Fire early warning system based on artificial intelligence

Abstract

The present invention relates to the field of fire warning, and in particular to a fire warning system based on artificial intelligence. The present invention obtains inspection images through an image inspection module, constructs a virtual reference frame through a boundary analysis module, and analyzes boundary difference characteristics based on image parameter differences of each segmented area after each virtual reference frame is segmented by an intersection boundary. The distribution of pixel points in a strip area around the intersection boundary in each virtual reference frame is analyzed through a continuous analysis module to determine the boundary distribution characteristics. Feature warning characterization parameters are calculated based on a calibration module to calibrate the corresponding inspection image. The inspection analysis module performs warning analysis in response to the calibration result of the inspection image. The present invention utilizes the relatively prominent characteristic of smoke characteristics against the sky background to adaptively analyze the inspection image, improve analysis efficiency, and improve the reliability and accuracy of fire warning for forests at night.

Description

Technical Field

[0001] The present invention relates to the field of fire warning, and in particular to a fire warning system based on artificial intelligence. Background Art

[0002] Traditional forest fire monitoring relies primarily on ground patrols, observation towers, and satellite remote sensing technology. However, these methods suffer from delayed response times, numerous blind spots, and low data resolution. For example, satellite imagery is limited by cloud cover and revisit cycles, making real-time early warning difficult. Manual inspections are also inefficient due to complex terrain and labor costs. Against this backdrop, drone technology, with its flexibility, high spatiotemporal resolution, and multi-sensor integration, is becoming a crucial addition to forest fire early warning systems. Drones can carry various sensors for inspections, analyze the acquired data, and promptly detect forest fire hazards.

[0003] For example, Chinese patent publication number CN108416963A discloses a deep learning-based forest fire warning method and system, which belongs to the field of fire safety. The method includes: S1, a drone senses the temperature and humidity information of the driving area in real time, captures and transmits forest images of the driving area to a ground station; and simultaneously sends location information to the ground station; S2, a fire warning judgment is made based on the temperature and humidity information, and a fire warning signal is sent to the ground station; S3, the ground station receives the fire warning signal and uses a deep learning algorithm to process the forest image to determine whether a fire is about to occur or exists; S4, the fire information is sent to the forest management center. By sensing temperature and humidity information and performing comparative processing, the possibility of fire in the area can be intuitively determined. If a fire is likely to occur, the forest image is processed using a deep learning algorithm to accurately identify fire pre-occurrence and existing fires, allowing relevant personnel to accurately and immediately be informed of the fire situation.

[0004] However, the prior art still has the following problems:

[0005] To ensure inspection efficiency when inspecting large areas of forest, the drone needs to maintain a certain altitude. In addition, there are many obstructions in the forest, which block infrared signals. It may be impossible to locate the heat source in the early stages of a fire. Therefore, it is necessary to determine the fire situation based on the smoke characteristics in the image by taking images. However, at night, the smoke characteristics are not obvious against the forest background and are difficult to identify, which can easily lead to misjudgment and low analysis efficiency. Summary of the Invention

[0006] To this end, the present invention provides an artificial intelligence-based fire warning system to solve the problem in the prior art that when obtaining images of large-scale forests for fire warning, the smoke features are not obvious against the forest background, making them difficult to identify, easily leading to misjudgment and low analysis efficiency.

[0007] To achieve the above objectives, the present invention provides an artificial intelligence-based fire warning system, which includes:

[0008] An image detection module, which includes a number of mobile inspection units for acquiring inspection images;

[0009] a boundary analysis module connected to the image detection module, configured to identify the intersection boundaries of predetermined feature areas based on the inspection image, construct a plurality of virtual reference frames along the intersection boundaries, and analyze boundary difference characteristics based on image parameter differences of the segmented areas after each virtual reference frame is segmented by the intersection boundaries;

[0010] a continuous analysis module connected to the boundary analysis module, configured to analyze the distribution of pixels in a strip around the intersection boundary of each virtual reference frame to determine boundary distribution characteristics;

[0011] a calibration module, connected to the boundary analysis module and the continuous analysis module respectively, for calculating characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics to calibrate the corresponding inspection image;

[0012] An early warning analysis module is connected to the calibration module and performs early warning analysis in response to the calibration results of the inspection image, including:

[0013] A unique virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the unique virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. The corresponding coordinate set of unique pixel points in the sky segmentation area of ​​the inspection image obtained at each moment is used to determine whether to issue a fire warning signal;

[0014] or, controlling the mobile inspection unit to move along a predetermined inspection route and continuously acquiring inspection images;

[0015] The segmented areas include sky segmented areas and non-sky segmented areas.

