An intelligent wildfire early warning method combining infrared imaging and smoke detection

Abstract

The present invention discloses an intelligent wildfire early warning method combining infrared imaging and smoke detection, comprising the following steps: S1: infrared video acquisition and adjustment, collecting infrared thermal imaging video of a target area by an infrared camera, the infrared camera being installed at a fixed position or on a mobile platform, and taking infrared images of the target area in real time; S2: smoke feature extraction, S3: multi-spectral fusion, S4: real-time monitoring and alarm, S5: environmental adaptability adjustment. The present invention realizes real-time monitoring of infrared images of a target area, and improves image quality by dynamic range adjustment and denoising processing, thereby more accurately extracting smoke features and enhancing the accuracy of wildfire detection.

Description

Technical Field

[0001] The present invention relates to the technical field of mountain fire monitoring, and in particular to an intelligent mountain fire early warning method combining infrared imaging and smoke detection. Background Art

[0002] In the field of intelligent wildfire warning, infrared thermal imaging technology is valued for its ability to work in adverse weather and nighttime conditions. However, existing technologies have some specific defects in practical applications. The following is a detailed analysis of the technical difficulties in infrared thermal imaging smoke detection technology based on motion and blur features:

[0003] Infrared thermal imaging videos usually have low contrast and poor texture, which requires smoke detection algorithms to be able to adapt to these conditions. Traditional Vibe detection algorithms use fixed thresholds, but it is difficult to balance smoke detection at different distances under different conditions, resulting in missed or false alarms;

[0004] Infrared detectors are greatly affected by the environment, such as detection angle, cloud thickness, vertical structure of the atmosphere, and terrain. These environmental and climatic factors will affect the detection results of satellite infrared remote sensing, causing false hot spots or failure to find real fire points, resulting in misjudgment and missed judgments;

[0005] In summary, infrared thermal imaging smoke detection technology based on motion and blur features has defects in environmental adaptability, anti-interference ability, balance between system response and detection ability, insufficient research on image enhancement algorithms, and performance evaluation methods in intelligent wildfire warning. Solving these problems is crucial to improving the accuracy and efficiency of wildfire warning systems. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent forest fire early warning method combining infrared imaging and smoke detection to solve the above problems.

[0007] The present invention is achieved through the following technical solutions:

[0008] An intelligent wildfire early warning method combining infrared imaging and smoke detection includes the following steps:

[0009] S1: Infrared video acquisition and adjustment, collecting infrared thermal imaging video of the target area through an infrared camera, the infrared camera is installed in a fixed position or on a mobile platform (such as a drone), and takes infrared images of the target area in real time;

[0010] Adjust the dynamic range of the collected infrared thermal imaging video to enhance the contrast of the image, and then perform denoising on the adjusted infrared image to reduce the impact of noise on infrared image detection;

[0011] S2: Smoke feature extraction, based on the improved ViBe algorithm to extract smoke features, the algorithm uses adaptive threshold and dynamic background model, separates the smoke area from the background through threshold segmentation method, and uses optical flow method to detect the movement direction and speed of the smoke;

[0012] S3: Multispectral fusion: Multispectral fusion of infrared images and visible light images is performed to enhance the accuracy of wildfire detection. The fusion method adopts weighted fusion based on image gradient;

[0013] S4: Real-time monitoring and alarm: The fused images are input into the pre-trained deep learning model for wildfire identification and alarm. The model uses a convolutional neural network (CNN) and is trained with a large-scale data set to ensure high accuracy and real-time performance.

[0014] S5: Environmental adaptability adjustment: Dynamically adjust algorithm parameters based on detected environmental parameters, including detection angle, cloud thickness, vertical structure of the atmosphere, etc., to improve the robustness of the system.

[0015] The present invention realizes real-time monitoring of infrared images of target areas and improves image quality through dynamic range adjustment and denoising processing, thereby more accurately extracting smoke features and enhancing the accuracy of wildfire detection.

