Fire point detection method based on satellite data and related equipment
By acquiring the latent spatial vector and covariance matrix of satellite time-series data and combining it with Mahalanobis distance to calculate the anomaly degree, the problem of inaccurate anomaly point identification in existing models is solved, and more efficient fire point detection is achieved.
Patent Information
- Application Number
- CN202511721982.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-02-27
AI Technical Summary
Existing fire detection models based on deep autoencoders assume that normal samples are distributed in an isotropic hypersphere when calculating latent space distances, leading to inaccurate identification of outliers.
The latent spatial vectors obtained from satellite time-series data are used to calculate the mean vector and covariance matrix. The latent spatial anomaly is calculated by combining Mahalanobis distance, and fire points are identified by a classifier.
It improves the accuracy of fire detection, reduces false positives for anomalies, and significantly enhances the robustness and detection accuracy of the model.
Smart Images
Figure CN121580228A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of satellite remote sensing data processing technology, and in particular to a fire detection method and related equipment based on satellite data. Background Technology
[0002] With the application of high-frequency Earth observation satellites, early fire detection using their time-series data has become mainstream. Among these, the anomaly detection framework based on deep autoencoders is one of the most advanced technical approaches currently available. The core idea of this framework is to train a deep learning model (such as an autoencoder based on Transformer, LSTM, etc.) on massive amounts of "normal" satellite time-series data without fire points. By learning how to effectively encode and reconstruct normal data, the model grasps the spatiotemporal variation patterns of the Earth's surface under various normal conditions.
[0003] However, this model suffers from a fundamental flaw that limits its performance in its calculation of "latent space distance": existing technologies generally use Euclidean distance to measure the distance between the test sample and the center (centroid / mean vector) of the normal sample cluster. The core problem with this method is that it implicitly assumes that normal samples are distributed in an "isotropic" hypersphere in the latent space. But in real-world scenarios, due to the complex correlations between different physical features, the distribution of normal sample data in the latent space is almost never a perfect sphere, but rather often a directional, stretchable "ellipsoid" or a more irregular shape, leading to inaccurate identification of outliers. Summary of the Invention
[0004] In view of this, the present invention provides a fire detection method and related equipment based on satellite data.
[0005] The specific technical solution of the first embodiment of the present invention is as follows: a fire point detection method based on satellite data, the method comprising: acquiring multiple satellite time-series data of a fire-free region; inputting the satellite time-series data into a preset encoder to obtain a latent spatial vector for each satellite time-series data; acquiring a mean vector between the latent spatial vectors, and acquiring a covariance matrix between the latent spatial vectors based on the mean vector and the latent spatial vector; acquiring target satellite time-series data of a target point to be detected, and acquiring a target latent spatial vector of the target satellite time-series data using the preset encoder; acquiring a latent spatial anomaly degree of the target satellite time-series data based on the target latent spatial vector, the mean vector, and the covariance matrix; inputting the latent spatial anomaly degree into a preset classifier to obtain a fire point detection result for the target point to be detected; the fire point detection result includes fire points and fire-free points.
[0006] Preferably, obtaining the mean vector among the potential spatial vectors includes: summing all the potential spatial vectors and dividing the sum by the total number of satellite time series data to obtain the mean vector.
[0007] Preferably, the covariance matrix is obtained using the following formula:
[0008] in, Let be the covariance matrix. The total number of satellite time-series data. Let be the latent spatial vector of the i-th satellite time series data. Let the mean vector be... This is the transpose operation for a vector.
[0009] Preferably, the latent space anomaly degree is obtained using the following formula:
[0010] in, The degree of anomaly in the latent space. Let be the target potential space vector. Let the mean vector be... Let be the inverse of the covariance matrix. This is the transpose operation for a vector.
[0011] Preferably, after obtaining the target potential spatial vector of the target satellite time series data using the preset encoder, the method further includes: obtaining the reconstruction error when the encoder obtains the target potential spatial vector; then, inputting the potential spatial anomaly degree into the preset classifier to obtain the fire detection result of the target point to be detected includes: inputting the potential spatial anomaly degree and the reconstruction error into the preset classifier to obtain the fire detection result of the target point to be detected.