[0016] Furthermore, the continuous analysis module is used to identify the intersection boundary of the predetermined feature area based on the inspection image, and construct a plurality of virtual reference frames along the intersection boundary, including:

[0017] Used to identify predetermined feature areas, including sky areas and non-sky areas;

[0018] Used to determine the boundary between the sky area and the non-sky area;

[0019] for calibrating a plurality of reference points along the intersection boundary, and constructing a plurality of virtual reference frames with each of the reference points as the center;

[0020] To determine the sky segmentation area and the non-sky segmentation area after the virtual reference frame is divided by the intersection boundary;

[0021] The bottom edge of the virtual reference frame is parallel to the bottom edge of the inspection image.

[0022] Furthermore, the continuous analysis module is used to analyze boundary difference features based on image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary, including:

[0023] To solve the image parameter standard deviation between the sky segmentation areas in each virtual reference frame respectively;

[0024] To respectively solve the standard deviation of image parameters between non-sky segmentation areas in each virtual reference frame;

[0025] for performing weighted summation of the image parameter standard deviation for the sky segmentation area and the image parameter standard deviation for the non-sky segmentation area to obtain the boundary difference feature;

[0026] The image parameter standard deviation is the mean of the brightness standard deviation and the chromaticity standard deviation.

[0027] Furthermore, the distribution of pixels in the strip area around the intersection boundary of each virtual reference frame is analyzed to determine the boundary distribution characteristics, including:

[0028] The virtual intersection boundary is moved in the virtual reference frame by a predetermined distance in the direction of the sky segmentation area to obtain a virtual intersection boundary, so that the virtual intersection boundary and the intersection boundary form the strip area;

[0029] Determine the characteristic pixel based on the grayscale difference ratio between the pixel and its adjacent pixels;

[0030] Used to determine the density of feature similarity points as boundary distribution features;

[0031] If the grayscale difference ratio corresponding to any adjacent pixel of a pixel is greater than a predetermined grayscale difference ratio threshold, the pixel is determined as a feature pixel.

[0032] Furthermore, the calibration module is used to calculate the characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics, including:

[0033] The ratio of the boundary difference feature to the predetermined boundary difference feature threshold is used to calculate the boundary difference factor;

[0034] The ratio of the boundary distribution feature to the predetermined boundary distribution feature threshold is used to calculate the boundary distribution factor;

[0035] The boundary difference factor and the boundary distribution factor are weighted and summed to obtain the characteristic warning characterization parameter.

[0036] Furthermore, the calibration module calibrates the inspection image including:

[0037] If the characteristic warning representation parameter is greater than or equal to a preset characteristic warning representation threshold, it is determined that the inspection image needs to be calibrated.

[0038] Furthermore, the early warning analysis module performs early warning analysis in response to the calibration results of the inspection image, wherein:

[0039] If the inspection image is calibrated, a specific virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the specific virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. Based on the coincidence degree of the corresponding coordinate set of the specific pixel points in the sky segmentation area in the inspection image obtained at each moment, it is determined whether to issue a fire warning signal;

[0040] If the inspection image is not calibrated, the mobile inspection unit is controlled to move along a predetermined inspection route to continuously acquire inspection images.

[0041] Furthermore, the early warning analysis module determines a specific virtual reference frame based on the inspection image, and determines the moving direction of the mobile inspection unit based on the specific virtual reference frame, including:

[0042] Used to verify the image parameter difference ratio between the sky segmentation area within the virtual reference frame and the remaining sky segmentation areas;

[0043] for determining the virtual reference frame corresponding to the maximum image parameter difference ratio as a specific virtual reference frame;

[0044] The direction of the specific virtual reference frame is determined, and the direction is determined as the moving direction.

[0045] Furthermore, the warning analysis module analyzes the changes in basic features of pixels in the sky segmentation area of ​​the inspection image at each moment to determine whether to issue a fire warning signal, including:

[0046] Used to identify unique pixels within the sky segmentation area;

[0047] To determine the coordinates of specific pixel points and obtain a coordinate set for each specific pixel point;

[0048] To compare the coordinate sets of corresponding specific pixel points in the inspection images at adjacent times and determine the degree of coincidence of the coordinate sets;

[0049] If the mean value of the coordinate set coincidence is less than or equal to the preset coordinate set coincidence threshold, it is determined that a fire warning signal needs to be issued;

[0050] If the grayscale value of a pixel does not fall within a predetermined grayscale range, the pixel is determined to be a unique pixel.

