Fire identification method, fire identification device, electronic equipment and readable storage medium

Abstract

The invention discloses a fire recognition method, a fire recognition device, electronic equipment and a readable storage medium. The method comprises the following steps: acquiring image data, temperature sensing data and audio data corresponding to a to-be-recognized area; extracting image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data and audio features corresponding to the audio data; and inputting the image features, the temperature sensing features and the audio features into a fire recognition model to obtain a fire recognition result corresponding to the to-be-recognized area, the fire recognition model being constructed based on a random forest algorithm.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and in particular, relates to a fire identification method, a fire identification device, an electronic device and a readable storage medium. Background Art

[0002] As a natural disaster with high suddenness and great destructive power, forest fires face great challenges in rescue and disposal. In complex natural environments, traditional manual monitoring and fire-fighting methods often fail to detect and control fires in a timely and effective manner due to factors such as inconvenience in movement and slow reaction time, resulting in low fire-fighting efficiency. With the advancement of technology, sensors carried by drones and aircraft have been widely used in the early detection and monitoring of forest fires, showing the advantages of high efficiency and speed. However, the relevant fire source positioning technology mainly relies on visual and optical information, ignoring other possible sources of information, resulting in low accuracy in fire source identification. Summary of the invention

[0003] The embodiments of the present application provide a fire identification method, a fire identification device, an electronic device and a readable storage medium, which can solve the problem that the related art only relies on visual and optical information to identify the fire source, resulting in low identification accuracy.

[0004] In a first aspect, an embodiment of the present application provides a fire identification method, the method comprising: acquiring image data, temperature sensing data and audio data corresponding to an area to be identified; extracting image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data and audio features corresponding to the audio data; inputting the image features, the temperature sensing features and the audio features into a fire identification model to obtain a fire identification result corresponding to the area to be identified, wherein the fire identification model is constructed based on a random forest algorithm.

[0005] In a second aspect, an embodiment of the present application provides a fire identification device, which includes: an acquisition module, used to acquire image data, temperature sensing data and audio data corresponding to an area to be identified; an extraction module, used to extract image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data, and audio features corresponding to the audio data; an identification module, which inputs the image features, the temperature sensing features and the audio features into a fire identification model to obtain a fire identification result corresponding to the area to be identified, wherein the fire identification model is constructed based on a random forest algorithm.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0008] In a fifth aspect, an embodiment of the present application provides a chip, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method described in the first aspect.

[0009] In a sixth aspect, a computer program product is provided, the computer program product comprising at least one computer program, and when the computer program is loaded and executed by a processor, the method as described in the first aspect is implemented.

[0010] In an embodiment of the present application, by acquiring image data, temperature sensing data and audio data corresponding to the area to be identified, then extracting image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data and audio features corresponding to the audio data, and finally inputting the image features, temperature sensing features and audio features into a fire recognition model, a fire recognition result corresponding to the area to be identified is obtained, wherein the fire recognition model is constructed based on a random forest algorithm, which realizes fire identification, fire source location and fire monitoring in a complex environment, can avoid being affected by complex environments, such as dense fog and heavy smoke, and effectively improves the fire recognition rate and detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of a fire identification method provided in an embodiment of the present application;

[0012] Figure 2 It is a schematic diagram of a noise filtering simulation result based on wavelet transform and Kalman filtering provided in an embodiment of the present application;

[0013] Figure 3 It is a schematic diagram of extracting texture features of a thermal image based on a gray-level co-occurrence matrix provided in an embodiment of the present application;

[0014] Figure 4 It is a schematic diagram of extracting edge features of a thermal imaging image based on an edge feature algorithm provided in an embodiment of the present application;

[0015] Figure 5 This is a schematic diagram of a fire identification based on random forest provided in an embodiment of the present application

[0016] Figure 6 is another flow chart of a fire identification method provided in an embodiment of the present application;

[0017] Figure 7 It is a structural schematic diagram of a fire identification device provided in an embodiment of the present application;

[0018] Figure 8 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] In conjunction with the accompanying drawings, a fire identification method, a fire identification device, an electronic device and a readable storage medium provided by the embodiments of the present application are described in detail below through specific embodiments and their application scenarios.

