Fire point identification method and system based on infrared imaging moonlet

Abstract

The invention provides a fire point identification method and system based on an infrared imaging moonlet, and relates to the technical field of image processing. The method comprises the following steps: acquiring an infrared image shot by a moonlet; dividing the infrared image into a plurality of to-be-identified areas in a window sliding mode; calculating a temperature intensity value of each to-be-identified area; judging whether the temperature intensity value of each to-be-identified area is greater than a preset temperature intensity value or not; if yes, taking the to-be-identified area as a fire point candidate area; otherwise, removing the to-be-identified area; fire point features and dynamic features of the fire point candidate areas are extracted; and outputting a fire point recognition result through a graph neural network model based on multi-view image features according to the fire point features and the dynamic features. According to the method, the sensitivity of fire point detection to time sequence change is improved by introducing the dynamic features, and the detection accuracy of the fire points is remarkably improved by combining the strong feature expression capability of the graph neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for identifying fire points based on an infrared imaging small satellite. Background Art

[0002] With the rapid development of earth observation technology, infrared imaging satellites have become important tools in fire monitoring, thermal anomaly detection, and emergency response due to their small size, low cost, and flexible launch. In fire monitoring applications, fire point identification is one of the core tasks of infrared imaging technology. The ability to quickly and accurately identify fire points is of great significance to reducing fire losses, protecting the ecological environment, and ensuring the safety of human life and property. However, infrared imaging satellites are limited by their observation resolution, imaging frequency, and transmission capacity, and there are still many challenges in achieving efficient fire point identification.

[0003] At present, technical research on fire point identification mainly focuses on the following aspects: (1) Fire point identification based on traditional algorithms: using the brightness temperature threshold, background temperature contrast and spatial texture features of infrared images to detect fire points; (2) Fire point identification based on machine learning: using traditional machine learning algorithms such as support vector machine (SVM) and random forest (RF) to classify fire points, which improves the accuracy and efficiency of detection to a certain extent.

[0004] However, existing technologies mostly rely on fixed temperature thresholds or single features to judge fire points, lack analysis of the dynamic changes of fire points, and have poor accuracy in dynamic fire point detection. At the same time, existing methods lack adaptability to complex scenes, making it difficult to distinguish fire points from interference factors, and lack the use of spatial features and multi-perspective data, which makes false detections and missed detections prone to occur. Summary of the invention

[0005] In order to solve the technical problems that the existing technologies mostly rely on fixed temperature thresholds or single features to judge fire points, lack analysis of dynamic changes of fire points, and have poor accuracy in dynamic fire point detection; at the same time, the existing methods are insufficiently adaptable to complex scenes, have difficulty in distinguishing fire points from interference factors, lack the use of spatial features and multi-view data, and are prone to false detection and missed detection, the present invention provides a fire point recognition method and system based on infrared imaging small satellite.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] First aspect:

[0008] An embodiment of the present invention provides a method for identifying fire points based on an infrared imaging satellite, comprising:

[0009] S1: Acquire infrared images taken by small satellites;

[0010] S2: Dividing the infrared image into a plurality of areas to be identified in a sliding window manner;

[0011] S3: Calculating the temperature intensity value of each of the areas to be identified;

[0012] S4: Determine whether the temperature intensity value of each of the to-be-identified areas is greater than a preset temperature intensity value; if so, use the to-be-identified area as a candidate fire point area; otherwise, remove the to-be-identified area;

[0013] S5: extracting fire point features and dynamic features of the fire point candidate area;

[0014] S6: According to the fire point features and the dynamic features, the fire point recognition results are output through a graph neural network model based on multi-view image features.

[0015] Second aspect:

[0016] An embodiment of the present invention provides a fire point identification system based on an infrared imaging small satellite, comprising:

[0017] processor;

[0018] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the fire point identification method based on the infrared imaging small satellite as described in the first aspect is implemented.

[0019] The third aspect:

[0020] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method for identifying fire points based on infrared imaging small satellites as described in the first aspect is implemented.

[0021] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0022] (1) In the embodiment of the present invention, the interference of complex background on fire point detection is reduced by sliding window division, spatial features and dynamic features are extracted to enhance the ability to distinguish fire points from interference factors, and graph neural networks are used to integrate multi-view features and spatial relationships, thereby improving the adaptability of the algorithm to complex scenes and effectively reducing the problems of false detection and missed detection.