[0051] Furthermore, the early warning analysis module is used to control the mobile inspection unit to move in the corresponding moving direction and maintain the position to continuously obtain the inspection image in the moving direction, including:

[0052] It is used to control the mobile inspection unit to move to the corresponding moving direction and then hover to maintain the position and continuously obtain the inspection image of the predetermined length of time in the moving direction.

[0053] Compared with the prior art, the present invention sets an image detection module, a boundary analysis module, a continuous analysis module, a calibration module and an early warning analysis module, obtains inspection images through the image inspection module, constructs a virtual reference frame through the boundary analysis module, and analyzes boundary difference characteristics based on the image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary, analyzes the distribution of pixel points in the strip area around the intersection boundary in each virtual reference frame through the continuous analysis module, determines the boundary distribution characteristics, calculates feature early warning characterization parameters based on the calibration module to calibrate the corresponding inspection image, and the inspection analysis module performs early warning analysis in response to the calibration results of the inspection image. The present invention utilizes the relatively prominent characteristics of smoke features against the sky background to adaptively analyze the inspection image, improve analysis efficiency, and improve the reliability and accuracy of forest fire warnings at night.

[0054] In particular, the present invention constructs several virtual reference frames along the intersection boundary, and analyzes the boundary difference characteristics based on the image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary. In actual situations, the inspection image is the image in front of the mobile inspection unit. Usually, the inspection image contains the sky, mountains and forests. Usually, smoke will float up. Due to heat attenuation, the smoke will become more and more scattered as it moves away from the forest until it dissipates. When approaching the forest, the smoke characteristics are relatively obvious in the inspection image. Moreover, since the sky contains fewer features, the smoke characteristics will be more obvious in the sky area of ​​the inspection image. Therefore, the present invention constructs a virtual reference frame, and then segments the sky segmentation area and the non-sky segmentation area, focusing on observing the sky near the forest. The segmented area can highlight the potential smoke characteristics against the background of a pure sky. That is, if there is smoke, the image parameters in each sky segmentation area will be different. Based on this, the present invention considers the standard deviation of the image parameters in the sky segmentation area. At the same time, combined with the standard deviation of the image parameters between the non-sky segmentation areas, a weighted method is used to obtain the boundary difference characteristics. It can only analyze the image within the virtual reference frame, reduce the introduction of interference, reduce the amount of data analysis, and quickly identify potential abnormal tendencies, thereby providing support for the subsequent calculation of feature warning characterization parameters, and then calibrating the inspection image, which is convenient for subsequent adaptive warning analysis based on the calibration structure, improving analysis efficiency, and improving the reliability and accuracy of forest fire warnings at night.

[0055] In particular, the present invention determines the boundary distribution characteristics based on the distribution of pixel points in the strip area around the intersection boundary, and further constructs the strip area in the virtual reference frame for the purpose of only considering the situation of pixel points near the intersection boundary. In actual conditions, both the sky segmentation area and the non-sky segmentation area will have some features that affect the judgment in the night environment, such as the looming outline of leaves in the forest, the outline of clouds or stars in the sky, etc. Therefore, it is best to only consider the local sky area close to the forest, that is, the strip area, to introduce fewer interference features, and further observe the distribution of pixel points therein to determine the density of relatively isolated characteristic pixel points therein as the boundary distribution feature, to characterize the relatively scattered features presented in the inspection image after the smoke dissipates, and then subsequently calculate the feature warning characterization parameters, comprehensively characterize the potential abnormal tendencies, and calibrate the corresponding inspection images, so as to facilitate the subsequent adaptive warning analysis based on the calibration structure, improve the analysis efficiency, and improve the reliability and accuracy of fire warning for forests at night.