[0021] Figure 1 A fire identification method provided by an embodiment of the present application is shown. The method can be executed by an electronic device, and the electronic device may include: a server and / or a terminal device. In other words, the method can be executed by software or hardware installed in the electronic device, and the method includes the following steps:

[0022] S110: Acquire image data, temperature data, and audio data corresponding to the area to be identified.

[0023] It is understandable that the related technology only relies on visual and optical information to identify the fire source. However, under special weather conditions such as fog and smoke, its recognition effect and accuracy are significantly limited. At the same time, the related technology also ignores the sound information of the fire scene, such as sound data such as popping sounds and material burning sounds, and cannot identify fires through diversified data sources, resulting in low recognition accuracy. In the embodiment of the present application, the data sources obtained include image data, temperature sensing data and audio data. Among them, fire recognition based on image data can perform fast and accurate visual analysis of the area to be identified. Fire recognition based on temperature sensing data can directly use the thermal radiation of the fire source or the ambient temperature for detection, thereby improving the recognition accuracy in the case of poor visibility or dense smoke. Fire recognition based on audio data can accurately capture the sound of the area to be identified, so that it can be further determined whether the sound of the area to be identified contains popping sounds and burning sounds related to the fire.

[0024] Among them, temperature sensing data refers to temperature information obtained through various sensors or thermal cameras, which may include but are not limited to thermal imaging images, thermal imaging videos, temperature sensor data, heat flux density data, ambient temperature data, etc. The image data can be a static image or a dynamic video.

[0025] Optionally, the audio data can be captured by a microphone array mounted on a drone, which is low-cost and has the characteristics of high precision and fast processing speed, making the capture and analysis of sound data more efficient.

[0026] S120: Extracting image features corresponding to the image data, temperature features corresponding to the temperature data, and audio features corresponding to the audio data.

[0027] It is understandable that image and audio data are usually high-dimensional data, and directly processing these raw data will result in excessive computational burden. Through feature extraction, the data can be converted into smaller and more easily processed feature representations while retaining important information, which will help improve the processing efficiency of the fire recognition model in the future.

[0028] The audio feature is an acoustic feature for identifying fire burning and explosion, and the image feature and temperature feature can be a texture feature and a color feature for identifying fire.

[0029] S130: Input the image features, the temperature features, and the audio features into a fire recognition model to obtain a fire recognition result corresponding to the area to be recognized.

[0030] Wherein, the fire identification model is constructed based on the random forest algorithm.

[0031] It can be understood that the input of the fire recognition model is the image features, the temperature features and the audio features, and the output is the fire recognition result, that is, whether a fire occurs in the area to be recognized is recognized by using a variety of different types of features, thereby improving the recognition accuracy.

[0032] The fire recognition model is constructed based on the random forest algorithm. The random forest algorithm improves the overall recognition performance by combining the prediction results of each tree, that is, each tree gives a category prediction for one observation. The recognition result can be expressed as:

[0033] y=mode{y1,y2,…,y t}

[0034] Among them, y i is the recognition result of i trees, and t is the total number of trees.

[0035] That is, random forests improve the accuracy and stability of recognition by integrating multiple decision trees. Since each decision tree is trained based on a different sample subset, the fire recognition model is able to learn the characteristics of different types of data from multiple dimensions, thereby reducing the risk of overfitting. In addition, random forests are highly robust to outliers and noisy data, and the errors of a single tree can be corrected in other trees in the forest, further improving the accuracy of recognition.

[0036] In an embodiment of the present application, by acquiring image data, temperature sensing data and audio data corresponding to the area to be identified, then extracting image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data and audio features corresponding to the audio data, and finally inputting the image features, temperature sensing features and audio features into a fire recognition model, a fire recognition result corresponding to the area to be identified is obtained, wherein the fire recognition model is constructed based on a random forest algorithm, which realizes fire identification, fire source location and fire monitoring in a complex environment, can avoid being affected by complex environments, such as dense fog and heavy smoke, and effectively improves the fire recognition rate and detection accuracy.

[0037] In one implementation, extracting audio features corresponding to the audio data includes: performing noise reduction processing on the audio data to obtain the noise-reduced audio data; and performing feature extraction on the noise-reduced audio data using Mel-frequency cepstral coefficients to obtain the audio features.