[0012] Preferably, after obtaining the covariance matrix between the potential spatial vectors, the method further includes: adding a perturbation to the covariance matrix to obtain an optimized covariance matrix; then, obtaining the latent spatial anomaly of the target satellite time series data based on the target potential spatial vector, the mean vector, and the covariance matrix includes: obtaining the latent spatial anomaly of the target satellite time series data based on the target potential spatial vector, the mean vector, and the optimized covariance matrix.
[0013] Preferably, the optimized covariance matrix is obtained using the following formula:
[0014] in, The optimized covariance matrix is... Let be the covariance matrix. As a preset positive number, It is the identity matrix. This constitutes the disturbance.
[0015] The specific technical solution of the second embodiment of the present invention is as follows: a fire point detection system based on satellite data, the system comprising: a first data acquisition module, an encoding module, a calculation module, a second data acquisition module, an anomaly calculation module, and a classification module; the first data acquisition module is used to acquire multiple satellite time-series data of fire-free areas; the encoding module is used to input the satellite time-series data into a preset encoder to obtain a latent spatial vector for each satellite time-series data; the calculation module is used to obtain a mean vector between the latent spatial vectors, and to obtain a covariance matrix between the latent spatial vectors based on the mean vector and the latent spatial vectors; the second data acquisition module is used to acquire target satellite time-series data of a target point to be detected, and to obtain a target latent spatial vector of the target satellite time-series data using the preset encoder; the anomaly calculation module is used to obtain a latent spatial anomaly of the target satellite time-series data based on the target latent spatial vector, the mean vector, and the covariance matrix; the classification module is used to input the latent spatial anomaly into a preset classifier to obtain a fire point detection result for the target point to be detected; the fire point detection result includes fire points and fire-free areas.
[0016] The specific technical solution of the third embodiment of the present invention is as follows: a fire detection device based on satellite data, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.
[0017] The specific technical solution of the fourth embodiment of the present invention is as follows: a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor performs the steps of any one of the methods described in the first embodiment of the present application.
[0018] Implementing the embodiments of the present invention will have the following beneficial effects: This invention acquires multiple satellite time-series data when there are no fire points, obtains a latent spatial vector through a preset encoder, and calculates the mean vector and covariance matrix. The covariance matrix can describe the dispersion of data in different directions, and can more accurately characterize the shape of the no-fire point sample cluster, no longer limited to the hypersphere assumption. For the target point to be detected, after acquiring its target satellite time-series data and target latent spatial vector, the latent spatial anomaly degree is obtained by combining the mean vector and covariance matrix. This method of obtaining latent spatial anomaly degree by combining the covariance matrix fully considers the actual distribution pattern of no-fire point samples in the latent space, thereby obtaining an accurate latent spatial anomaly degree. Inputting the latent spatial anomaly degree into a preset classifier can effectively improve the accuracy of fire point detection and avoid misjudgment of anomalies due to incorrect assumptions. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating the steps of a fire detection method based on satellite data. Figure 2 This is a schematic diagram illustrating the existing method for calculating Euclidean distance. Figure 3 This is a diagram illustrating the implementation effect of Euclidean distance in existing technologies. Figure 4 This is a schematic diagram illustrating the calculation method of Mahalanobis distance in this invention; Figure 5 This is a diagram illustrating the implementation effect of the Mahalanobis distance in this invention; Figure 6 Flowchart for the offline modeling phase; Figure 7 This is a flowchart of the online testing phase. Figure 8 This is a schematic diagram illustrating the calculation method based on the covariance matrix; Figure 9 This is a schematic diagram of the ROC comparison curves; Figure 10 This is a schematic diagram of a fire detection system based on satellite data. Among them, 201 is the first data acquisition module; 202 is the encoding module; 203 is the calculation module; 204 is the second data acquisition module; 205 is the anomaly calculation module; and 206 is the classification module. Detailed Implementation
[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0022] The terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such processes, methods, products, or apparatus.
[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0024] Please see Figure 1 This is a flowchart illustrating the steps of a fire detection method based on satellite data in the first embodiment of this application, aimed at effectively improving the accuracy of fire detection. The method includes: Step 101: Obtain multiple satellite time series data for areas without fire points; Step 102: Input the satellite time series data into a preset encoder to obtain the potential spatial vector of each satellite time series data; Step 103: Obtain the mean vector between the latent space vectors, and obtain the covariance matrix between the latent space vectors based on the mean vector and the latent space vectors; Step 104: Obtain the target satellite time series data of the target point to be detected, and use the preset encoder to obtain the target potential spatial vector of the target satellite time series data; Step 105: Obtain the latent spatial anomaly degree of the target satellite time series data based on the target latent spatial vector, the mean vector, and the covariance matrix; Step 106: Input the potential spatial anomaly degree into a preset classifier to obtain the fire detection result of the target point to be detected; the fire detection result includes fire points and no fire points.