[0056] In particular, the present invention performs early warning analysis in response to the calibration results of the inspection images. The calibrated inspection images have a high tendency to have abnormalities. Therefore, a virtual reference frame where the abnormalities are more obvious is determined in a targeted manner. After the moving direction is determined based on this, the drone is controlled to approach for verification. Inspection images at multiple moments are captured in a hovering state. The properties of the dynamic diffusion of smoke and the relatively prominent properties of smoke features in the sky segmentation area are utilized. The overlap of the corresponding coordinate sets of specific pixel points is considered to verify whether the smoke is flowing and spreading, and then determine whether a fire warning signal is issued, thereby improving the reliability and accuracy of fire warnings for forests at night. The drone is controlled to approach for verification only for the marked inspection images, thereby improving the efficiency of inspecting large areas of forests. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of the structure of a fire warning system based on artificial intelligence according to an embodiment of the invention;

[0058] Figure 2 A logic block diagram of calibrating inspection images according to an embodiment of the invention;

[0059] Figure 3 A logic block diagram of performing early warning analysis in response to calibration results of inspection images according to an embodiment of the invention;

[0060] Figure 4 This is a logic block diagram of an embodiment of the invention for determining whether to issue a fire warning signal. DETAILED DESCRIPTION

[0061] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0062] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] See also Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the structure of a fire warning system based on artificial intelligence according to an embodiment of the invention. Figure 2 This is a logic block diagram of calibrating inspection images according to an embodiment of the invention. Figure 3 This is a logic block diagram of an embodiment of the invention for performing early warning analysis in response to the calibration results of the inspection image. Figure 4 This is a logic block diagram of an embodiment of the invention for determining whether to issue a fire warning signal. The fire warning system based on artificial intelligence of the present invention includes:

[0064] An image detection module, which includes a number of mobile inspection units for acquiring inspection images;

[0065] a boundary analysis module connected to the image detection module, configured to identify the intersection boundaries of predetermined feature areas based on the inspection image, construct a plurality of virtual reference frames along the intersection boundaries, and analyze boundary difference characteristics based on image parameter differences of the segmented areas after each virtual reference frame is segmented by the intersection boundaries;

[0066] a continuous analysis module connected to the boundary analysis module, configured to analyze the distribution of pixels in a strip around the intersection boundary of each virtual reference frame to determine boundary distribution characteristics;

[0067] a calibration module, connected to the boundary analysis module and the continuous analysis module respectively, for calculating characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics to calibrate the corresponding inspection image;

[0068] An early warning analysis module is connected to the calibration module and performs early warning analysis in response to the calibration results of the inspection image, including:

[0069] A unique virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the unique virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. The corresponding coordinate set of unique pixel points in the sky segmentation area of ​​the inspection image obtained at each moment is used to determine whether to issue a fire warning signal;

[0070] or, controlling the mobile inspection unit to move along a predetermined inspection route and continuously acquiring inspection images;

[0071] The segmented areas include sky segmented areas and non-sky segmented areas.

[0072] Specifically, there is no limitation on the specific structure of the mobile inspection unit. The mobile inspection unit may be a drone, and the corresponding inspection images are obtained by carrying a photographic device on the drone. Those skilled in the art may select the drone model according to their needs, which will not be elaborated here.

[0073] Specifically, the shooting direction of the photographic device is preferably the same as the moving direction of the mobile inspection unit, so that the acquired inspection image can include the sky area.

[0074] Specifically, there is no limitation on the specific structures of the boundary analysis module, the continuous analysis module, the calibration module and the early warning analysis module, and they can all be composed of logic components, including field programmable processors, computers or microprocessors in computers.

[0075] Specifically, the continuous analysis module is used to identify the intersection boundaries of predetermined feature areas based on the inspection image, and to construct a number of virtual reference frames along the intersection boundaries, including:

[0076] Used to identify predetermined feature areas, including sky areas and non-sky areas;

[0077] Used to determine the boundary between the sky area and the non-sky area;

[0078] for calibrating a plurality of reference points along the intersection boundary, and constructing a plurality of virtual reference frames with each of the reference points as the center;

[0079] To determine the sky segmentation area and the non-sky segmentation area after the virtual reference frame is divided by the intersection boundary;

[0080] The bottom edge of the virtual reference frame is parallel to the bottom edge of the inspection image.

[0081] Specifically, there is no limitation on the method of identifying the sky area. An image processing model or algorithm that can realize the corresponding function can be pre-trained, and logical components can be imported to realize the corresponding function. This will not be repeated here.

[0082] Specifically, the intersection boundary is essentially the boundary between the mountain and the forest and the sky in the inspection image, which will not be described in detail.

[0083] Specifically, the lateral distances between the set reference points are equal, and the lateral distances are determined based on the bottom length of the inspection image and are selected between 0.1 times and 0.2 times the bottom length.

