Fire point identification system and method based on remote sensing technology and artificial intelligence algorithm

Abstract

The invention discloses a fire point recognition system and method based on a remote sensing technology and an artificial intelligence algorithm. According to the invention, fire detection is carried out through satellite remote sensing and ground mobile remote sensing images. The method mainly comprises the following steps: acquiring a satellite remote sensing image, processing the satellite remote sensing image by using a first convolutional neural network (CNN), generating a satellite vector diagram for identifying a fire point, automatically positioning ground remote sensing equipment according to the generated satellite vector diagram, and acquiring more accurate fire point condition information according to ground remote sensing. During execution, when the computer executes the instruction, a plurality of processors are called and coupled with the memory, and operations are executed through an artificial intelligence algorithm and a plurality of sets of machine learning algorithms, so that relevant data near a fire point are further judged according to the artificial intelligence algorithm, the fire situation is confirmed, the disaster situation is guided, and the data is evaluated.

Description

Technical Field

[0001] A fire point identification system and method based on remote sensing technology and artificial intelligence algorithm. Background Art

[0002] The use of satellite remote sensing technology to monitor and warn of forest fires has become one of the important ways to solve the problem of outdoor forest fire disasters. Monitoring through satellite remote sensing can avoid the various concurrent disasters caused by powered ground sensors such as cameras. Satellite remote sensing warning systems can observe fire conditions in real time to control the spread of wildfires and improve the prevention of loss of life and property. However, systems based on satellite images are prone to high false alarms and are not usually used as the only source for detecting wildfires. Other near-ground systems use sensors or videos on the ground, which are much better in terms of false alarm rates, but have the disadvantage of only covering part of the land susceptible to wildfires and are expensive to install and operate. Summary of the invention

[0003] The present invention discloses a method for jointly monitoring fires based on satellite remote sensing and ground mobile remote sensing, which solves the problems of high false alarms in satellite image-based systems and the limitations of near-ground fixed sensors, accurately identifies the exact location of the fire point through an artificial intelligence algorithm, and analyzes and evaluates the comprehensive situation of the fire.

[0004] In order to achieve the purpose of the above invention, the technical solution adopted by the present invention is: acquiring satellite remote sensing images; using the first convolutional neural network (CNN) to process the satellite remote sensing images and identify the satellite vectors of multiple fire points; and automatically guiding the ground mobile remote sensing equipment sensor to a certain position according to the satellite vector.

[0005] The implementation of the present invention requires a device including a memory and a plurality of processors, the processors being coupled to the memory and operating to execute the method steps of the system analysis output. Figure 1 The system method and cloud computing nodes that can be used to implement the present invention are described.

[0006] There are two main sensing methods for outdoor fire point monitoring described in the present invention: satellite-based remote sensing provides wide coverage and drone-based remote sensing provides local high-precision coverage. Through the technology of combining these two monitoring methods with remote sensing images, and using artificial intelligence algorithms, satellite remote sensing images are used as early warning prompts to control the positioning route of drones to maximize the implementation of discovering and covering fire points. The artificial intelligence algorithm uses multi-scale CNN (convolutional neural network) to extract texture maps from satellite remote sensing images and is implemented using a machine reinforcement learning system.

[0007] The artificial intelligence algorithm integrates the sensing based on satellite remote sensing images with the mobile sensor network of the drone. The mobile sensor is any type of controllable mobile sensor, and the analysis results of the satellite remote sensing image provide the initial prompts for the operation of the drone. Reinforcement learning is then used to guide the flight path of the drone. The system is guided by various acquired action optimization functions.

[0008] The function reinforcement learning visualizes the state space as a two-dimensional grid for UAV navigation and guides the flight path of the UAV based on reinforcement learning optimization. In the reinforcement learning method, the UAV learns to use the full coverage image provided by satellite remote sensing and its own zero coverage image to optimize the navigation route.

[0009] In the reinforcement learning, the satellite remote sensing imagery and the output of the drone cover the same area, but each pixel in the corresponding output represents a different area size. The satellite model output pixel covers a larger area than the drone model output pixel, but there is a unique satellite remote sensing output pixel that covers the target pixel of the drone output. The multi-scale product neural network processes the satellite and drone historical images at different life stages to be trained sequentially or shadowed separately. The input data set includes all images from satellite remote sensing and drone remote sensing, and the corresponding images output the demand indication route for fire point monitoring through reinforcement learning.