[0025] Specifically, a specific fire-free area is selected, and multiple satellite time-series data of this area are continuously acquired over a period of time using multiple satellites, covering information from different time periods and different bands. The acquired satellite time-series data is input into a pre-set deep neural network encoder, which performs feature extraction and dimensionality reduction on the data to obtain the latent spatial vector corresponding to each satellite time-series data. The mean vector of all latent spatial vectors is calculated, which represents the average feature of the data in the fire-free area. Then, based on the mean vector and each latent spatial vector, the covariance matrix between the latent spatial vectors is calculated using a specific formula. This matrix reflects the correlation and dispersion between the data. The target satellite time-series data of the target point to be detected is acquired, and its target latent spatial vector is also obtained using the pre-set encoder. Based on the target latent spatial vector, mean vector, and covariance matrix, the latent spatial anomaly of the target satellite time-series data is calculated using methods such as Mahalanobis distance. The latent spatial anomaly is input into a pre-set classifier, which determines whether the target point to be detected is a fire point or a fire-free point according to a pre-set threshold and classification rules, and outputs the corresponding fire detection result.
[0026] The method in this embodiment acquires multiple satellite time-series data when there are no fire points, obtains a latent spatial vector through a preset encoder, and calculates the mean vector and covariance matrix. The covariance matrix can describe the dispersion of data in different directions, and can more accurately characterize the shape of the no-fire point sample cluster, no longer limited to the hypersphere assumption. For the target point to be detected, after acquiring its target satellite time-series data and target latent spatial vector, the latent spatial anomaly degree is obtained by combining the mean vector and covariance matrix. This method of obtaining latent spatial anomaly degree by combining the covariance matrix fully considers the actual distribution pattern of no-fire point samples in the latent space, thereby obtaining an accurate latent spatial anomaly degree. Inputting the latent spatial anomaly degree into a preset classifier can effectively improve the accuracy of fire point detection and avoid misjudgment of anomalies due to incorrect assumptions.
[0027] In a specific embodiment, obtaining the mean vector among the potential spatial vectors includes: summing all the potential spatial vectors and dividing the sum by the total number of satellite time series data to obtain the mean vector.
[0028] Specifically, the mean vector is obtained using the following formula:
[0029] in, Let be the mean vector of all fire-free samples in the latent space. This represents the total number of fire-free samples. i For the sample index, traverse from 1 to... ; It represents the latent space vector of the i-th fireless sample, that is, the low-dimensional feature vector obtained after encoding a single fireless sample by an autoencoder.
[0030] In a specific embodiment, the covariance matrix is obtained using the following formula:
[0031] in, Let be the covariance matrix. The total number of satellite time-series data. Let be the latent spatial vector of the i-th satellite time series data. Let the mean vector be... This is the transpose of a vector. Specifically, the covariance matrix quantifies the degree of variation in different dimensions of satellite time-series data and the correlation between them. By calculating the mean of the product of the deviations of each latent spatial vector and the mean vector, we can understand the dispersion of the data in various directions and the degree of correlation between data in different directions.
[0032] In a specific embodiment, the latent space anomaly degree is obtained using the following formula:
[0033] in, The degree of anomaly in the latent space. Let be the target potential space vector. Let the mean vector be... This is the inverse of the covariance matrix, and it serves to "correct" the distance based on the data distribution pattern. This is a vector transpose operation. Specifically, by comprehensively considering the deviation between the target latent space vector and the mean vector, and combining it with the inverse of the covariance matrix, this calculation method can accurately quantify the degree of anomaly of the target in the latent space. It not only considers the simple distance between the target and the mean, but also incorporates the correlation information between various dimensions of the data, i.e., the covariance matrix, making the measurement of anomaly more comprehensive and accurate. The essence of Mahalanobis distance is to use the inverse of the covariance matrix before calculating the distance. A "correction" (rotation and scaling) is performed on the coordinate space. This correction straightens and stretches the originally tilted "ellipsoidal" data distribution into a standard "unit sphere." Calculating Euclidean distance in this corrected new space is equivalent to calculating Mahalanobis distance in the original space. Therefore, Mahalanobis distance is a distance that eliminates the effects of inter-dimensional correlation and scale differences, truly reflecting the "degree of deviation" of a point relative to a data cluster.