[0016] Preferably, in S1, when the infrared camera collects infrared thermal imaging video of the target area, at least two infrared cameras are provided in the same target area, one of which is a dynamic following camera that mainly monitors the movement trajectory of animals in the target area, and the other is a fixed-position camera that mainly monitors the infrared spectrum of the target area.

[0017] The present invention sets at least two infrared cameras, a dynamic following camera and a fixed position camera, in the same target area. The beneficial effect of this claim is that it can simultaneously monitor the animal's movement trajectory and infrared spectrum, provide more comprehensive monitoring data, and reduce false alarms and missed alarms.

[0018] Preferably, the one infrared camera is a dynamic following camera system, which can automatically track and monitor the movement trajectory of animals in the target area, and it captures the infrared radiation emitted by animals in real time through an infrared sensor array, and determines the position and movement direction of the animals through algorithm processing to achieve automatic tracking;

[0019] The dynamic following camera system of the present invention can automatically track and monitor the movement trajectory of animals in the target area, and its beneficial effect is to improve the monitoring efficiency of animal behavior, reduce the burden of manual monitoring, and improve the tracking accuracy through algorithm processing.

[0020] Preferably, the animal movement trajectory data captured by the dynamic following camera is fused with the infrared spectrum data monitored by the fixed-position camera, and the occurrence of false hot spots is prevented by analyzing the relationship between animal behavior and ambient temperature changes.

[0021] The present invention fuses animal movement trajectory data with infrared spectrum data to analyze the relationship between animal behavior and ambient temperature changes. The beneficial effect is that it can more accurately distinguish between real fire conditions and false hotspot phenomena, thereby improving the reliability of the early warning system.

[0022] Preferably, the animal movement trajectory data captured by the dynamic following camera is fused with the infrared spectrum data monitored by the fixed-position camera, and a convolutional neural network (CNN) model is specifically used for target recognition and tracking, and in multi-target tracking, a data association algorithm is introduced to improve the tracking accuracy.

[0023] The present invention uses a convolutional neural network (CNN) model for target recognition and tracking, and introduces a data association algorithm, which has the beneficial effect of improving the accuracy of multi-target tracking and enhancing the real-time monitoring capability of the system.

[0024] Preferably, in S1, the dynamic range of the collected infrared thermal imaging video is adjusted, and the contrast of the image is enhanced using logarithmic transformation. The specific formula is:

[0025] I enhanced (x, y) = c·log(1+I original (x, y));

[0026] Among them, I original (x, y) is the pixel value of the original image;

[0027] I enhanced (x, y) is the enhanced image pixel value;

[0028] C constant, used to adjust the degree of enhancement;

[0029] In S1, the adjusted infrared image is denoised, and a bilateral filter is used to reduce the influence of noise on infrared image detection;

[0030] The formula for the bilateral filter is:

[0031]

[0032] Where I(x, y) is the image pixel value;

[0033] is the spatial Gaussian kernel;

[0034] is the range Gaussian kernel;

[0035] Ω is the filter window;

[0036] W(p) is the normalization coefficient.

[0037] The present invention uses logarithmic transformation to enhance image contrast and bilateral filter to reduce noise, which has the beneficial effect of improving the quality of infrared images and providing clearer image data for subsequent smoke feature extraction and wildfire identification.

[0038] Preferably, in S3, the infrared image and the visible light image are aligned, and a method based on feature point matching is used, and the feature points are extracted by a Super Rapid Robust Feature (SURF) algorithm, which specifically includes the following steps:

[0039] S3.1 Feature extraction: Extract SURF feature points from infrared images and visible light images;

[0040] S3.2 Feature matching: Match feature points using the nearest neighbor matching method;

[0041] S3.3 Geometric transformation: The geometric transformation matrix between the infrared image and the visible light image is estimated by the RANSAC algorithm.