[0023] (2) In the embodiment of the present invention, by extracting the dynamic features and features of the fire point and combining them with the graph neural network model for comprehensive analysis, the fire point identification is freed from the reliance on a single temperature threshold or static features. At the same time, the introduction of dynamic features improves the sensitivity of fire point detection to time series changes. Combined with the powerful feature expression ability of the graph neural network, the detection accuracy of dynamic fire points is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A schematic diagram of a flow chart of a method for identifying fire points based on an infrared imaging satellite according to an embodiment of the present invention;

[0026] Figure 2 A schematic structural diagram of a fire point identification system based on an infrared imaging small satellite provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0028] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0029] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0030] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0031] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0032] Reference Manual Attached Figure 1 , showing a flow chart of a fire point identification method based on an infrared imaging small satellite provided in an embodiment of the present invention.

[0033] The embodiment of the present invention provides a method for identifying fire points based on infrared imaging satellites. The method can be implemented by a fire point identification device based on infrared imaging satellites. The fire point identification device based on infrared imaging satellites can be a terminal or a server. The processing flow of the method for identifying fire points based on infrared imaging satellites can include the following steps:

[0034] S1: Get infrared images taken by small satellites.

[0035] S2: Divide the infrared image into multiple areas to be identified in a sliding window manner.

[0036] Among them, the sliding window is a movable window, whose size is usually fixed. The window slides on the data set to process the elements in the window and move the window as needed to process the entire data set.

[0037] Specifically, we need to define the size of the window (width and height) and the step size of the window sliding according to the size of the input infrared image. It is usually set to a fixed-size rectangular area for extracting the local area. The step size is an integer value, which represents the horizontal and vertical movement of the window each time it slides, ensuring that the window does not exceed the image boundary when sliding on the image.

[0038] In the image, the window starts from the upper left corner and then gradually slides horizontally or vertically until the entire image is covered. Each time it slides, the image area in the current window is extracted and saved as a region to be recognized.

[0039] In the present invention, the sliding window can gradually scan the entire image and extract the features of each local area. In this way, the model can obtain information from each small area of ​​the image instead of relying solely on the overall features of the image. At the same time, the sliding window will traverse each area of ​​the image and scan the entire image through overlapping windows to ensure that each local area can be processed and avoid information loss.

[0040] S3: Calculate the temperature intensity value of each area to be identified.

[0041] Among them, the temperature intensity value usually refers to the temperature of a certain image area or a certain pixel, reflecting the thermal information of the area or pixel. In infrared imaging, the grayscale value of each pixel can represent the thermal radiation intensity of the location. By converting these grayscale values ​​into temperature values, we can get the temperature intensity value of the area.

[0042] In a possible implementation, S3 specifically includes:

[0043] S301: Using a device calibration formula, convert the grayscale value of each pixel in each of the to-be-identified areas into a temperature value:

[0044] Among them, the calibration formula is used to convert the grayscale value (or digital value) of each pixel in the image into the corresponding temperature value. Since the image data obtained by the infrared imaging device is usually the radiation intensity or grayscale level, these data need to be mapped to the actual temperature through the calibration formula. Equipment calibration is an important step to ensure the accuracy and reliability of infrared measurement.

[0045]

[0046] Among them, T(x,y) represents the temperature value of the pixel, I(x,y) represents the grayscale value of the pixel, x represents the horizontal coordinate of the pixel in the area to be identified, y represents the vertical coordinate of the pixel in the area to be identified, a represents the first linear calibration parameter provided by the infrared imaging device, and b represents the second linear calibration parameter provided by the infrared imaging device.

[0047] In the present invention, by converting the gray value into a temperature value using a device calibration formula, errors caused by nonlinear response of the imaging device or environmental factors can be eliminated, thereby improving the accuracy and reliability of infrared imaging results.

[0048] S302: Calculate the temperature intensity value of each area to be identified based on the temperature values ​​of all pixels in each area to be identified:

[0049]

[0050] Among them, T l Represents the lth region to be identified R l The temperature intensity value, N represents the total number of pixels, (x, y) represents the coordinates of the pixel, R l Indicates the lth area to be identified.

[0051] In the present invention, by calculating the temperature intensity value of the region, we can effectively identify the high temperature area in the image and distinguish the fire point from other irrelevant areas. At the same time, the calculation of the temperature intensity value helps to conduct more refined regional analysis, especially when monitoring a large area, it can help to timely identify the temperature abnormal area, thereby providing more accurate information for decision-making.

