of smoke signs by introducing a convolutional neural network architecture with spatial attention to generate secondary detection results; when the secondary detection result is smoke, output the input image and select the signs of smoke on the input image.

[0019] Preferably, the fire monitoring and analysis unit is also used to obtain the focal length information of the camera and the angle data of the configured angle sensor, so as to infer the location where the smoke signs appear; based on the location where the smoke signs appear, infer whether there is a fire risk according to the environmental information of the monitoring scene, and generate a speculation result; mark the smoke signs that pose a fire risk according to the speculation result, and generate an analysis result.

[0020] Preferably, the fire hazard alarm unit is also used to generate warning information based on the analysis results, and send the warning information to the control center and the user terminal through the 5G network; conduct subsequent monitoring of the location where the marked smoke signs appear, and dynamically track the smoke signs, and update the location where the smoke signs appear in real time; generate new warning information based on the updated location of the smoke signs and the image data generated by subsequent monitoring, and send it through the 5G network.

[0021] Preferably, the remote control unit is also used to transmit real-time captured image data to firefighters and on-site staff through the 5G network to achieve remote command and dispatch; use AR technology to provide augmented reality information to achieve real-time remote collaboration between on-site staff and remote operators; remotely adjust the camera's parameter settings and scanning strategies according to the environmental information of the monitoring scene.

[0022] Preferably, the fire range analysis unit is also used to integrate wind speed and heat transfer models, smoke diffusion models, dynamically simulate the spread of fire according to real-time image data and analysis results, and generate a real-time expansion map of the fire; based on the real-time expansion map, the fire range within a preset time period is estimated through the Rothermel model according to the real-time image data; the fire range is transmitted to firefighters and on-site staff through the 5G network for planning potential threat areas.

[0023] Preferably, the system background monitoring unit is also used to view image data and warning information in real time, and initiate emergency response measures based on the analysis results; superimpose the real-time expansion map of the fire on the geographic information system to display the specific location affected by the fire; indicate the spread of the fire through different colors and shadows, and display the scope covered by the fire within a preset time period.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The system of the present invention uses a camera to capture and obtain image data in real time, and uses a deep learning algorithm to analyze the data, so as to effectively identify signs of smoke, especially in the early stages of a fire, and can promptly detect possible fire risks. Unlike traditional smoke sensors and temperature sensors, this system can still maintain a high recognition accuracy under changing environmental conditions and has a low false alarm rate. At the same time, the system also obtains environmental information of the monitoring scene, conducts a comprehensive analysis based on smoke signs, accurately determines the fire risk, and transmits the analysis results to the control center and user terminals in real time through the 5G network. This remote alarm and real-time data transmission not only improves the reaction speed, but also ensures that emergency personnel can respond quickly and take prevention and control measures in a timely manner.

[0026] The system of the present invention supports remote monitoring and management functions. Users can view real-time data at any time, remotely adjust system settings or perform emergency processing on emergencies, thus avoiding the problem that traditional fire monitoring systems require on-site manual operation.

[0027] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a framework diagram of the fire early warning system based on AI image recognition of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] Embodiment 1:

[0031] See also Figure 1 As shown, a fire early warning system based on AI image recognition includes: an image acquisition unit, a fire danger alarm unit, an image recognition unit, a remote control unit, a fire monitoring and analysis unit, a fire range analysis unit, and a system background monitoring unit;

[0032] The image acquisition unit is used to use a camera to shoot the monitoring scene in real time and obtain real-time video stream or static image data;

[0033] The image recognition unit is used to optimize the collected image data, analyze the optimized image data using a deep learning algorithm, and identify possible signs of smoke;

[0034] The fire monitoring and analysis unit is used to obtain environmental information of the monitoring scene, infer whether there is a fire risk based on smoke signs, mark smoke signs that indicate a fire risk, and generate analysis results;

[0035] The fire alarm unit is used to send alarms to the control center and user terminals through the 5G network based on the analysis results, and to take and save real-time photos of the fire area;

[0036] The remote control unit is used to provide an interface based on the 5G network to remotely monitor, manage and control the system, view real-time data, respond to emergencies, and optimize system settings;

[0037] The fire range analysis unit is used to estimate the fire range by using the fire spread prediction model based on the real-time image data combined with the analysis results;

[0038] The system background monitoring unit is used to provide real-time fire monitoring functions, displaying image data, analysis results and warning information of each monitoring scene.