[0084] Specifically, the virtual reference frame is a rectangular frame, the bottom side of which is less than or equal to the horizontal distance, and the side length is determined based on the maximum vertical width of the sky area and is set between 0.3 times and 0.5 times the maximum vertical width.

[0085] Specifically, the continuous analysis module is used to analyze boundary difference features based on image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary, including:

[0086] To solve the image parameter standard deviation between the sky segmentation areas in each virtual reference frame respectively;

[0087] To respectively solve the standard deviation of image parameters between non-sky segmentation areas in each virtual reference frame;

[0088] for performing weighted summation of the image parameter standard deviation for the sky segmentation area and the image parameter standard deviation for the non-sky segmentation area to obtain the boundary difference feature;

[0089] The image parameter standard deviation is the mean of the brightness standard deviation and the chromaticity standard deviation.

[0090] In implementation, the weight of the image parameter standard deviation for the sky segmentation area is 0.8, and the weight of the image parameter standard deviation for the non-sky segmentation area is 0.2 during weighted summation.

[0091] In practice, before calculating the standard deviation of the image parameters, normalization processing is required to scale the brightness or chromaticity values ​​to the same range, which will not be described in detail.

[0092] The present invention constructs several virtual reference frames along the intersection boundary, and analyzes the boundary difference characteristics based on the image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary. In actual situations, the inspection image is the image in front of the mobile inspection unit. Usually, the inspection image contains the sky, mountains and forests. Usually, smoke will float up. Due to heat attenuation, the smoke will become more and more scattered as it moves away from the forest until it dissipates. When approaching the forest, the smoke characteristics are relatively obvious in the inspection image. Moreover, since the sky contains fewer features, the smoke characteristics will be more obvious in the sky area of ​​the inspection image. Therefore, the present invention constructs a virtual reference frame, and then segments the sky segmentation area and the non-sky segmentation area, focusing on observing the sky segmentation area near the forest. The segmented area can highlight the potential smoke characteristics against the background of a pure sky. That is, if there is smoke, the image parameters in each sky segmentation area will be different. Based on this, the present invention considers the standard deviation of the image parameters in the sky segmentation area. At the same time, combined with the standard deviation of the image parameters between the non-sky segmentation areas, a weighted method is used to obtain the boundary difference characteristics. It can only analyze the image within the virtual reference frame, reduce the introduction of interference, reduce the amount of data analysis, quickly identify potential abnormal tendencies, and provide support for the subsequent calculation of feature warning characterization parameters, and then calibrate the inspection image, so as to facilitate the subsequent adaptive warning analysis based on the calibration structure, improve the analysis efficiency, and improve the reliability and accuracy of forest fire warning at night.

[0093] Specifically, it is used to analyze the distribution of pixels in the strip area around the intersection boundary of each virtual reference frame and determine the boundary distribution characteristics, including:

[0094] The virtual intersection boundary is moved in the virtual reference frame by a predetermined distance in the direction of the sky segmentation area to obtain a virtual intersection boundary, so that the virtual intersection boundary and the intersection boundary form the strip area;

[0095] Determine the characteristic pixel based on the grayscale difference ratio between the pixel and its adjacent pixels;

[0096] Used to determine the density of feature similarity points as boundary distribution features;

[0097] If the grayscale difference ratio corresponding to any adjacent pixel of a pixel is greater than a predetermined grayscale difference ratio threshold, the pixel is determined as a feature pixel.

[0098] Specifically, the predetermined distance is 0.5 times the virtual reference frame.

[0099] Specifically, the purpose of setting the grayscale difference ratio threshold is to characterize the situation where the difference between a pixel and its adjacent pixels is large, and the grayscale difference ratio threshold is selected within the interval [0.3, 0.5].

[0100] It is understandable that there is at least one adjacent pixel point around the characteristic pixel point that is significantly different from it, reflecting the relative isolation of the characteristic pixel point. For example, the characteristic pixel point may be a scattered single pixel point formed after the smoke diffuses, or it may be a pixel point on a continuous smoke line, or it may be other situations, which will not be elaborated here.