[0042] The present invention aligns infrared images and visible light images through a method based on feature point matching, which has the beneficial effect of improving the accuracy of multi-spectral fusion and enhancing the accuracy and reliability of wildfire detection.

[0043] Preferably, in S5, the algorithm parameters are dynamically adjusted according to the detected environmental parameters to improve the robustness of the system. The specific adjustment method is as follows:

[0044] S5.1 Threshold adjustment: adjust the coefficient k in the adaptive threshold T(x, y) according to the environmental parameters;

[0045] S5.2 Filter parameters: Adjust the spatial Gaussian kernel σ of the bilateral filter according to the environmental parameters s and the range Gaussian kernel σ r ;

[0046] S5.3 Weight adjustment: Adjust the weights w1(x, y) and w2(x, y) in image fusion according to the environmental parameters.

[0047] The present invention dynamically adjusts algorithm parameters according to environmental parameters, which has the beneficial effect of improving the adaptability and robustness of the system and ensuring that high detection accuracy can be maintained under different environmental conditions.

[0048] Preferably, in S4, an attention mechanism is introduced in the fully connected layer of the convolutional neural network to improve the real-time performance of the model. Through the attention mechanism, the contribution of different feature maps is dynamically adjusted to improve the real-time performance of the model. While ensuring accuracy, a lightweight convolutional neural network structure is used to reduce the amount of calculation.

[0049] The present invention introduces an attention mechanism into the fully connected layer of the convolutional neural network, which has the beneficial effect of improving the real-time performance and accuracy of the model, while using a lightweight network structure to reduce the amount of calculation and improve the efficiency of the system.

[0050] Preferably, in S2, the background model is established by statistically analyzing the distribution of pixel values ​​of historical frames, and the specific formula is:

[0051] B(x,y,t)={I(x,y,t-1),I(x,y,t-2),...,I(x,y,tN)}

[0052] Where B(x, y, t) is the background model at position (x, y) at time t;

[0053] I(x, y, t) is the pixel value of the current frame;

[0054] N is the number of frames of the background model;

[0055] In S2, the smoke area is separated from the background by threshold segmentation. The adaptive threshold T(x, y) is calculated by the following formula:

[0056] T(x,y)=μ(x,y)+k·σ(x,y);

[0057] Among them, μ(x, y) is the mean of the background model;

[0058] σ(x, y) is the standard deviation of the background model;

[0059] k is the coefficient;

[0060] In S2, the optical flow method is used to detect the movement direction and speed of the smoke. The optical flow method calculates the motion vector of the pixel by solving the optical flow equation. The specific formula is:

[0061] Optical flow equation:

[0062] Where I is the image pixel value;

[0063] V is the motion vector;

[0064] is the gradient of the image.

[0065] The present invention establishes a background model by statistically analyzing the distribution of pixel values ​​in historical frames, and uses adaptive threshold and optical flow method to detect smoke. The beneficial effect of the present invention is that the accuracy of smoke feature extraction and the efficiency of smoke motion detection are improved, providing strong technical support for wildfire early warning.

[0066] Through the technical solution provided by the present invention, the present invention has at least the following technical effects:

[0067] The present invention significantly improves the image quality and enhances the accuracy of smoke feature extraction by real-time acquisition and processing of infrared images, combined with dynamic range adjustment and denoising technology. By setting at least two infrared cameras, simultaneous monitoring of animal movement trajectories and infrared spectra is achieved, reducing false alarms and missed alarms. The introduction of a dynamic following camera system and a data association algorithm improves the accuracy of target tracking and the real-time monitoring capability of the system. The application of logarithmic transformation and bilateral filters further improves the image quality, while the method based on feature point matching and the dynamic adjustment of environmental parameters enhance the adaptability and robustness of the system. In addition, the introduction of the attention mechanism in the convolutional neural network and the use of a lightweight network structure not only improve the real-time and accuracy of the model, but also reduce the amount of calculation and improve the overall efficiency of the system. The integration of these technologies provides an efficient, accurate and reliable solution for wildfire warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0069] Figure 1 is a flow chart of the steps of the present invention;

[0070] Figure 2 is a specific step flow chart of S3 of the present invention;

[0071] Figure 3 Schematic diagram of the types of cameras used for infrared video acquisition in the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the embodiments. The schematic implementation mode of the present invention and its description are only used to explain the present invention and are not intended to limit the present invention. The embodiments described below are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention.