[0052] S4: Determine whether the temperature intensity value of each area to be identified is greater than a preset temperature intensity value. If so, the area to be identified is used as a candidate fire point area. Otherwise, the area to be identified is eliminated.

[0053] In the present invention, by setting a preset temperature intensity threshold, those areas with temperatures below the preset threshold can be effectively filtered out, thereby reducing background noise and interference from non-target areas. At the same time, by further analyzing and processing only the areas that meet the conditions, the efficiency of the overall system can be improved. The temperature intensity value judgment removes areas in the image that are not related to the target, so that the remaining areas are concentrated in the fire point area of ​​interest, further improving the accuracy of the detection results.

[0054] S5: Extract the fire point features and dynamic features of the fire point candidate area.

[0055] Among them, the fire point feature is the attribute of the fire area extracted through static image analysis, which mainly describes the shape, size, boundary characteristics of the flame, etc. These features help identify the fire area and distinguish the fire point from other high-temperature areas.

[0056] Among them, dynamic features are extracted from time series infrared images to describe the changing behavior of fire points over time. Fire points usually have dynamic characteristics, such as flame expansion and morphological changes.

[0057] In a possible implementation manner, the fire point features include a boundary chain code, a circularity of the flame, and a contour line. The method for extracting the fire point features specifically includes:

[0058] Among them, the boundary chain code is used to describe the shape of the boundary by encoding the boundary of the object. The boundary chain code can capture the geometric information of the fire point contour, especially the shape of the flame.

[0059] Among them, circularity is an indicator that measures the similarity of an object's shape to a circle. For a fire point, a flame is usually considered to be an approximately circular or elliptical area, so the circularity of the flame helps to identify the fire point.

[0060] Among them, the contour line is a curve that describes the edge shape of an object. In fire point recognition, the contour line of the flame helps to identify the specific boundary shape of the fire point.

[0061] S501: Convert the candidate region of the fire point into the HSV color space:

[0062]

[0063] Among them, V represents the color space of the candidate area of ​​the fire point, x represents a single pixel in the color space, x(H) represents the hue of the pixel, x(S) represents the saturation of the pixel, and x(V) represents the brightness component.

[0064] S502: Determine the flame area in the fire point candidate area based on the value range of the HSV color space.

[0065] Specifically, based on the color range, the pixels in the image that belong to the flame color are divided into flame regions.

[0066] S503: Use the Laplace operator to extract the boundary chain code of the flame area.

[0067] Among them, the Laplacian operator is an important mathematical operator, which is widely used in various fields of mathematics, physics, engineering and computer science. It is mainly used to describe the local changes of functions. In image processing and computer vision, the Laplacian operator is often used for tasks such as edge detection and image filtering.

[0068] S504: Counting the number of pixels in the flame area as the area size of the flame.

[0069] S505: Calculate the circularity of the flame according to the perimeter of the boundary and the area of ​​the flame.

[0070]

[0071] Among them, Q represents the circularity of the flame, L represents the circumference of the boundary, and S represents the area of ​​the flame.

[0072] S506: In a clockwise direction, a plurality of boundary points on the boundary of the flame region are represented as:

[0073]

[0074] Among them, z i represents the complex number representation of the i-th boundary point, x i Represents the horizontal coordinate of the i-th boundary point on the flame area boundary, y i represents the ordinate of the i-th boundary point on the flame area boundary, and j represents the imaginary unit.

[0075] S507: Perform discrete Fourier transform on each boundary point to extract the frequency characteristics of each boundary point in the flame area:

[0076] Among them, discrete Fourier transform (DFT) is a discrete form of Fourier transform, which converts discrete signals from time domain (or space domain) to frequency domain and is widely used in signal processing, image processing, audio analysis and other fields.

[0077]

[0078] Among them, F w represents the Fourier coefficient of the ωth frequency component, and N represents the total number of boundary points.

[0079] S508: Arrange the frequencies from low to high, select the Fourier coefficients corresponding to the frequencies at the front, and use the inverse discrete Fourier transform to reconstruct the contour of the flame area:

[0080]

[0081] in, represents the complex representation of the i-th boundary point after reconstruction, and k represents the number of selected Fourier coefficients.

[0082] In a possible implementation manner, the dynamic feature extraction method specifically includes:

[0083] S509: Based on the flame features of the fire point candidate area, construct an n×m dynamic feature matrix, where n represents the number of continuous images and m represents the number of flame features.