[0038] It is understandable that noise reduction processing can remove irrelevant interference sounds in audio data, which helps to improve the accuracy of feature extraction, so that the features extracted from the audio can more accurately reflect useful signal features instead of being masked by noise. At the same time, noise will affect the performance of the feature extraction algorithm, so that the extracted features contain a lot of useless noise information, thereby affecting the accuracy of the fire identification model. After noise processing, the Mel-frequency cepstral coefficients can be used to extract features from the denoised audio data to obtain the audio features. Extracting audio features through Mel-frequency cepstral coefficients can effectively convert high-dimensional original audio data into low-dimensional feature representations, effectively reducing the dimension of the features while retaining the main information of the audio signal.

[0039] The method of extracting features of the denoised audio data using Mel-frequency cepstral coefficients to obtain the audio features may include the following steps:

[0040] Step 1: Perform pre-emphasis processing on the audio data.

[0041] It can be expressed as:

[0042]

[0043] The value range of α can be 0.95 to 0.97.

[0044] Step 2: Divide the pre-emphasized audio data s(n) into frames of fixed length and apply a window function to each frame.

[0045] It can be expressed as:

[0046]

[0047] Step 3: Window each frame Perform fast Fourier transform to convert to frequency domain.

[0048] It can be expressed as:

[0049]

[0050] Step 4: Apply the Mel filter bank to filter the result of FFT, and the center frequency of each filter is distributed according to the Mel scale.

[0051] It can be expressed as:

[0052]

[0053] Step 5: Take the logarithm of the energy of each filter to simulate the logarithmic perception of sound intensity by the human ear.

[0054] It can be expressed as:

[0055] L j =log(E j )

[0056] Step 6: Perform discrete cosine transform on the logarithmic values ​​of all Mel filter energies and extract the cepstral coefficients.

[0057] It can be expressed as:

[0058]

[0059] Among them, α i is the i-th cepstral coefficient. Optionally, the embodiment of the present application may select the first 13 coefficients as the extracted audio features.

[0060] In one implementation, the denoising process of the audio data to obtain the denoised audio data includes: performing a wavelet transform on each audio frame of the audio data to obtain a time-frequency local feature corresponding to each audio frame, wherein the time-frequency local feature is used to represent the local features of the audio frame in the time dimension and the frequency dimension; and performing an inverse wavelet transform on each of the time-frequency local features to obtain the denoised audio data.

[0061] It is understandable that the noise reduction processing of audio data in the embodiment of the present application is implemented by wavelet transform. By using wavelet transform decomposition, the local features of audio data at different times and frequencies can be obtained, and the local features can be expressed as:

[0062]

[0063] Among them, c k and d jk are the approximation coefficient and detail coefficient, φ jk and ψ jk are the discrete forms of scaling function and wavelet function respectively. That is to say, when performing wavelet transform, the audio data is decomposed into local features of different time and frequency, which are represented by detail coefficients and approximation coefficients.

[0064] Among them, in another implementation, before obtaining the denoised audio data by performing inverse wave transform on each of the time-frequency local features, the method also includes: enhancing the time-frequency local features corresponding to each audio frame through a Kalman filter.

[0065] It is understandable that in the application of fire detection, the detail coefficient d that reflects the fire characteristics is usually concerned. jk Therefore, the Kalman filter algorithm is used to filter out the noise. Kalman filtering is a method of state estimation for dynamic systems, which estimates the internal state of the process through a series of measurements, including statistical noise and other uncertainties. In the application of combining wavelet transform and Kalman filtering, Kalman filtering is mainly used to optimize the detail coefficient d obtained by wavelet transform. jk These detail coefficients reflect the high-frequency part of the signal at each decomposition level, and are usually associated with transient information in the signal (such as the explosion sound or cracking wood at a fire scene). By optimizing these coefficients, these important features of the signal can be restored more accurately while filtering out noise and interference. For example, Figure 2 The figure shows the simulation results of noise filtering based on wavelet transform and Kalman filter. Each step of Kalman filter can be used to optimize each level of detail coefficient d obtained after wavelet transform. jk In its prediction stage, the state prediction and covariance prediction can be expressed as:

[0066]