[0073] In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is apparent to those of ordinary skill in the art that these specific details do not have to be employed to practice the present invention. In other embodiments, in order to avoid confusing the present invention, known structures, materials or methods are not specifically described. The materials, instruments and reagents used in the following examples, unless otherwise specified, can be obtained from commercial sources. The technical means used in the examples, unless otherwise specified, are conventional means well known to those skilled in the art.

[0074] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0075] Example

[0076] An intelligent wildfire early warning method combining infrared imaging and smoke detection includes the following steps:

[0077] S1: Infrared video acquisition and adjustment, collecting infrared thermal imaging video of the target area through an infrared camera, the infrared camera is installed in a fixed position or on a mobile platform (such as a drone), and takes infrared images of the target area in real time;

[0078] Adjust the dynamic range of the collected infrared thermal imaging video to enhance the contrast of the image, and then perform denoising on the adjusted infrared image to reduce the impact of noise on infrared image detection;

[0079] S2: Smoke feature extraction, based on the improved ViBe algorithm to extract smoke features, the algorithm uses adaptive threshold and dynamic background model, separates the smoke area from the background through threshold segmentation method, and uses optical flow method to detect the movement direction and speed of the smoke;

[0080] S3: Multispectral fusion: Multispectral fusion of infrared images and visible light images is performed to enhance the accuracy of wildfire detection. The fusion method adopts weighted fusion based on image gradient;

[0081] S4: Real-time monitoring and alarm: The fused images are input into the pre-trained deep learning model for wildfire identification and alarm. The model uses a convolutional neural network (CNN) and is trained with a large-scale data set to ensure high accuracy and real-time performance.

[0082] S5: Environmental adaptability adjustment: Dynamically adjust algorithm parameters based on detected environmental parameters, including detection angle, cloud thickness, vertical structure of the atmosphere, etc., to improve the robustness of the system.

[0083] The present invention realizes real-time monitoring of infrared images of target areas and improves image quality through dynamic range adjustment and denoising processing, thereby more accurately extracting smoke features and enhancing the accuracy of wildfire detection.

[0084] In this embodiment, in S1, when the infrared camera collects infrared thermal imaging video of the target area, at least two infrared cameras are provided in the same target area, one of which is a dynamic following camera, which mainly monitors the movement trajectory of animals in the target area, and the other is a fixed-position camera, which mainly monitors the infrared spectrum of the target area.

[0085] The present invention sets at least two infrared cameras, a dynamic following camera and a fixed position camera, in the same target area. The beneficial effect of this claim is that it can simultaneously monitor the animal's movement trajectory and infrared spectrum, provide more comprehensive monitoring data, and reduce false alarms and missed alarms.

[0086] In this embodiment, the one infrared camera is a dynamic following camera system, which can automatically track and monitor the movement trajectory of animals in the target area. It captures the infrared radiation emitted by animals in real time through an infrared sensor array, and determines the position and movement direction of the animals through algorithm processing to achieve automatic tracking;

[0087] The dynamic tracking camera system should be highly automated and intelligent, able to identify and distinguish different types of animal behaviors. The system should be able to adapt to different lighting conditions and weather changes to maintain tracking continuity and accuracy.

[0088] The dynamic following camera system of the present invention can automatically track and monitor the movement trajectory of animals in the target area, and its beneficial effect is to improve the monitoring efficiency of animal behavior, reduce the burden of manual monitoring, and improve the tracking accuracy through algorithm processing.