[0101] The present invention determines the boundary distribution characteristics based on the distribution of pixel points in the strip area around the intersection boundary. The purpose of further constructing the strip area in the virtual reference frame is to only consider the situation of the pixel points near the intersection boundary. In actual conditions, both the sky segmentation area and the non-sky segmentation area will have some features that affect the judgment in the night environment, such as the looming outline of leaves in the forest, the outline of clouds or stars in the sky, etc. Therefore, it is best to only consider the local sky area close to the forest, that is, the strip area, to introduce fewer interference features, and further observe the distribution of pixel points therein to determine the density of relatively isolated characteristic pixel points therein as the boundary distribution feature, to characterize the relatively scattered features presented in the inspection image after the smoke dissipates, and then subsequently calculate the feature warning characterization parameters, comprehensively characterize the potential abnormal tendencies, and calibrate the corresponding inspection images, so as to facilitate the subsequent adaptive warning analysis based on the calibration structure, improve the analysis efficiency, and improve the reliability and accuracy of fire warning for forests at night.

[0102] Specifically, the calibration module is used to calculate the characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics, including:

[0103] The ratio of the boundary difference feature to the predetermined boundary difference feature threshold is used to calculate the boundary difference factor;

[0104] The ratio of the boundary distribution feature to the predetermined boundary distribution feature threshold is used to calculate the boundary distribution factor;

[0105] The boundary difference factor and the boundary distribution factor are weighted and summed to obtain the characteristic warning characterization parameter.

[0106] Specifically, the boundary difference feature threshold and the boundary distribution feature threshold are pre-selected and set, where:

[0107] Obtain in advance several inspection image samples of fires in the sky area and the non-sky area, calculate the boundary difference feature and the boundary distribution feature based on each image sample, and then solve the boundary difference feature mean and the boundary distribution feature mean;

[0108] The boundary difference feature threshold is set as the product of the boundary difference feature mean and the offset parameter, and the boundary distribution feature mean is set as the product of the boundary distribution feature mean and the offset parameter. The offset parameter is selected in the interval [1.15, 1.3].

[0109] Specifically, the weight of the boundary difference factor in the weighted sum is 0.6, and the weight of the boundary distribution factor is 0.4.

[0110] Specifically, the calibration module calibrates the inspection image including:

[0111] If the characteristic warning representation parameter is greater than or equal to a preset characteristic warning representation threshold, it is determined that the inspection image needs to be calibrated.

[0112] Specifically, the characteristic warning characterization threshold is selected in the interval [1.05, 1.25].

[0113] Specifically, the early warning analysis module performs early warning analysis in response to the calibration results of the inspection image, wherein:

[0114] If the inspection image is calibrated, a specific virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the specific virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. Based on the coincidence degree of the corresponding coordinate set of the specific pixel points in the sky segmentation area in the inspection image obtained at each moment, it is determined whether to issue a fire warning signal;

[0115] If the inspection image is not calibrated, the mobile inspection unit is controlled to move along a predetermined inspection route to continuously acquire inspection images.

[0116] Specifically, the mobile inspection unit may move along a predetermined inspection route to inspect the forest in the target area. The inspection route is pre-set by those skilled in the art and will not be described in detail here.

[0117] Specifically, the early warning analysis module determines a specific virtual reference frame based on the inspection image, and determines the moving direction of the mobile inspection unit based on the specific virtual reference frame, including:

[0118] Used to verify the image parameter difference ratio between the sky segmentation area within the virtual reference frame and the remaining sky segmentation areas;

[0119] for determining the virtual reference frame corresponding to the maximum image parameter difference ratio as a specific virtual reference frame;

[0120] The direction of the specific virtual reference frame is determined, and the direction is determined as the moving direction.

[0121] In an implementation, the image parameter difference ratio is the average of the chrominance difference ratio and the luminance difference ratio.

[0122] The difference ratio is the ratio of the absolute value of the difference between two values ​​to the mean of the two values.

[0123] Specifically, in implementation, the shooting direction is the same as the moving direction, so the center of the inspection image is the current moving direction. Based on this, the direction corresponding to the center of the specific virtual reference frame relative to the moving inspection unit is determined as the direction of the virtual reference frame.

[0124] Determining the direction of coordinates in a captured image relative to the shooting position is an existing technology and is widely used in the field of drones, so it will not be described in detail here.

[0125] Specifically, the warning analysis module analyzes the changes in the basic features of the pixels in the sky segmentation area of ​​the inspection image at each moment to determine whether to issue a fire warning signal, including:

[0126] Used to identify unique pixels within the sky segmentation area;

[0127] To determine the coordinates of specific pixel points and obtain a coordinate set for each specific pixel point;

[0128] To compare the coordinate sets of corresponding specific pixel points in the inspection images at adjacent times and determine the degree of coincidence of the coordinate sets;

[0129] If the mean value of the coordinate set coincidence is less than or equal to the preset coordinate set coincidence threshold, it is determined that a fire warning signal needs to be issued;

[0130] If the grayscale value of a pixel does not fall within a predetermined grayscale range, the pixel is determined to be a unique pixel.