[0084] S510: Calculate the mean and mean square error of each element in the dynamic feature matrix respectively as the description of the dynamic feature:

[0085]

[0086] Among them, E(j) represents the mean of the j-th feature in n consecutive images, n represents the number of consecutive images, and X(i,j) represents the element in the i-th row and j-th column of the state feature matrix.

[0087]

[0088] Among them, S(j) represents the mean square error of the j-th feature in n consecutive images.

[0089] In the present invention, by combining the fire point features and dynamic features, the system can perform comprehensive analysis from both static images and dynamic behaviors, thereby more accurately identifying the fire point. At the same time, by integrating the static and dynamic features of the fire point, it is possible to reduce false alarms caused by other heat sources or background noise, while improving the detection capability of missed alarms.

[0090] S6: According to the fire point characteristics and dynamic characteristics, the fire point recognition results are output through the graph neural network model based on multi-view image features.

[0091] Among them, the Multi-view Image Feature Graph Neural Network Model is a method that applies Graph Neural Network (GNN) to extract and fuse features from multi-view images. Its main purpose is to take advantage of the advantages of graph neural networks and combine information from multiple image perspectives to improve the accuracy of image understanding, target recognition or classification.

[0092] In a possible implementation, the graph neural network model includes a shared module; S6 specifically includes:

[0093] S601: Each node of the graph neural network model represents a pair of library images for verifying the similarity between images.

[0094] Among them, the reference image refers to a set of preprocessed and stored images, which are used as reference or comparison images to assist the analysis, comparison, recognition or classification of the current image. The reference image usually contains known features, targets or background information, and can provide correlation or similarity information with the image to be analyzed.

[0095] S602: Construct an undirected graph G(V,E) based on the library image and multiple candidate fire point regions, where V represents a node set. , v i represents the i-th node, i=1,2,…q, q represents the total number of nodes, and E represents the edges between nodes.

[0096] Among them, an undirected graph is a graph model without direction, in which the edge connects two nodes and has no direction. In an undirected graph, the relationship between nodes is symmetrical and is widely used in social networks, computer networks, road systems, chemical molecular structures and other fields.

[0097] S603: Use the shared module to encode the relationship features of each pair of library images, and extract the fire point features and dynamic features of each pair of library images.

[0098] S604: Process the fire point features and dynamic features of each pair of library images through the fully connected layer of the shared module to generate a fire point feature vector and a dynamic feature vector for node input.

[0099] It should be noted that the shared module includes CNN of network flow sd, which is used to extract image relationship features, and fully connected layers (FCs) for further feature processing.

[0100] S605: Perform differential operation on the fire point feature vector and dynamic feature vector extracted from each pair of library images to generate differential features:

[0101]

[0102] Among them, d i represents the differential feature vector of the i-th node, C2 represents the fire point feature and dynamic feature vector of the second image in the library image pair, and C1 represents the fire point feature and dynamic feature vector of the first image in the library image pair.

[0103] S606: Use the differential features as input features of the graph neural network nodes, and update the node features through the message passing mechanism of the graph neural network.

[0104] In a possible implementation manner, S606 specifically includes:

[0105] S6061: Use the image similarity estimation function to calculate the edge weight:

[0106]

[0107] Among them, W ij represents the edge weight between the i-th node and the j-th node, represents the similarity score between the image pair of the i-th node and the image pair of the j-th node, g i represents the image pair of the i-th node, g j represents the image pair of the j-th node, and exp represents the exponential function.

[0108] S6062: Input the node relationship features into the two-layer fully connected network to generate deep messages:

[0109]

[0110] Among them, t i represents the depth message, F( ) represents a two-layer fully connected network, Represents the initial relationship feature of the i-th node.

[0111] S6063: Based on the edge weight, iteratively update the weighted sum of its own features and adjacent node features:

[0112]

[0113] in, represents the differential eigenvector of the i-th node after the k-th iteration, α represents the updated weight coefficient, represents the differential eigenvector of the i-th node after the k-1th iteration, N represents the total number of nodes, W ij represents the edge weight between the i-th node and the j-th node, represents the depth message passed from the jth node in the k-1th iteration.

[0114] In the present invention, by calculating the edge weights between nodes (i.e., the similarity scores between images from different perspectives), the model can more flexibly adjust the information transmission and fusion between each perspective, thereby improving the accuracy of image understanding. At the same time, through the deep message passing mechanism in the graph neural network, the node can gradually fuse the information of neighboring nodes and update the feature representation. This allows the information of each perspective to be effectively integrated, thereby improving the accurate representation of the fire point features.