[0067] Among them, F jk Represents the state transition matrix, used to transition from the previous state Predict current state Represents the estimated value of the detail coefficient at decomposition level j at the current time k; represents the prior estimate of the detail coefficient predicted by the state transition model; P jk represents the estimation error covariance matrix, which represents the uncertainty of the current detail coefficient estimate; represents the predicted error covariance matrix, which is used to evaluate the prediction uncertainty of the detail coefficients before the observation is updated; Q jk The process noise covariance matrix describes the uncertainty of state prediction caused by model imperfections and other external factors. In the update phase, the Kalman gain calculation, state update and covariance update are expressed as:

[0068]

[0069] Among them, H jk Represents the observation matrix, which maps the state space to the observation space and is used to convert the state estimate into a matrix that can be compared with the actual observation value y jk Direct comparison form; y jk The actual observation value represents the actual measurement of the detail coefficient at decomposition level j at time k; R jk The observation noise covariance matrix represents the uncertainty in the observation process; K jk Kalman gain, used to calculate the jk Update state estimate Weighing uncertainty in predictions and observations.

[0070] After wavelet decomposition or Kalman filtering, the reconstruction of the audio signal is achieved by inverse wavelet transform, which re-synthesizes multiple time-frequency local features into a signal in the time domain, expressed as:

[0071]

[0072] in, Represents the filtered audio data or audio signal, d j ' k are the approximation and detail coefficients adjusted by the Kalman filter.

[0073] In one implementation, extracting image features corresponding to the image data includes: using a convolutional neural network to extract features from the image data to obtain initial image features; using a local binary pattern to binarize a local neighborhood of each pixel in the image data to obtain a first texture feature of the image data; and performing feature fusion on the initial image features and the first texture features to obtain the image features.

[0074] The process of extracting features from the image data using a convolutional neural network to obtain initial image features may include the following steps:

[0075] Step 1: Use bilinear interpolation, where the difference is to estimate the value of the unknown pixel point through the value of the known pixel point.

[0076] It can be expressed as:

[0077] I resize =resize(I,width=224,height=224)

[0078] Step 2: The image is normalized.

[0079] It can be expressed as:

[0080]

[0081] Step 3: Set up CNN conv Represents the function from image to feature map, the extracted features F.

[0082] It can be expressed as: F = CNN conv (I norm )

[0083] Step 4: Use global average pooling to reduce feature dimensions.

[0084] It can be expressed as:

[0085] F pool =GAP(F)

[0086] Among them, F pool It is the initial image feature extracted using the convolutional neural network.

[0087] Using a local binary pattern to perform binarization processing on a local neighborhood of each pixel in the image data to obtain a first texture feature of the image data may include the following steps:

[0088] Step 1: For each pixel p at position (x, y), calculate the LBP value.

[0089] It can be expressed as:

[0090]

[0091] Among them, g c is the gray value of the central pixel, gp is the gray value of the neighboring pixel, and the function s(x) is a sign function, which can be defined as:

[0092]

[0093] Step 2: Construct a histogram H of LBP features and calculate the frequency of each LBP value.

[0094] It can be expressed as:

[0095]

[0096] After obtaining the initial image feature and the first texture feature, the initial image feature and the first texture feature can be fused to obtain the image feature: final =[F pool ; H].

[0097] In one implementation, the temperature sensing data includes a thermal imaging image, and the temperature sensing features include thermal imaging features; extracting the temperature sensing features corresponding to the temperature sensing data includes: performing feature extraction on the thermal imaging image based on a grayscale co-occurrence matrix to obtain a second texture feature corresponding to the thermal imaging image, wherein the second texture feature includes at least one of the following: contrast, homogeneity, angular second moment, entropy; performing feature extraction on the thermal imaging image based on an edge detection algorithm to obtain an edge feature corresponding to the thermal imaging image; performing feature fusion on the second texture feature and the edge feature to obtain the image feature.

[0098] It is understandable that thermal imaging technology is not limited by light conditions and can directly use the thermal radiation of the fire source for detection at night or in low visibility conditions. Therefore, the temperature sensing data in the embodiment of the present application may include a thermal imaging image. Figure 3 As shown, the second texture feature of the thermal image is extracted based on the gray level co-occurrence matrix (GLCM), as shown in Figure 4 As shown, the edge features of the thermal imaging image are extracted based on the edge feature algorithm.