[0034] In a specific embodiment, after obtaining the target potential spatial vector of the target satellite time series data using the preset encoder, the method further includes: obtaining the reconstruction error when the encoder obtains the target potential spatial vector; then, inputting the potential spatial anomaly degree into the preset classifier to obtain the fire detection result of the target point to be detected includes: inputting the potential spatial anomaly degree and the reconstruction error into the preset classifier to obtain the fire detection result of the target point to be detected.
[0035] Specifically, the latent spatial vector of the target point to be detected is first extracted by the encoder, and the Mahalanobis distance between the vector and the mean vector is calculated as the latent spatial anomaly. At the same time, the reconstruction error after the input data is reconstructed by the decoder is calculated. The latent spatial anomaly and the reconstruction error are jointly input into a preset random forest classifier. The classifier learns the association rules between the two and fire points based on the training data, and finally outputs a binary classification result of whether the target point to be detected is a fire point, thus achieving efficient and accurate fire point detection.
[0036] Covariance matrix singularity handling: When the number of training samples is insufficient or the latent space dimension is too high, the calculated covariance matrix may be close to singular (non-invertible), making it impossible to calculate the covariance matrix. This invention uses a regularization method to solve this problem: after obtaining the covariance matrix between the latent space vectors, a perturbation is added to the covariance matrix to obtain an optimized covariance matrix; then, the step of obtaining the latent space anomaly degree of the target satellite time series data based on the target latent space vector, the mean vector, and the covariance matrix includes: obtaining the latent space anomaly degree of the target satellite time series data based on the target latent space vector, the mean vector, and the optimized covariance matrix.
[0037] In practical implementation, the optimized covariance matrix is obtained using the following formula:
[0038] in, The optimized covariance matrix is... Let be the covariance matrix. As a preset positive number, such as It is a very small positive number (e.g.) ), It is the identity matrix. This constitutes the disturbance.
[0039] For high-dimensional latent spaces (e.g., dimension) For datasets with large amounts of data, calculating and inverting the covariance matrix can be time-consuming. This invention can optimize this process using the following methods: Incremental update: When modeling offline, an incremental covariance update algorithm is used to process data in batches, avoiding loading all data into memory at once.
[0040] Cholesky decomposition: When calculating Mahalanobis distance, the covariance matrix... Perform Cholesky decomposition ( , (where covariance matrix is used). Solving a system of linear equations instead of directly inverting the matrix results in a more stable numerical solution and faster computation speed.
[0041] Principal Component Dimensionality Reduction: A principal component analysis (PCA) can be selectively performed on the latent space to reduce dimensionality, retaining more than 99% of the energy, and calculating covariance and Mahalanobis distance at a lower dimension, significantly reducing computational complexity.
[0042] c) Enhanced dynamic adaptability: To adapt to the slow changes in the surface environment due to seasonal factors, the following can be adopted: Sliding window update: Periodically (e.g., monthly) using the latest normal data (fire point data), the mean vector and covariance matrix are updated through a sliding window mechanism to allow the model to adapt to the slow drift of the data distribution.
[0043] Time decay factor: When calculating a statistical model, more recent normal samples are given higher weights, making the model more sensitive to the latest "normal" state.
[0044] In specific embodiments, to illustrate the beneficial effects of this application, two comparative methods are used to compare the specific implementation effects: a) A baseline model using Euclidean distance; b) The Mahalanobis distance model of the present invention is adopted.
[0045] Experimental results: On a test set containing a large number of exposed rock masses (which are prone to false alarms of thermal anomalies), method a achieved a precision of 0.75 and a recall of 0.88; method b improved its precision to 0.85 while maintaining a stable recall of 0.87. ROC curve comparisons show that method b's AUC value increased from 0.89 to 0.94, demonstrating a significant enhancement in its overall discriminative ability.