[0115] S607: Use a linear classifier to classify each updated node feature and output a similarity score:

[0116] Among them, the linear classifier is a classification model based on a linear decision boundary, which is used to classify samples in a data set into two or more categories. It maps the input features to a category label by constructing a linear function. Linear classifiers are widely used in binary classification and multi-classification problems in machine learning, and usually have the advantages of high computational efficiency and simple implementation.

[0117]

[0118] Among them, f(d i ) represents the similarity score of the i-th updated node in the graph neural network model, represents the differential eigenvector of the i-th node after the k-th iteration.

[0119] S608: Determine the fire point division result according to the similarity score of each updated node.

[0120] In a possible implementation manner, the fire point division result in S608 is specifically:

[0121]

[0122] Among them, r represents the fire point identification result, r=1 means that the area corresponding to the node is the fire point area, r=0 means that the area corresponding to the node is the non-fire point area, and B represents the preset threshold.

[0123] In one possible implementation, the graph neural network model is optimized using a cross entropy loss function.

[0124] Among them, the cross entropy loss function is a widely used loss function in classification problems, especially suitable for binary and multi-classification tasks. It measures the performance of the classification model by calculating the difference between the true label and the probability distribution predicted by the model. The cross entropy loss function has the advantages of high optimization efficiency, simple calculation and wide application, and is an indispensable tool in machine learning and deep learning.

[0125] The cross entropy loss function is specifically:

[0126]

[0127] Among them, L represents the total value of the cross entropy loss function, n represents the total number of nodes, and y i represents the true label of the i-th node, y i = 1 means that the image pair of the i-th node is similar, y i = 0 means that the image pair of the i-th node is not similar, f(d i ) represents the similarity score of the i-th updated node in the graph neural network model.

[0128] In the present invention, multi-view images can provide different views of the target from multiple angles, making the identification of fire points more accurate and robust. Using the graph neural network model, these multi-view features can be effectively fused through the graph structure. At the same time, by analyzing the changes in the fire point area in continuous images, dynamic features (such as flame expansion, morphological changes, etc.) can help monitor the changes of fire points in real time, predict the spread trend of fire, and improve the timeliness and accuracy of fire point detection.

[0129] Optionally, the graph neural network model based on multi-view image features is optimized by a particle swarm optimization algorithm with the goal of minimizing the cross entropy loss function.

[0130] S701: Using the inverse of the cross entropy loss function as the fitness function of the particle swarm optimization algorithm.

[0131] S702: Setting the particle swarm optimization algorithm parameters, wherein the particle swarm optimization algorithm parameters include population size, inertia weight, acceleration factor, mutation probability, neighborhood range, mutation factor, and maximum number of iterations.

[0132] S703: Initialize the position and velocity of each particle in the population, where each particle represents a feasible graph neural network model parameter solution.

[0133] S704: Calculate the fitness of each particle in the neighborhood of each particle, and select the particle with the highest fitness value as the local optimal position in the neighborhood.

[0134] S705: Update the speed and position of each particle according to the local optimal position:

[0135]

[0136]

[0137] Among them, V i(t+1) represents the velocity of the ith particle at the t+1 iteration, w represents the inertia weight, c1, c2 represent constants, r1, r2 represent constants, Pbest i represents the position where the fitness value of the i-th particle is optimal, X i indicates that Gbest indicates that X i (t) represents the position of the i-th particle at the t-th iteration, w max Indicates the maximum value of the inertia weight, w min represents the minimum value of the inertia weight, t represents the tth iteration, T max Indicates the maximum number of iterations.

[0138]

[0139] Among them, X i (t+1) represents the position of the i-th particle at the t+1-th iteration.

[0140] Optionally, a bounds check is performed after the position update to ensure that the particle's velocity or position does not exceed the predefined search space.

[0141] S706: Generate a random number rand, rand∈(0,1), and determine whether the random number rand is less than the preset mutation probability. If so, perform a first mutation operation on the current individual optimal position, i.e., the local optimal position, to obtain a first test position, and enter S707; otherwise, perform a second mutation operation on the current individual optimal position, to obtain a second test position, and enter S708.

[0142] In a possible implementation, the formula of the first mutation operation is specifically:

[0143]

[0144] Where T1 represents the first test position, X i represents the current position of the i-th particle, F represents the mutation factor, F∈(0,1), Pbest r represents the individual optimal position of the first particle randomly selected from the population.