[0089] In this embodiment, the animal movement trajectory data captured by the dynamic following camera is fused with the infrared spectrum data monitored by the fixed position camera, and the relationship between the animal behavior and the ambient temperature change is analyzed to prevent the occurrence of false hot spots.

[0090] The present invention fuses animal movement trajectory data with infrared spectrum data to analyze the relationship between animal behavior and ambient temperature changes. The beneficial effect is that it can more accurately distinguish between real fire conditions and false hotspot phenomena, thereby improving the reliability of the early warning system.

[0091] In this embodiment, the animal movement trajectory data captured by the dynamic following camera is fused with the infrared spectrum data monitored by the fixed-position camera, and a convolutional neural network (CNN) model is used for target recognition and tracking. In multi-target tracking, a data association algorithm is introduced to improve the tracking accuracy.

[0092] The present invention uses a convolutional neural network (CNN) model for target recognition and tracking, and introduces a data association algorithm, which has the beneficial effect of improving the accuracy of multi-target tracking and enhancing the real-time monitoring capability of the system.

[0093] In this embodiment, in S1, the dynamic range of the collected infrared thermal imaging video is adjusted, and the contrast of the image is enhanced using logarithmic transformation. The specific formula is:

[0094] I enhanced (x, y) = c·log(1+I original (x, y));

[0095] Among them, I original (x, y) is the pixel value of the original image;

[0096] I enhanced (x, y) is the enhanced image pixel value;

[0097] C constant, used to adjust the degree of enhancement;

[0098] In S1, the adjusted infrared image is denoised, and a bilateral filter is used to reduce the influence of noise on infrared image detection;

[0099] The formula for the bilateral filter is:

[0100]

[0101] Where I(x, y) is the image pixel value;

[0102] is the spatial Gaussian kernel;

[0103] is the range Gaussian kernel;

[0104] Ω is the filter window;

[0105] W(p) is the normalization coefficient.

[0106] The present invention uses logarithmic transformation to enhance image contrast and bilateral filter to reduce noise, which has the beneficial effect of improving the quality of infrared images and providing clearer image data for subsequent smoke feature extraction and wildfire identification.

[0107] In this embodiment, in S3, the infrared image and the visible light image are aligned, and a method based on feature point matching is used. The feature points are extracted by a Super Rapid Robust Feature (SURF) algorithm, which specifically includes the following steps:

[0108] S3.1 Feature extraction: Extract SURF feature points from infrared images and visible light images;

[0109] S3.2 Feature matching: Match feature points using the nearest neighbor matching method;

[0110] S3.3 Geometric transformation: The geometric transformation matrix between the infrared image and the visible light image is estimated by the RANSAC algorithm.

[0111] The present invention aligns infrared images and visible light images through a method based on feature point matching, which has the beneficial effect of improving the accuracy of multi-spectral fusion and enhancing the accuracy and reliability of wildfire detection.

[0112] In this embodiment, in S5, the algorithm parameters are dynamically adjusted according to the detected environmental parameters to improve the robustness of the system. The specific adjustment method is as follows:

[0113] S5.1 Threshold adjustment: adjust the coefficient k in the adaptive threshold T(x, y) according to the environmental parameters;

[0114] S5.2 Filter parameters: Adjust the spatial Gaussian kernel σ of the bilateral filter according to the environmental parameters s and the range Gaussian kernel σ r ;

[0115] S5.3 Weight adjustment: Adjust the weights w1(x, y) and w2(x, y) in image fusion according to the environmental parameters.

[0116] The present invention dynamically adjusts algorithm parameters according to environmental parameters, which has the beneficial effect of improving the adaptability and robustness of the system and ensuring that high detection accuracy can be maintained under different environmental conditions.