[0039] The image acquisition unit is also used to monitor the environmental information of the scene. The camera is powered by a mobile power supply and is equipped with a wind-solar complementary power generation module;

[0040] Determine the camera's scanning strategy (such as continuous scanning strategy, fixed-point scanning strategy, etc.) based on the camera's parameter indicators and the environmental information of the monitoring scene;

[0041] After shooting, the collected image data is saved as a local file and uploaded to the cloud for storage via the 5G network.

[0042] The image recognition unit is also used to use the Mosaic data enhancement method to perform random cropping, random scaling, and random distribution operations on the image data, and splice them into a new picture to obtain an input image;

[0043] The input image is optimized using a dark channel prior algorithm, and the size of the input image is adjusted by adaptive image scaling according to a preset image size;

[0044] The dark channel calculation method is to first find the minimum RGB component of each pixel in the image to obtain a grayscale image of the same size as one channel of the original image, and then use a window with a radius of r to perform a minimum value filter on the grayscale image. The formula of the dark channel function is as follows:

[0045]

[0046] In the formula, x represents the pixel of the image, c represents each channel of the color image, Ω represents a window centered on the pixel x, and f dark and f c They represent the dark channel image and the image to be processed respectively.

[0047] Use YOLOv5s to analyze the adjusted input image and identify possible signs of smoke;

[0048] We crawled and screened forest fire smoke images from the Internet, used smoke bombs to simulate fire conditions, and took photos with cameras to obtain forest fire smoke images. Considering that the smoke image dataset obtained in this way is not enough to support YOLOv5s training, we used smoke synthesis to expand the dataset, and marked the real smoke data collected to obtain label data.

[0049] The convolutional neural network architecture uses real smoke images that were previously used for YOLOv5s training, and obtains the training data set by cropping the smoke areas of these images. The training data set is expanded using data augmentation, and the data augmentation methods used (such as horizontal flip, vertical flip, rotation, etc.) are tested on the convolutional neural network architecture, and the enhancement effects of smoke image features using different data augmentation methods are compared, and then the amount of smoke and non-smoke images in the training data set is redistributed.

[0050] A convolutional neural network architecture with spatial attention is used to perform secondary detection of smoke signs and generate secondary detection results.

[0051] The convolutional neural network architecture uses stacked convolutions to extract input smoke sign features, and uses pooling after each convolution layer to reduce computational complexity and expand the receptive field. In the middle layer, convolution and pooling operations of different scales are parallelized to extract multi-level features, and spatial attention is introduced to adjust the distribution of network weights, and then they are combined to widen the network width and the receptive field.

[0052] When the secondary detection result is smoke, the input image is output and smoke signs are selected on the input image.

[0053] The fire monitoring and analysis unit is also used to obtain the focal length information of the camera and the angle data of the configured angle sensor, and then infer the location where the smoke signs appear;

[0054] Based on the location where smoke signs appear, infer whether there is a fire risk based on the environmental information of the monitored scene and generate inference results;

[0055] The environmental information of the monitoring scene includes meteorological data (such as temperature, humidity, wind speed and direction, etc.) and geographical and historical data (such as fire-prone areas, historical fire data, etc.).

[0056] Based on the location of smoke signs and the environmental information of the monitoring scene, the possibility of fire can be comprehensively judged in the following ways:

[0057] Humidity matches smoke location: If the temperature around the location where smoke signs appear is abnormally high, especially high temperatures in a colder area and low humidity in the area, the possibility of fire will increase significantly. It can be combined with meteorological data to confirm whether it is a dry season or a fire-prone period.

[0058] Auxiliary analysis of wind speed and direction: According to the changes in wind speed and direction, if the location of the smoke sign is consistent with the wind direction and the wind speed is high, it means that the smoke is likely caused by fire. Strong winds can accelerate the spread of fire and cause smoke to spread rapidly.