[0131] Specifically, the predetermined grayscale range is determined based on the inspection image that has been taken, and the average grayscale value of each pixel point in the sky area of ​​the inspection image that has been taken is determined. The grayscale range is a closed interval, the lower limit of the interval is 0.75 times the average grayscale value, and the upper limit of the interval is 1.25 times the average grayscale value.

[0132] Specifically, the calculation process of the coincidence degree of the coordinate set includes:

[0133] The number of coordinates that appear in both coordinate sets is calculated, and the ratio of the number to the total number of coordinates in the coordinate sets is determined as the coincidence degree of the coordinate sets.

[0134] Specifically, the coordinate set coincidence threshold is set to characterize the changes in specific pixel points in the inspection image at each moment. Smoke characteristics are highly dynamic, and the coincidence is usually low. In implementation, the coincidence threshold is selected within the range of [0.3.0.5].

[0135] Specifically, the early warning analysis module is used to control the mobile inspection unit to move in the corresponding moving direction and maintain the position to continuously obtain the inspection image in the moving direction, including:

[0136] It is used to control the mobile inspection unit to move to the corresponding moving direction and then hover to maintain the position and continuously obtain the inspection image of the predetermined length of time in the moving direction.

[0137] The scheduled duration is 10 seconds, and an inspection image is acquired once per second.

[0138] There is no limit on the moving distance, and the purpose is to get close to the features within the specific virtual reference frame, which will not be elaborated here.

[0139] The present invention performs early warning analysis in response to the calibration results of the inspection images. The calibrated inspection images have a high tendency to have abnormalities. Therefore, a virtual reference frame where the abnormalities are more obvious is determined in a targeted manner. After the moving direction is determined based on the virtual reference frame, the drone is controlled to approach for verification. Inspection images at multiple moments are captured in a hovering state. The properties of the dynamic diffusion of smoke and the relatively prominent properties of smoke features in the sky segmentation area are used. The overlap of the corresponding coordinate sets of specific pixel points is considered to verify whether the smoke is flowing and spreading, and then determine whether a fire warning signal is issued, thereby improving the reliability and accuracy of fire warnings for forests at night. The drone is controlled to approach for verification only for the marked inspection images, thereby improving the efficiency of inspecting large areas of forests.

[0140] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0141] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. Fire warning system based on artificial intelligence, characterized by: include: An image detection module, which includes a number of mobile inspection units for acquiring inspection images; a boundary analysis module connected to the image detection module, configured to identify the intersection boundaries of predetermined feature areas based on the inspection image, construct a plurality of virtual reference frames along the intersection boundaries, and analyze boundary difference characteristics based on image parameter differences of the segmented areas after each virtual reference frame is segmented by the intersection boundaries; a continuous analysis module connected to the boundary analysis module, configured to analyze the distribution of pixels in a strip around the intersection boundary of each virtual reference frame to determine boundary distribution characteristics; a calibration module, connected to the boundary analysis module and the continuous analysis module respectively, for calculating characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics to calibrate the corresponding inspection image; An early warning analysis module is connected to the calibration module and performs early warning analysis in response to the calibration results of the inspection image, including: A unique virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the unique virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. The corresponding coordinate set of unique pixel points in the sky segmentation area of ​​the inspection image obtained at each moment is used to determine whether to issue a fire warning signal; or, controlling the mobile inspection unit to move along a predetermined inspection route and continuously acquiring inspection images; The segmented area includes a sky segmented area and a non-sky segmented area; The early warning analysis module determines a specific virtual reference frame based on the inspection image, and determines the moving direction of the mobile inspection unit based on the specific virtual reference frame, including: Used to verify the image parameter difference ratio between the sky segmentation area within the virtual reference frame and the remaining sky segmentation areas; for determining the virtual reference frame corresponding to the maximum image parameter difference ratio as a specific virtual reference frame; The direction of the specific virtual reference frame is determined, and the direction is determined as the moving direction.