[0099] Among them, for the thermal image I, the gray level co-occurrence matrix P(i,j|d,θ) is defined as the probability that a pixel with gray value i in the image co-occurs with a pixel with gray value j at distance d and direction θ. Common θ directions include 0° (horizontal), 45°, 90° (vertical) and 135°. The embodiment of the present application can extract features such as contrast, homogeneity, angular second-order moment and entropy of the thermal image based on GLCM, which can be expressed as:

[0100] Contrast ratio:

[0101] Among them, contrast reflects the clarity of the image and the depth changes of the texture.

[0102] Homogeneity:

[0103] Among them, homogeneity measures the smoothness of image texture.

[0104] Angular second moment:

[0105] Among them, the second-order angular moment reflects the uniformity and repeatability of image texture.

[0106] entropy:

[0107] Among them, entropy measures the complexity of image texture.

[0108] In addition, the embodiment of the present application uses an edge detection algorithm to extract edge features of the thermal imaging image. The edge detection algorithm may include two matrices, one for the horizontal direction G x The other one is used for edge detection in the vertical direction G y :

[0109]

[0110] For each pixel position (x, y), calculate the Sobel operator G x The response of Gy and Gy is used to obtain the edge strength of each pixel position:

[0111]

[0112] In one implementation, before the image features, the temperature features, and the audio features are input into a fire recognition model to obtain a fire recognition result corresponding to the area to be recognized, the method further includes: generating multiple sample subsets from a target training data set by guided sampling, wherein each of the sample subsets includes a first number of training samples; constructing a corresponding decision tree for each of the sample subsets; and constructing the fire recognition model based on the multiple decision trees.

[0113] Each training sample includes image training data, temperature sensing training data and audio training data.

[0114] It is understandable that if Figure 5 As shown, in order to improve the combination ability of multi-sensor parameters and the generalization ability of the model, the fire recognition model in the embodiment of the present application is constructed based on the random forest algorithm. The random forest consists of multiple decision trees, each of which is constructed by randomly extracting samples from the original data set through bootstrap sampling. The construction process of each tree is to select a sample subset D of size N from the target training data set by sampling with replacement. i To train the i-th tree, where N and i are both integers greater than 0.

[0115] Furthermore, in another implementation, the constructing of the corresponding decision tree includes: in each node splitting process, determining the optimal splitting point by performing feature selection and splitting point evaluation on each node, and dividing the sample subset into a first subset and a second subset; recursively repeating the above splitting process for the first subset and the second subset until a stopping condition is met; wherein the stopping condition includes one of the following: the depth of the decision tree is greater than or equal to a first threshold; the number of training samples in the first subset or the number of training samples in the second subset is less than a second threshold.

[0116] It can be understood that through feature extraction, a total feature set corresponding to the sample subset can be obtained, and the number of features included in the total feature set is M. In each splitting process, m features are randomly selected, where m is usually much smaller than the total number of features M. For each node, the optimal split point based on entropy is calculated using the selected m features:

[0117]

[0118] Among them, E(T) is the entropy of node T, N L and N R is the number of samples of the left and right child nodes after the split. Each tree grows to the maximum possible depth without pruning. The node splitting stop condition can be that the depth of the decision tree is greater than or equal to the first threshold or the number of training samples in the first subset or the number of training samples in the second subset is less than the second threshold.

[0119] In addition, in the embodiment of the present application, the performance of the tree can be verified using the out-of-bag (OOB) of random forest to estimate the generalization error of the model:

[0120]

[0121] Among them, 1 represents the indicator function, y i represents the true label of sample i, Represents the prediction on the OOB sample.

[0122] Based on the above embodiments, the present application also provides another flow chart of a fire identification method, such as Figure 6As shown in the figure, the multi-dimensional sensor parameters include audio data, image data and thermal imaging data, and then the wavelet transform and Kalman filter methods are combined to filter out noise and improve the clarity of audio data; the algorithm based on Mel Frequency Cepstrum Coefficient (MFCC) is used to extract audio features; the technology based on gray level co-occurrence matrix and Sobel operator is used to extract the features of thermal imaging data; and the convolutional neural network (CNN) and local binary pattern (LBP) algorithm are used to extract image features. In order to further enhance the generalization ability and accuracy of the fire detection model, a fire recognition model based on the random forest algorithm is used to identify whether a fire occurs in the area to be identified, as well as the fire situation when a fire occurs. The fire recognition model integrates feature data from three different sources to achieve high-precision fire detection. This multi-source information fusion method not only improves the accuracy of detection, but also enhances the application ability of the system in complex environments.