[0059] Based on the inference, smoke signs that indicate fire risk are marked and analysis results are generated.

[0060] The fire alarm unit is also used to generate early warning information based on the analysis results and send the early warning information to the control center and user terminals through the 5G network;

[0061] Follow up monitoring of the locations where the marked smoke signs appear, and dynamically track the smoke signs, updating the locations where the smoke signs appear in real time;

[0062] Based on the updated location of smoke signs and image data generated by subsequent monitoring, new warning information is generated and sent via the 5G network.

[0063] Embodiment 2:

[0064] See also Figure 1 As shown, this embodiment is basically the same as the above embodiment, except that the remote control unit is also used to transmit the real-time captured image data to firefighters and on-site staff through the 5G network to achieve remote command and dispatch;

[0065] Using AR technology to provide augmented reality information for real-time remote collaboration between on-site staff and remote operators;

[0066] Remotely adjust the camera's parameter settings and scanning strategies based on the environmental information of the monitoring scene.

[0067] The fire range analysis unit is also used to integrate wind speed and heat transfer models, smoke diffusion models, dynamically simulate the spread of fire according to real-time image data and analysis results, and generate a real-time expansion map of the fire;

[0068] Based on the real-time expansion map, the fire range within the preset time period is estimated by the Rothermel model according to the real-time image data;

[0069] The Rothermel model is a mathematical model commonly used in forest fire propagation simulation. The model is mainly used to estimate the rate of forest fire spread, especially in forest fire management and early warning systems. The core of the Rothermel model is to predict the spread rate of fire based on the relationship between the burning rate of fire and environmental conditions, fuel characteristics and meteorological conditions.

[0070] The fire scope is transmitted to firefighters and on-site staff via the 5G network to map out potential threat areas.

[0071] The system background monitoring unit is also used to view image data and warning information in real time, and initiate emergency response measures based on the analysis results;

[0072] Overlaying real-time fire expansion maps onto a GIS to show specific locations of fire impacts;

[0073] The spread of fire is represented by different colors and shadows, showing the area covered by the fire within a preset time period.

[0074] As can be seen from the above, the system of the present invention uses a camera to capture and obtain image data in real time, and uses a deep learning algorithm to analyze the data, so as to effectively identify signs of smoke, especially in the early stages of a fire, and can promptly detect possible fire risks. Unlike traditional smoke sensors and temperature sensors, this system can still maintain a high recognition accuracy under changing environmental conditions and has a low false alarm rate. At the same time, the system also obtains environmental information of the monitoring scene, combines it with smoke signs for comprehensive analysis, accurately determines the fire risk, and transmits the analysis results to the control center and user terminals in real time through the 5G network. This remote alarm and real-time data transmission not only improves the reaction speed, but also ensures that emergency personnel can respond quickly and take prevention and control measures in a timely manner.

[0075] The system of the present invention supports remote monitoring and management functions. Users can view real-time data at any time, remotely adjust system settings or perform emergency processing on emergencies, thus avoiding the problem that traditional fire monitoring systems require on-site manual operation.

[0076] Embodiment 3:

[0077] The embodiment of the present invention further provides a computer-readable storage medium, on which is stored a program of a fire early warning system based on AI image recognition as any of the above items, which, when executed by a processor, implements each process of the above early warning system embodiment and can achieve the same technical effect, which will not be described here to avoid repetition. The computer-readable storage medium includes a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0078] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0079] In the drawings of the embodiments disclosed in the present invention, only the structures involved in the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0080] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0081] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fire early warning system based on AI image recognition, characterized in that: include: Image acquisition unit, fire danger alarm unit, image recognition unit, remote control unit, fire monitoring and analysis unit; The image acquisition unit is used to use a camera to shoot the monitoring scene in real time and obtain real-time video stream or static image data; The image recognition unit is used to optimize the collected image data, analyze the optimized image data using a deep learning algorithm, and identify possible signs of smoke; The fire monitoring and analysis unit is used to obtain environmental information of the monitoring scene, infer whether there is a fire risk based on the smoke signs, mark the smoke signs that indicate the fire risk, and generate analysis results; The fire danger alarm unit is used to alarm the control center and the user terminal through the 5G network according to the analysis result, and to take pictures and save the fire area in real time; The remote control unit is used to provide an interface based on the 5G network, and a remote monitoring, management and control system.