[0046] For a detailed classification of Euclidean distance in the existing technology, please refer to [link / reference needed]. Figure 2 and Figure 3 . Figure 2 This demonstrates the limitations of existing Euclidean distance technology. Figure 3 The point cloud in the image represents the actual distribution of "normal samples" in the latent space. It can be seen that it is an elliptical distribution, not a circle. Figure 3 Centroid: This is the center point of all normal samples. Figure 3Dashed concentric circles: These are the equidistant lines of the Euclidean distance. Euclidean distance assumes that all points equidistant from the centroid lie on a circle. Figure 3 Point P1: Located in the elliptical distribution, it is close to the minor axis but deviates from the main direction. Since d1 is small, the Euclidean distance considers point P1 to be normal because it is within the circular threshold. Figure 3 Point P2: Located at a relatively far point along the major axis of the elliptical distribution. Due to the large d2, the Euclidean distance considers point P2 to be an anomaly because it exceeds the circular threshold.
[0047] For details regarding the classification of Mahalanobis distance in this application, please refer to [link / reference needed]. Figure 4 and Figure 5 . Figure 4 This demonstrates the advantages of the Mahalanobis distance in this invention. Figure 5 Midpoint cloud: This represents the actual distribution of "normal samples" in the latent space. It can be seen that it is an elliptical distribution, not a circle. Figure 5 Centroid: This is the center point of all normal samples. Figure 5 The dashed concentric circles represent the equidistant lines of the Mahalanobis distance. By learning the covariance (i.e., shape and orientation) of a normal data distribution, Mahalanobis distance generates a discrimination boundary consistent with the shape of the data distribution. This allows it to more accurately determine whether a point is a true outlier (deviation from the distribution direction) or merely on the edge of the normal distribution (along the distribution direction). Figure 5 Point P1: According to Mahalanobis distance, although it is close to the centroid line, it deviates from the main direction of the normal data distribution. Mahalanobis distance considers it to be inconsistent with the normal distribution pattern, and therefore correctly judges it as "abnormal". Figure 5 Point P2: According to Mahalanobis distance, although it is far from the centroid line, it is located in the main direction of the normal data distribution. Mahalanobis distance considers it to be consistent with the normal distribution pattern, so it will correctly judge it as "normal".
[0048] Figure 6 This is a detailed flowchart (including optimization strategies) of the offline modeling stage of the present invention. Figure 7 This is a detailed flowchart of the online detection stage of the present invention, wherein, Let covariance matrix be the variance matrix. This is the optimized covariance matrix. Figure 8 This is a schematic diagram illustrating the calculation method based on the covariance matrix in this invention. Figure 9 ROC curves for different technical solutions are shown. Based on the above implementation scheme, the present invention has the following significant advantages: 1. Achieves adaptive data distribution and more accurate measurement: By introducing the covariance matrix and Mahalanobis distance, this invention enables anomaly measurement to "understand" the distribution pattern of normal data, and is no longer affected by different scales and correlations between dimensions, resulting in more scientific and accurate measurement results.
[0049] 2. Significantly improves model robustness and reduces false positives and false negatives: As described in the background analysis, this invention effectively avoids false positives and false negatives caused by uneven data distribution, especially when dealing with complex and marginal "difficult samples," demonstrating stronger robustness. Preliminary comparative experiments show that after adopting this invention, the F1 score on specific difficult sample sets can be improved by about 15%, and the false positive rate can be reduced by more than 40%.
[0050] 3. The technical solution has universality: This method is not only applicable to fire detection, but its core idea (modeling the covariance of the latent space and using Mahalanobis distance) can be widely applied to any time-series or image anomaly detection task based on autoencoders, and has strong technical promotion value.
[0051] 4. Comprehensive engineering implementation considerations: This solution fully considers practical engineering problems such as the singularity of the covariance matrix, computational efficiency, and dynamic adaptability, and proposes clear solutions, making it highly feasible.
[0052] In a specific embodiment, the second embodiment of this application provides a schematic diagram of a fire detection system based on satellite data. The system includes: a first data acquisition module 201, an encoding module 202, a calculation module 203, a second data acquisition module 204, an anomaly calculation module 205, and a classification module 206. The first data acquisition module 201 is used to acquire multiple satellite time-series data of fire-free areas. The encoding module 202 is used to input the satellite time-series data into a preset encoder to obtain a latent spatial vector for each satellite time-series data. The calculation module 203 is used to obtain the mean vector among the latent spatial vectors and, based on the mean... The covariance matrix between the latent space vector and the vector is obtained; the second data acquisition module 204 is used to acquire the target satellite time series data of the target point to be detected, and to acquire the target latent space vector of the target satellite time series data using the preset encoder; the anomaly calculation module 205 is used to acquire the latent space anomaly of the target satellite time series data according to the target latent space vector, the mean vector and the covariance matrix; the classification module 206 is used to input the latent space anomaly into the preset classifier to obtain the fire point detection result of the target point to be detected; the fire point detection result includes fire point and no fire point.