[0123] Figure 7 A schematic diagram of the structure of a fire identification device provided by an embodiment of the present application is shown as follows: Figure 7 As shown, the fire identification device 700 may include: an acquisition module 710 , an extraction module 720 and an identification module 730 .

[0124] In this embodiment, the acquisition module 710 is used to acquire image data, temperature sensing data and audio data corresponding to the area to be identified; the extraction module 720 is used to extract image features corresponding to the image data, temperature sensing features corresponding to the temperature sensing data and audio features corresponding to the audio data; the identification module 730 is used to input the image features, the temperature sensing features and the audio features into a fire identification model to obtain a fire identification result corresponding to the area to be identified, wherein the fire identification model is constructed based on a random forest algorithm.

[0125] In one implementation, the extraction module 720 is specifically used to: perform noise reduction processing on the audio data to obtain the noise-reduced audio data; and perform feature extraction on the noise-reduced audio data using Mel-frequency cepstral coefficients to obtain the audio features.

[0126] In one implementation, the denoising process of the audio data to obtain the denoised audio data includes: performing a wavelet transform on each audio frame of the audio data to obtain a time-frequency local feature corresponding to each audio frame, wherein the time-frequency local feature is used to represent the local features of the audio frame in the time dimension and the frequency dimension; and performing an inverse wavelet transform on each of the time-frequency local features to obtain the denoised audio data.

[0127] In one implementation, before obtaining the denoised audio data by performing inverse wave transform on each of the time-frequency local features, the method further includes: performing enhancement processing on the time-frequency local features corresponding to each audio frame through a Kalman filter.

[0128] In one implementation, the extraction module 720 is specifically used to: use a convolutional neural network to extract features from the image data to obtain initial image features; use a local binary pattern to binarize the local neighborhood of each pixel in the image data to obtain a first texture feature of the image data; and perform feature fusion on the initial image features and the first texture features to obtain the image features.

[0129] In one implementation, the temperature sensing data includes a thermal imaging image, and the temperature sensing feature includes a thermal imaging feature; the extraction module 720 is specifically used to: perform feature extraction on the thermal imaging image based on a grayscale co-occurrence matrix to obtain a second texture feature corresponding to the thermal imaging image, wherein the second texture feature includes at least one of the following: contrast, homogeneity, angular second moment, entropy; perform feature extraction on the thermal imaging image based on an edge detection algorithm to obtain an edge feature corresponding to the thermal imaging image; perform feature fusion on the second texture feature and the edge feature to obtain the image feature.

[0130] In one implementation, the fire identification device 200 may further include: a construction module, used to generate multiple sample subsets from the target training data set by guided sampling, wherein each of the sample subsets includes a first number of training samples; for each of the sample subsets, construct a corresponding decision tree; and based on the multiple decision trees, construct the fire identification model.

[0131] In one implementation, the construction of the corresponding decision tree includes: in each node splitting process, by performing feature selection and splitting point evaluation on each node, determining the optimal splitting point, and dividing the sample subset into a first subset and a second subset; recursively repeating the above splitting process for the first subset and the second subset until a stopping condition is met; wherein the stopping condition includes one of the following: the depth of the decision tree is greater than or equal to a first threshold; the number of training samples in the first subset or the number of training samples in the second subset is less than a second threshold.

[0132] A fire identification device provided in an embodiment of the present application can implement each process implemented in the fire identification method embodiment, and will not be described again here to avoid repetition.

[0133] The fire identification device in the embodiment of the present application may be a device, or a component, an integrated circuit, or a chip in an electronic device. The embodiment of the present application does not specifically limit this.