[0010] The convolutional neural network includes multiple computer processors that can work together to implement one or more machine learning algorithms, which can be synchronous or asynchronous. In the convolutional neural network, it combines all input values ​​together to implement supervised, unsupervised or semi-supervised machine learning.

[0011] The reinforcement learning system satellite vectors and drone vectors basically comb through satellite and drone imagery and reposition the drone in the air when necessary. It combines feature maps extracted from separate drone motion sensors and satellite models, where the combined map includes features from both models and is continuously optimized. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 A cloud computing node diagram that can be used to implement the system method and implementation of the present invention; Figure 2 Flow chart for detecting wide area fires using mobile sensors; Figure 3 Flowchart of the image convolutional neural network (CNN) for satellite remote sensing image CNN and UAV remote sensing image CNN.

Claims

1. A fire point identification system and method based on remote sensing technology and artificial intelligence algorithm, characterized in that: The system automatically acquires satellite remote sensing images; uses the first convolutional neural network (CNN) to process the satellite remote sensing images, and generates satellite vectors to identify the fire points in the monitoring area; after positioning according to the satellite vectors, the ground mobile remote sensing monitoring equipment automatically acquires the positioned position, and acquires the ground remote sensing images of the positioning area to further determine the relevant data near the fire point and confirm the fire situation. The ground mobile remote sensing monitoring equipment is a remotely controlled drone device, and the device is equipped with an artificial intelligence monitoring system.

2. According to claim 1, a ground remote sensing image is obtained from a ground mobile remote sensing monitoring device, the image is processed using a second convolutional neural network (CNN), and a ground remote sensing vector is generated to identify the fire situation in the monitoring area; according to the satellite vector obtained according to claim 1 and the ground remote sensing vector obtained above, the location of the fire point in the detection area is narrowed down. Each pixel in the satellite vector and the ground mobile remote sensing vector captures the intensity data of the fire point in the corresponding geographical area.

3. According to claim 1, the artificial intelligence monitoring system can observe the location of suspected fire points, the positioning of the surrounding social building living areas and the fire location, the assessment of the disaster losses caused by the fire, and the impact of meteorological conditions such as temperature and humidity in the fire area on the intensity of the fire.

4. According to claim 3, the artificial intelligence monitoring system includes action control and state control. The action control is a set of defined allowable actions, and the state control is the current environment encountered by the system, which can be customized according to one or more factors of sensor direction, aperture control, field of view control and spectrum control.

5. According to claim 1, the artificial intelligence monitoring system is generated based on the current state of the first reinforcement learning system of the actions selected by the first convolutional neural network, and an indication of the success or failure of the selected action sequence to achieve a defined goal.

6. According to claim 2, the model of the satellite remote sensing image processed by the first convolutional neural network (CNN) is outputted at multiple scales, and the ground mobile remote sensing monitoring equipment uses the second convolutional neural network (CNN) to process the image at multiple scales, and finally the tensors of the satellite remote model outputted by the first convolutional neural network (CNN) and the ground remote sensing equipment model processed by the second convolutional neural network (CNN) are combined and connected.

7. The method of claim 6, wherein the multi-scale output of the satellite CNN model and the multi-scale output of the mobile ground remote sensing CNN model are combined by combining feature maps extracted from mobile sensors of ground mobile remote sensing monitoring equipment and satellite remote sensing models. The first convolutional neural network (CNN) outputs the result after reinforcement learning Q-Learning and Q-table.

8. According to claim 5, the artificial intelligence monitoring system implements a second reinforcement learning system to control the remote satellite, wherein the state space and action space of the second reinforcement learning system are specific to satellite control, and the second reinforcement learning system initially learns using movement vectors and then learns using movement vectors and satellite vectors simultaneously. The satellite control includes a first control to determine an area to be sensed, a second control to determine an extent of a spatial area to be sensed, and a third control to determine a frequency range to be sensed.

9. A fire point identification system and method based on remote sensing technology and artificial intelligence algorithm, which requires a non-transmittable computer-readable medium containing computer-executable instructions. When the computer executes these instructions, the computer performs the following operations: acquiring satellite images, processing the satellite images using a first convolutional neural network (CNN), generating satellite vectors identifying multiple fire points, and automatically orienting a mobile sensor to a mobile position according to the satellite vectors.

10. According to claim 9, the non-transmissive computer-readable medium comprises a plurality of processors and a memory, wherein the plurality of processors are connected to the memory and are capable of performing operations including acquiring satellite images, using a first convolutional neural network to process the satellite images to identify satellite vectors to a plurality of fire points, and automatically guiding a mobile sensor to a mobile position according to the satellite vectors.

Images