[0145] In a possible implementation, the formula of the second mutation operation is specifically:

[0146]

[0147] Among them, T2 represents the second test position, Lbest i Pbest represents the position with the highest fitness value of all particles in the neighborhood of the i-th particle, F represents the mutation factor, F∈(0,1), rrepresents the individual optimal position of the first particle randomly selected from the population, Pbest s represents the individual optimal position of the second particle randomly selected from the population.

[0148] In the present invention, by introducing the first mutation and the second mutation operation when updating the particle position, the diversity of the search space is increased, and the particle swarm can jump out of the local optimal solution, thereby improving the global optimization ability. At the same time, the mutation operation enables particles to try more potential solutions, thereby increasing the probability of searching for the global optimal solution. The first mutation operation expands the search range by randomly selecting the perturbation of the individual optimal position; the second mutation operation combines the optimal position of fitness in the neighborhood to refine the search and improve the accuracy of the solution.

[0149] S707: According to the first test position, update the local optimal position of the particle and the global optimal position of the particle, and proceed to S709.

[0150] In a possible implementation manner, S707 specifically includes:

[0151] S7071: Calculate the fitness value of the particle corresponding to the first test position.

[0152] S7072: According to the fitness value of the particle corresponding to the first test position, the first target position for the next iteration is selected by the following formula:

[0153]

[0154] Where T1 represents the first test position, f(T1) represents the fitness value of the first test position, and f(X i (t)) represents the fitness value of the position of the i-th particle at the t-th iteration.

[0155] S7073: Based on the first target position, update the individual optimal position using the following formula:

[0156]

[0157] Among them, Pbest i represents the optimal position of the individual optimal position of the i-th particle in all iterations, f(X i (t+1)) represents the fitness value of the position of the i-th particle at the t+1th iteration, f(Pbest i ) represents the fitness value of the historical individual optimal position of the i-th particle.

[0158] S7074: Based on the updated individual optimal position, update the global optimal position using the following formula:

[0159]

[0160] Among them, Gbest represents the global optimal position, and f(Gbest) represents the fitness value of the global optimal position.

[0161] S708: According to the second test position, update the local optimal position of the particle and the global optimal position of the particle, and proceed to S709.

[0162] In a possible implementation manner, S708 specifically includes:

[0163] S7081: Calculate the fitness value of the particle corresponding to the second test position.

[0164] S7082: According to the fitness value of the particle corresponding to the first test position, the second target position for the next iteration is selected by the following formula:

[0165]

[0166] Where T2 represents the second test position, f(T2) represents the fitness value of the second test position, and f(X i (t)) represents the fitness value of the position of the i-th particle at the t-th iteration.

[0167] S7083: Based on the second target position, update the individual optimal position using the following formula:

[0168]

[0169] Among them, Pbest i represents the optimal position of the individual optimal position of the i-th particle in all iterations, f(X i (t+1)) represents the fitness value of the position of the i-th particle at the t+1th iteration, f(Pbest i ) represents the fitness value of the historical individual optimal position of the i-th particle.

[0170] S7084: Based on the updated individual optimal position, update the global optimal position using the following formula:

[0171]

[0172] Among them, Gbest represents the global optimal position, and f(Gbest) represents the fitness value of the global optimal position.

[0173] S709: Add 1 to the number of iterations and determine whether the number of iterations reaches the maximum number of iterations. If so, end the training and output the graph neural network model parameters corresponding to the particle at the global optimal position. Otherwise, return to S704.

[0174] In the present invention, the particle swarm optimization algorithm is a swarm intelligence optimization method, which performs global search by simulating the foraging behavior of bird flocks. After introducing the mutation operation, the particle swarm can jump out of the local optimal solution, explore a wider search space, and improve the global optimization ability. Through global optimization, a more suitable graph neural network parameter configuration can be found to improve the performance of the model. At the same time, by introducing the mutation operation, the particle swarm optimization achieves a balance between global search and local search. Global search ensures extensive exploration, while local optimization refines the search area and further improves the quality of understanding.

[0175] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0176] (1) In the embodiment of the present invention, the interference of complex background on fire point detection is reduced by sliding window division, spatial features and dynamic features are extracted to enhance the ability to distinguish fire points from interference factors, and graph neural networks are used to integrate multi-view features and spatial relationships, thereby improving the adaptability of the algorithm to complex scenes and effectively reducing the problems of false detection and missed detection.