[0117] In this embodiment, in S4, an attention mechanism is introduced in the fully connected layer of the convolutional neural network to improve the real-time performance of the model. Through the attention mechanism, the contribution of different feature maps is dynamically adjusted to improve the real-time performance of the model. While ensuring accuracy, a lightweight convolutional neural network structure is used to reduce the amount of calculation.

[0118] In this embodiment, the SE module implements the relationship modeling between channels through global average pooling (GAP) and a fully connected layer. The specific steps are as follows:

[0119] Global Average Pooling (GAP): compresses each feature map into a scalar value;

[0120] Fully connected layer: The compressed features are processed through two fully connected layers. The first layer reduces the dimension and the second layer restores the original dimension. The expression is as follows:

[0121] s=σ(g(z,W))=σ(W2δ(W1z))

[0122] Among them, g represents the fully connected operation, δ is the ReLU activation function, σ is the Sigmoid activation function, W1 and W2 are the weights of the fully connected layer;

[0123] Recalibration: Use the obtained channel weights to weight the original feature map;

[0124]

[0125] in, Represents element-wise multiplication.

[0126] In this embodiment, the ECA module provides a more efficient channel attention mechanism, which uses one-dimensional convolution to replace the fully connected layer in the SE module, greatly reducing the amount of computation.

[0127] It includes adaptive kernel size, which automatically selects the kernel size of one-dimensional convolution according to the number of channels. The expression is as follows;

[0128]

[0129] Among them, k is the adaptive kernel size and c is the number of channels.

[0130] And through one-dimensional convolution, operations are directly performed on the original channels to avoid information loss;

[0131] y=σ(conv(y))

[0132] Among them, conv is a one-dimensional convolution operation.

[0133] Finally, local cross-channel interactions are used to capture the dependencies between local channels through one-dimensional convolution.

[0134] In this embodiment, in order to reduce the amount of calculation, a lightweight CNN structure can be used, for example, using depthwise separable convolution instead of standard convolution. This structure first applies depthwise convolution and then pointwise convolution.

[0135] Specifically, they include depthwise convolution and pointwise convolution, where depthwise convolution applies convolution independently to each input channel;

[0136] Point-wise convolution is to apply 1x1 convolution on the result of depth convolution and merge channels;

[0137] Through the combination of depthwise convolution and point-by-point convolution, it can reduce the amount of computation and the number of parameters while maintaining the performance of convolutional neural networks.

[0138] The present invention introduces an attention mechanism into the fully connected layer of the convolutional neural network, which has the beneficial effect of improving the real-time performance and accuracy of the model, while using a lightweight network structure to reduce the amount of calculation and improve the efficiency of the system.

[0139] In this embodiment, in S2, the background model is established by statistically analyzing the distribution of pixel values ​​of historical frames. The specific formula is:

[0140] B(x,y,t)={I(x,y,t-1),I(x,y,t-2),...,I(x,y,tN)}

[0141] Where B(x, y, t) is the background model at position (x, y) at time t;

[0142] I(x, y, t) is the pixel value of the current frame;

[0143] N is the number of frames of the background model;

[0144] In S2, the smoke area is separated from the background by threshold segmentation. The adaptive threshold T(x, y) is calculated by the following formula:

[0145] T(x,y)=μ(x,y)+k·σ(x,y);

[0146] Among them, μ(x, y) is the mean of the background model;

[0147] σ(x, y) is the standard deviation of the background model;

[0148] k is the coefficient;

[0149] In S2, the optical flow method is used to detect the movement direction and speed of the smoke. The optical flow method calculates the motion vector of the pixel by solving the optical flow equation. The specific formula is:

[0150] Optical flow equation:

[0151] Where I is the image pixel value;

[0152] V is the motion vector;

[0153] is the gradient of the image.

[0154] The present invention establishes a background model by statistically analyzing the distribution of pixel values ​​in historical frames, and uses adaptive threshold and optical flow method to detect smoke. The beneficial effect of the present invention is that the accuracy of smoke feature extraction and the efficiency of smoke motion detection are improved, providing strong technical support for wildfire early warning.