[0134] The fire identification device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0135] Optional, such as Figure 8 As shown, the embodiment of the present application further provides an electronic device 800, including a processor 810, a memory 820, and a program or instruction stored in the memory 820 and executable on the processor 810. When the program or instruction is executed by the processor 810, each process of the above-mentioned fire identification method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0136] The embodiment of the present application further provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned fire identification method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0137] The processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0138] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned fire identification method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0139] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0140] The embodiment of the present application also provides a computer program product, which includes at least one computer program. When the computer program is loaded and executed by a processor, the various processes of the above-mentioned fire identification method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0141] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0142] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0143] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A fire identification method, characterized in that: include: Acquire image data, temperature data and audio data corresponding to the area to be identified; Extracting image features corresponding to the image data, temperature features corresponding to the temperature data, and audio features corresponding to the audio data; The image features, the temperature features and the audio features are input into a fire recognition model to obtain a fire recognition result corresponding to the area to be recognized, wherein the fire recognition model is constructed based on a random forest algorithm.

2. The method according to claim 1, characterized in that: Extracting audio features corresponding to the audio data includes: Performing noise reduction processing on the audio data to obtain the audio data after noise reduction; The audio features are obtained by extracting features from the denoised audio data using Mel-frequency cepstral coefficients.

3. The method according to claim 2, characterized in that The performing noise reduction processing on the audio data to obtain the noise-reduced audio data includes: By performing wavelet transform on each audio frame of the audio data, a time-frequency local feature corresponding to each audio frame is obtained, wherein the time-frequency local feature is used to represent the local features of the audio frame in the time dimension and the frequency dimension; By performing inverse wavelet transform on each of the time-frequency local features, the audio data after noise reduction is obtained.

4. The method according to claim 3, characterized in that Before obtaining the denoised audio data by performing inverse wavelet transform on each of the time-frequency local features, the method further includes: The time-frequency local features corresponding to each audio frame are enhanced by using a Kalman filter.

5. The method according to claim 1, characterized in that Extracting image features corresponding to the image data includes: Using a convolutional neural network to extract features from the image data to obtain initial image features; Using a local binary pattern to perform binarization processing on a local neighborhood of each pixel in the image data to obtain a first texture feature of the image data; The initial image feature and the first texture feature are fused to obtain the image feature.

6. The method according to claim 1, characterized in that The temperature sensing data includes a thermal imaging image, and the temperature sensing feature includes a thermal imaging feature; Extracting the temperature sensing features corresponding to the temperature sensing data includes: Performing feature extraction on the thermal image based on a gray level co-occurrence matrix to obtain a second texture feature corresponding to the thermal image, wherein the second texture feature includes at least one of the following: contrast, homogeneity, angular second moment, and entropy; Extracting features from the thermal image based on an edge detection algorithm to obtain edge features corresponding to the thermal image; The second texture feature and the edge feature are fused to obtain the image feature.

7. The method according to claim 1, characterized in that Before inputting the image features, the temperature features and the audio features into a fire recognition model to obtain a fire recognition result corresponding to the area to be recognized, the method further includes: Generating a plurality of sample subsets from a target training data set by guided sampling, wherein each of the sample subsets includes a first number of training samples; For each of the sample subsets, construct a corresponding decision tree; Based on the multiple decision trees, the fire identification model is constructed.

8. The method according to claim 7, characterized in that The constructing of the corresponding decision tree comprises: In each node splitting process, the optimal splitting point is determined by performing feature selection and splitting point evaluation on each node, and the sample subset is divided into a first subset and a second subset; Recursively repeating the above splitting process on the first subset and the second subset until a stopping condition is met; The stop condition includes one of the following: The depth of the decision tree is greater than or equal to a first threshold; The number of training samples in the first subset or the number of training samples in the second subset is less than a second threshold.

9. A fire identification device, characterized in that: include: An acquisition module, used to acquire image data, temperature data and audio data corresponding to the area to be identified; An extraction module, used to extract image features corresponding to the image data, temperature features corresponding to the temperature data, and audio features corresponding to the audio data; An input module is used to input the image features, the temperature features and the audio features into a fire recognition model to obtain a fire recognition result corresponding to the area to be recognized, wherein the fire recognition model is constructed based on a random forest algorithm.

10. An electronic device, characterized in that: include: A processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the fire identification method according to any one of claims 1 to 8.

11. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the fire identification method according to any one of claims 1 to 8 are implemented.

12. A computer program product, characterized in that The computer program product comprises at least one computer program, and when the computer program is loaded and executed by a processor, the steps of the fire identification method according to any one of claims 1 to 8 are implemented.

Images