[0177] (2) In the embodiment of the present invention, by extracting the dynamic features and features of the fire point and combining them with the graph neural network model for comprehensive analysis, the fire point identification is freed from the reliance on a single temperature threshold or static features. At the same time, the introduction of dynamic features improves the sensitivity of fire point detection to time series changes. Combined with the powerful feature expression ability of the graph neural network, the detection accuracy of dynamic fire points is significantly improved.

[0178] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a fire point identification system based on an infrared imaging small satellite provided by the present invention.

[0179] The present invention also provides a fire point identification system 20 based on infrared imaging small satellite, which is applied to the above-mentioned fire point identification method based on infrared imaging small satellite, comprising:

[0180] Processor 201.

[0181] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, the fire point identification method based on the infrared imaging small satellite as in the method embodiment is implemented.

[0182] The fire point identification system 20 based on infrared imaging small satellite provided by the present invention can execute the above-mentioned fire point identification method based on infrared imaging small satellite and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0183] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0184] (1) In the embodiment of the present invention, the interference of complex background on fire point detection is reduced by sliding window division, spatial features and dynamic features are extracted to enhance the ability to distinguish fire points from interference factors, and graph neural networks are used to integrate multi-view features and spatial relationships, thereby improving the adaptability of the algorithm to complex scenes and effectively reducing the problems of false detection and missed detection.

[0185] (2) In the embodiment of the present invention, by extracting the dynamic features and features of the fire point and combining them with the graph neural network model for comprehensive analysis, the fire point identification is freed from the reliance on a single temperature threshold or static features. At the same time, the introduction of dynamic features improves the sensitivity of fire point detection to time series changes. Combined with the powerful feature expression ability of the graph neural network, the detection accuracy of dynamic fire points is significantly improved.

[0186] It should be understood that the processor in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0187] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0188] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0189] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0190] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0191] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0192] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0193] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0194] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0195] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0196] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0197] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0198] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for identifying fire points based on an infrared imaging small satellite as described in the method embodiment is implemented.

[0199] A computer-readable storage medium provided by the present invention can implement the steps and effects of the fire point identification method based on infrared imaging small satellite in the above method embodiment. To avoid repetition, the present invention will not go into details.

[0200] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0201] (1) In the embodiment of the present invention, the interference of complex background on fire point detection is reduced by sliding window division, spatial features and dynamic features are extracted to enhance the ability to distinguish fire points from interference factors, and graph neural networks are used to integrate multi-view features and spatial relationships, thereby improving the adaptability of the algorithm to complex scenes and effectively reducing the problems of false detection and missed detection.

[0202] (2) In the embodiment of the present invention, by extracting the dynamic features and features of the fire point and combining them with the graph neural network model for comprehensive analysis, the fire point identification is freed from the reliance on a single temperature threshold or static features. At the same time, the introduction of dynamic features improves the sensitivity of fire point detection to time series changes. Combined with the powerful feature expression ability of the graph neural network, the detection accuracy of dynamic fire points is significantly improved.

[0203] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0204] There are a few points to note:

[0205] (1) The drawings of the embodiments of the present invention only involve structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0206] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0207] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0208] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying fire points based on infrared imaging satellites, characterized in that: include: S1: Acquire infrared images taken by small satellites; S2: Dividing the infrared image into a plurality of areas to be identified in a sliding window manner; S3: Calculating the temperature intensity value of each of the areas to be identified; S4: determining whether the temperature intensity value of each of the to-be-identified areas is greater than a preset temperature intensity value; if so, taking the to-be-identified area as a candidate fire point area; Otherwise, the area to be identified is removed; S5: extracting fire point features and dynamic features of the fire point candidate area; S6: According to the fire point features and the dynamic features, the fire point recognition results are output through a graph neural network model based on multi-view image features.

2. The method for identifying fire points based on infrared imaging satellite according to claim 1, characterized in that: The S3 specifically includes: S301: Using a device calibration formula, convert the grayscale value of each pixel in each of the to-be-identified areas into a temperature value: ； Wherein, T(x,y) represents the temperature value of the pixel, I(x,y) represents the gray value of the pixel, x represents the abscissa of the pixel in the area to be identified, y represents the ordinate of the pixel in the area to be identified, a represents the first linear calibration parameter provided by the infrared imaging device, and b represents the second linear calibration parameter provided by the infrared imaging device; S302: Calculate the temperature intensity value of each of the to-be-identified areas according to the temperature values ​​of all pixels in each of the to-be-identified areas: ； Among them, T l Represents the lth region to be identified R l The temperature intensity value, N represents the total number of pixels, (x, y) represents the coordinates of the pixel, R l Indicates the lth area to be identified.