[0155] The establishment of the background model takes into account long-term environmental changes, such as seasonal changes and changes in lighting conditions. The calculation of the adaptive threshold of the present invention should be based on the statistical characteristics of the background model to ensure effective separation of smoke areas.

[0156] It is understood that in this embodiment, dynamic range adjustment is an image processing technology used to improve the visual effect of an image, especially in high-contrast scenes. It enhances dark details and compresses bright details by stretching the pixel value distribution of the image, making details that were originally difficult to discern under extreme lighting conditions visible. In infrared images, since the sensor is sensitive to temperature changes, dynamic range adjustment is particularly important because it can help distinguish objects of different temperatures and improve the accuracy of subsequent analysis;

[0157] In this embodiment, denoising is an important step to reduce random noise in the image, which may come from electronic noise of the sensor, environmental interference or image compression. In infrared images, denoising is particularly important because noise may mask important temperature differences, thereby affecting the detection of smoke and other thermal features. The bilateral filter is able to reduce noise while keeping edges sharp because it takes into account the spatial proximity of pixel values ​​and the similarity of pixel values.

[0158] Optical flow is a method for calculating the motion information of objects in image sequences. It estimates the direction and speed of an object's motion by tracking feature points in consecutive frames.

[0159] The ViBe (Visual Background Extractor) algorithm is an algorithm for background subtraction, which distinguishes the foreground from the background by maintaining a background model. The ViBe algorithm is parameter-free and does not require a pre-set threshold.

[0160] RANSAC (Random Sample Consensus) is an iterative method for estimating mathematical model parameters from a set of data containing outliers. In computer vision, RANSAC is often used to estimate geometric transformations, such as translation, rotation, or affine transformation between images, from a set of point pairs that may contain incorrect matches. RANSAC fits the model by randomly selecting a subset of data and testing the applicability of this model to the entire data set, thereby effectively eliminating the influence of outliers.

[0161] The optional implementation modes of the embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation modes. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical scheme of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention.

[0162] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0163] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor (processor) execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0164] In addition, various implementation modes of the embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed by the embodiments of the present invention.

Claims

1. An intelligent forest fire early warning method combining infrared imaging and smoke detection, characterized in that: The steps include: S1: Infrared video acquisition and adjustment, collecting infrared thermal imaging video of the target area through an infrared camera, the infrared camera is installed in a fixed position or on a mobile platform, and captures the infrared image of the target area in real time; Adjust the dynamic range of the collected infrared thermal imaging video to enhance the contrast of the image, and then perform denoising on the adjusted infrared image to reduce the impact of noise on infrared image detection; S2: Smoke feature extraction, based on the improved ViBe algorithm to extract smoke features, the algorithm uses adaptive threshold and dynamic background model, separates the smoke area from the background through threshold segmentation method, and uses optical flow method to detect the movement direction and speed of the smoke; S3: Multispectral fusion: Multispectral fusion of infrared images and visible light images is performed to enhance the accuracy of wildfire detection. The fusion method adopts weighted fusion based on image gradient; S4: Real-time monitoring and alarm: The fused images are input into a pre-trained deep learning model for wildfire identification and alarm. The model uses a convolutional neural network and is trained with a large-scale data set to ensure high accuracy and real-time performance. S5: Environmental adaptability adjustment: Dynamically adjust algorithm parameters based on detected environmental parameters, including detection angle, cloud thickness, and vertical structure of the atmosphere to improve the robustness of the system.

2. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S1, when the infrared camera collects infrared thermal imaging video of the target area, at least two infrared cameras are arranged in the same target area, one of which is a dynamic following camera, which mainly monitors the movement trajectory of animals in the target area, and the other is a fixed-position camera, which mainly monitors the infrared spectrum of the target area.

3. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 2 is characterized in that: The infrared camera is a dynamic following camera system, which can automatically track and monitor the movement trajectory of animals in the target area. It captures the infrared radiation emitted by animals in real time through an infrared sensor array, and determines the position and movement direction of the animals through algorithm processing to achieve automatic tracking.

4. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 3 is characterized in that: The animal movement trajectory data captured by the dynamic following camera is integrated with the infrared spectrum data monitored by the fixed-position camera. By analyzing the relationship between animal behavior and ambient temperature changes, the occurrence of false hot spots can be prevented.

5. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 4 is characterized in that: The animal movement trajectory data captured by the dynamic following camera is fused with the infrared spectrum data monitored by the fixed-position camera. Specifically, a convolutional neural network model is used for target recognition and tracking. In multi-target tracking, a data association algorithm is introduced to improve the tracking accuracy.

6. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S1, the dynamic range of the acquired infrared thermal imaging video is adjusted, and the contrast of the image is enhanced using logarithmic transformation. The specific formula is: I enhanced (x,y)=c·log(1+I original (x,y)); Among them, I original (x, y) is the pixel value of the original image; I enhanced (x, y) is the enhanced image pixel value; C constant, used to adjust the degree of enhancement; In S1, the adjusted infrared image is denoised, and a bilateral filter is used to reduce the influence of noise on infrared image detection; The formula for the bilateral filter is: Where I(x, y) is the image pixel value; is the spatial Gaussian kernel; is the range Gaussian kernel; Ω is the filter window; W(p) is the normalization coefficient.

7. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S3, the infrared image is aligned with the visible light image using a feature point matching method. The feature points are extracted using a fast and robust feature algorithm, which specifically includes the following steps: S3.1 Feature extraction: Extract SURF feature points from infrared images and visible light images; S3.2 Feature matching: Match feature points using the nearest neighbor matching method; S3.3 Geometric transformation: The geometric transformation matrix between the infrared image and the visible light image is estimated by the RANSAC algorithm.

8. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S5, the algorithm parameters are dynamically adjusted according to the detected environmental parameters to improve the robustness of the system. The specific adjustment method is as follows: S5.1 Threshold adjustment: adjust the coefficient k in the adaptive threshold T(x, y) according to the environmental parameters; S5.2 Filter parameters: Adjust the spatial Gaussian kernel σ of the bilateral filter according to the environmental parameters s and the range Gaussian kernel σ r ; S5.3 Weight adjustment: Adjust the weights w1(x, y) and w2(x, y) in image fusion according to the environmental parameters.

9. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S4, an attention mechanism is introduced in the fully connected layer of the convolutional neural network to improve the real-time performance of the model. Through the attention mechanism, the contribution of different feature maps is dynamically adjusted to improve the real-time performance of the model. While ensuring accuracy, a lightweight convolutional neural network structure is used to reduce the amount of calculation.

10. The intelligent forest fire early warning method combining infrared imaging and smoke detection according to claim 1, characterized in that: In S2, the background model is established by statistically analyzing the distribution of pixel values ​​in historical frames. The specific formula is: B(x,y,t)={I(x,y,t-1),I(x,y,t-2),...,I(x,y,tN)} Where B(x, y, t) is the background model at position (x, y) at time t; I(x, y, t) is the pixel value of the current frame; N is the number of frames of the background model; In S2, the smoke area is separated from the background by threshold segmentation method, and the adaptive threshold T(x, y) is calculated by the following formula; T(x,y)=μ(x,y)+k·σ(x,y); Among them, μ(x, y) is the mean of the background model; σ(x, y) is the standard deviation of the background model; k is the coefficient; In S2, the optical flow method is used to detect the movement direction and speed of the smoke. The optical flow method calculates the motion vector of the pixel by solving the optical flow equation. The specific formula is: Optical flow equation: Where I is the image pixel value; V is the motion vector; is the gradient of the image.

Images