2. According to claim 1, a fire early warning system based on AI image recognition is characterized in that: The image acquisition unit is also used for: According to the environmental information of the monitoring scene, the camera is powered by a mobile power supply and is equipped with a wind-solar complementary power generation module; Determine the scanning strategy of the camera based on the parameter index of the camera and the environmental information of the monitoring scene; After the shooting is completed, the collected image data is saved as a local file and uploaded to the cloud for storage via the 5G network.

3. The fire early warning system based on AI image recognition according to claim 2 is characterized in that: The image recognition unit is also used for: Using Mosaic data augmentation method to randomly crop, randomly scale and randomly distribute the image data, splice them into a new picture, and obtain an input image; The input image is optimized using a dark channel prior algorithm, and the size of the input image is adjusted by adaptive image scaling according to a preset image size; Using YOLOv5s to analyze the adjusted input image to identify possible signs of smoke; Performing secondary detection on the smoke signs by introducing a convolutional neural network architecture with spatial attention to generate secondary detection results; When the secondary detection result is smoke, the input image is output and the smoke sign is framed on the input image.

4. The fire early warning system based on AI image recognition according to claim 3 is characterized in that: The fire monitoring and analysis unit is also used for: Obtaining focal length information of the camera and angle data of a configured angle sensor, and then inferring the location where the smoke sign appears; Based on the location where the smoke sign appears, infer whether there is a fire risk according to the environmental information of the monitoring scene, and generate an inference result; The smoke signs indicating the fire risk are marked according to the inference results to generate the analysis results.

5. The fire early warning system based on AI image recognition according to claim 4 is characterized in that: The fire alarm unit is also used for: Generate warning information according to the analysis result, and send the warning information to the control center and the user terminal through the 5G network; Perform subsequent monitoring on the marked location where the smoke sign appears, and dynamically track the smoke sign to update the location where the smoke sign appears in real time; Based on the updated location of the smoke signs and the image data generated by subsequent monitoring, new warning information is generated and sent via the 5G network.

6. The fire early warning system based on AI image recognition according to claim 5 is characterized in that: The remote control unit is also used for: The image data captured in real time is transmitted to firefighters and on-site staff through the 5G network for remote command and dispatch; Using AR technology to provide augmented reality information for achieving real-time remote collaboration between the on-site staff and the remote operators; The parameter settings of the camera and the scanning strategy are remotely adjusted according to the environmental information of the monitoring scene.

7. The fire early warning system based on AI image recognition according to claim 6 is characterized in that: The system also includes: a fire range analysis unit and a system background monitoring unit; The fire range analysis unit is used to estimate the fire range by using a fire spread prediction model based on the image data taken in real time and the analysis result; The system background monitoring unit is used to provide a real-time fire monitoring function and display the image data, analysis results and warning information of each monitoring scene.

8. The fire early warning system based on AI image recognition according to claim 7 is characterized in that: The fire range analysis unit is also used for: Integrate wind speed and heat transfer models and smoke diffusion models, dynamically simulate the spread of fire according to the real-time image data and the analysis results, and generate a real-time expansion map of the fire; Based on the real-time expansion map, the fire range within a preset time period is estimated by using a Rothermel model according to the image data taken in real time; The fire range is transmitted to the firefighters and the on-site staff via the 5G network for planning potential threat areas.

9. The fire early warning system based on AI image recognition according to claim 8, characterized in that: The system background monitoring unit is also used for: View the image data and the warning information in real time, and initiate emergency response measures based on the analysis results; superimposing a real-time expansion map of the fire onto a geographic information system to display specific locations affected by the fire; The spread of the fire is represented by different colors and shadows, showing the scope covered by the fire within a preset time period.

Images