3. The method for identifying fire points based on infrared imaging satellite according to claim 1, characterized in that: The fire point features include boundary chain code, flame circularity and contour line.

4. The method for identifying fire points based on infrared imaging satellite according to claim 3 is characterized in that: The method for extracting the dynamic features specifically includes: S509: constructing an n×m dynamic feature matrix based on the flame features of the fire point candidate area, where n represents the number of continuous images and m represents the number of flame features; S510: Calculate the mean and mean square error of each element in the dynamic feature matrix respectively as the description of the dynamic feature: ； Where E(j) represents the mean value of the jth feature in n consecutive images, n represents the number of consecutive images, and X(i,j) represents the element in the i-th row and j-th column of the state feature matrix; ； Among them, S(j) represents the mean square error of the j-th feature in n consecutive images.

5. The method for identifying fire points based on infrared imaging satellite according to claim 1, characterized in that: The graph neural network model includes a shared module; S6 specifically includes: S601: Each node of the graph neural network model represents a pair of library images for verifying the similarity between images; S602: construct an undirected graph G(V,E) based on the library image and the plurality of candidate fire point regions, where V represents a node set, , v i represents the i-th node, i=1,2,…q, q represents the total number of nodes, and E represents the edge between nodes; S603: using a shared module to perform relationship feature encoding on each pair of library images, and extracting the fire point features and dynamic features of each pair of library images; S604: Processing the fire point features and dynamic features of each pair of library images through the fully connected layer of the shared module to generate a fire point feature vector and a dynamic feature vector for node input; S605: Perform differential operation on the fire point feature vector and dynamic feature vector extracted from each pair of library images to generate differential features: ； Among them, d i represents the differential feature vector of the i-th node, C2 represents the fire point feature and dynamic feature vector of the second image in the library image pair, and C1 represents the fire point feature and dynamic feature vector of the first image in the library image pair; S606: Using the differential feature as an input feature of a graph neural network node, and updating the node feature through a message passing mechanism of the graph neural network; S607: Use a linear classifier to classify each updated node feature and output a similarity score: ； Among them, f(d i ) represents the similarity score of the i-th updated node in the graph neural network model, represents the differential eigenvector of the i-th node after the k-th iteration; S608: Determine the fire point division result according to the similarity score of each updated node.

6. The method for identifying fire points based on infrared imaging satellite according to claim 5, characterized in that: The S606 specifically includes: S6061: Use the image similarity estimation function to calculate the edge weight: ； Among them, W ij represents the edge weight between the i-th node and the j-th node, represents the similarity score between the image pair of the i-th node and the image pair of the j-th node, g i represents the image pair of the i-th node, g j represents the image pair of the jth node, exp represents the exponential function; S6062: Input the node relationship features into the two-layer fully connected network to generate deep messages: ； Among them, t i represents the depth message, F( ) represents a two-layer fully connected network, Represents the initial relationship feature of the i-th node; S6063: Based on the edge weight, iteratively update the weighted sum of its own features and adjacent node features: ； in, represents the differential eigenvector of the i-th node after the k-th iteration, α represents the updated weight coefficient, represents the differential eigenvector of the i-th node after the k-1th iteration, N represents the total number of nodes, W ij represents the edge weight between the i-th node and the j-th node, represents the depth message passed from the jth node in the k-1th iteration.

7. The method for identifying fire points based on infrared imaging satellite according to claim 5, characterized in that: The fire point division result in S608 is specifically: ； Among them, r represents the fire point identification result, r=1 means that the area corresponding to the node is the fire point area, r=0 means that the area corresponding to the node is the non-fire point area, and B represents the preset threshold.

8. The method for identifying fire points based on infrared imaging satellite according to claim 1, characterized in that: The graph neural network model is optimized by using the cross entropy loss function.

9. The method for identifying fire points based on infrared imaging satellite according to claim 8, characterized in that: The cross entropy loss function is specifically: ； Among them, L represents the total value of the cross entropy loss function, n represents the total number of nodes, and y i represents the true label of the i-th node, y i = 1 means that the image pair of the i-th node is similar, y i = 0 means that the image pair of the i-th node is not similar, f(d i ) represents the similarity score of the i-th updated node in the graph neural network model.

10. A fire point identification system based on infrared imaging satellite, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the fire point identification method based on infrared imaging small satellite as described in any one of claims 1 to 9 is implemented.

Images