2. The fire warning system based on artificial intelligence according to claim 1 is characterized in that: The continuous analysis module is used to identify the intersection boundary of the predetermined feature area based on the inspection image, and construct a plurality of virtual reference frames along the intersection boundary, including: Used to identify predetermined feature areas, including sky areas and non-sky areas; Used to determine the boundary between the sky area and the non-sky area; for calibrating a plurality of reference points along the intersection boundary, and constructing a plurality of virtual reference frames with each of the reference points as the center; To determine the sky segmentation area and the non-sky segmentation area after the virtual reference frame is divided by the intersection boundary; The bottom edge of the virtual reference frame is parallel to the bottom edge of the inspection image.

3. The artificial intelligence-based fire warning system according to claim 2 is characterized in that: The continuous analysis module is used to analyze boundary difference features based on image parameter differences of each segmented area after each virtual reference frame is segmented by the intersection boundary, including: To solve the image parameter standard deviation between the sky segmentation areas in each virtual reference frame respectively; To respectively solve the standard deviation of image parameters between non-sky segmentation areas in each virtual reference frame; for performing weighted summation of the image parameter standard deviation for the sky segmentation area and the image parameter standard deviation for the non-sky segmentation area to obtain the boundary difference feature; The image parameter standard deviation is the mean of the brightness standard deviation and the chromaticity standard deviation.

4. The artificial intelligence-based fire warning system according to claim 3 is characterized in that: It is used to analyze the distribution of pixels in the strip area around the intersection boundary of each virtual reference frame and determine the boundary distribution characteristics including: The virtual intersection boundary is moved in the virtual reference frame by a predetermined distance in the direction of the sky segmentation area to obtain a virtual intersection boundary, so that the virtual intersection boundary and the intersection boundary form the strip area; Determine the characteristic pixel based on the grayscale difference ratio between the pixel and its adjacent pixels; Used to determine the density of feature similarity points as boundary distribution features; If the grayscale difference ratio corresponding to any adjacent pixel of a pixel is greater than a predetermined grayscale difference ratio threshold, the pixel is determined as a feature pixel.

5. The artificial intelligence-based fire warning system according to claim 4 is characterized in that: The calibration module is used to calculate the characteristic warning characterization parameters based on the boundary difference characteristics and the boundary distribution characteristics, including: The ratio of the boundary difference feature to the predetermined boundary difference feature threshold is used to calculate the boundary difference factor; The ratio of the boundary distribution feature to the predetermined boundary distribution feature threshold is used to calculate the boundary distribution factor; The boundary difference factor and the boundary distribution factor are weighted and summed to obtain the characteristic warning characterization parameter.

6. The artificial intelligence-based fire warning system according to claim 5, characterized in that: The calibration module calibrates the inspection image including: If the characteristic warning representation parameter is greater than or equal to a preset characteristic warning representation threshold, it is determined that the inspection image needs to be calibrated.

7. The artificial intelligence-based fire warning system according to claim 6, characterized in that: The early warning analysis module performs early warning analysis in response to the calibration results of the inspection image, wherein: If the inspection image is calibrated, a specific virtual reference frame is determined based on the inspection image, and the moving direction of the mobile inspection unit is determined based on the specific virtual reference frame to control the mobile inspection unit to move in the corresponding moving direction and maintain its position. Based on the coincidence degree of the corresponding coordinate set of the specific pixel points in the sky segmentation area in the inspection image obtained at each moment, it is determined whether to issue a fire warning signal; If the inspection image is not calibrated, the mobile inspection unit is controlled to move along a predetermined inspection route to continuously acquire inspection images.

8. The fire warning system based on artificial intelligence according to claim 1 is characterized in that: The warning analysis module analyzes the changes in the basic features of the pixels in the sky segmentation area of ​​the inspection image at each moment to determine whether to issue a fire warning signal, including: Used to identify unique pixels within the sky segmentation area; To determine the coordinates of specific pixel points and obtain a coordinate set for each specific pixel point; To compare the coordinate sets of corresponding specific pixel points in the inspection images at adjacent times and determine the degree of coincidence of the coordinate sets; If the mean value of the coordinate set coincidence is less than or equal to the preset coordinate set coincidence threshold, it is determined that a fire warning signal needs to be issued; If the grayscale value of a pixel does not fall within a predetermined grayscale range, the pixel is determined to be a unique pixel.

9. The fire warning system based on artificial intelligence according to claim 1 is characterized in that: The early warning analysis module is used to control the mobile inspection unit to move in the corresponding moving direction and maintain the position to continuously obtain the inspection image in the moving direction, including: It is used to control the mobile inspection unit to move to the corresponding moving direction and then hover to maintain the position and continuously obtain the inspection image of the predetermined length of time in the moving direction.

Images