[0053] In this embodiment, the system acquires multiple satellite time-series data when there are no fire points, obtains a latent spatial vector through a preset encoder, and calculates the mean vector and covariance matrix. The covariance matrix can describe the dispersion of data in different directions, and can more accurately characterize the shape of the no-fire point sample cluster, no longer limited to the hypersphere assumption. For the target point to be detected, after acquiring its target satellite time-series data and target latent spatial vector, the latent spatial anomaly is obtained by combining the mean vector and covariance matrix. This method of obtaining latent spatial anomaly by combining the covariance matrix fully considers the actual distribution pattern of no-fire point samples in the latent space, thereby obtaining an accurate latent spatial anomaly. Inputting the latent spatial anomaly into a preset classifier can effectively improve the accuracy of fire point detection and avoid misjudgment of anomalies due to incorrect assumptions.
[0054] In a specific embodiment, the third embodiment of this application provides a fire detection device based on satellite data, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.
[0055] In a specific embodiment, the fourth embodiment of this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method as described in any one of the first embodiments of this application.
[0056] The above embodiments merely illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A fire detection method based on satellite data, characterized in that, The method includes: Acquire time-series satellite data from areas without fire points; The satellite time series data is input into a preset encoder to obtain the potential spatial vector of each satellite time series data. Obtain the mean vector among the latent space vectors, and obtain the covariance matrix among the latent space vectors based on the mean vector and the latent space vectors; Acquire target satellite time series data of the target point to be detected, and use the preset encoder to acquire the target potential spatial vector of the target satellite time series data; The latent spatial anomaly degree of the target satellite time series data is obtained based on the target latent spatial vector, the mean vector, and the covariance matrix; The potential spatial anomaly degree is input into a preset classifier to obtain the fire detection result of the target point to be detected; the fire detection result includes fire points and no fire points.
2. The fire detection method based on satellite data as described in claim 1, characterized in that, The step of obtaining the mean vector among the potential space vectors includes: The mean vector is obtained by summing all the potential spatial vectors and dividing the sum by the total number of satellite time series data.
3. The fire detection method based on satellite data as described in claim 1, characterized in that, The covariance matrix is obtained using the following formula: in, Let be the covariance matrix. The total number of satellite time-series data. Let be the latent spatial vector of the i-th satellite time series data. Let the mean vector be... This is the transpose operation for a vector.
4. The fire detection method based on satellite data as described in claim 1, characterized in that, The latent space anomaly degree is obtained using the following formula: in, The degree of anomaly in the latent space. Let be the target potential space vector. Let the mean vector be... Let be the inverse of the covariance matrix. This is the transpose operation for a vector.
5. The fire detection method based on satellite data as described in claim 1, characterized in that, After obtaining the target potential spatial vector of the target satellite time series data using the preset encoder, the method further includes: The reconstruction error of the encoder when acquiring the target latent space vector is obtained; The step of inputting the latent spatial anomaly degree into a preset classifier to obtain the fire detection result of the target point to be detected includes: The potential spatial anomaly degree and the reconstruction error are input into a preset classifier to obtain the fire detection result of the target point to be detected.
6. The fire detection method based on satellite data as described in claim 1, characterized in that, After obtaining the covariance matrix between the latent space vectors, the method further includes: By adding a perturbation to the covariance matrix, an optimized covariance matrix is obtained. The step of obtaining the latent spatial anomaly degree of the target satellite time series data based on the target latent spatial vector, the mean vector, and the covariance matrix includes: The latent spatial anomaly degree of the target satellite time series data is obtained based on the target latent spatial vector, the mean vector, and the optimized covariance matrix.
7. The fire detection method based on satellite data as described in claim 6, characterized in that, The optimized covariance matrix is obtained using the following formula: in, The optimized covariance matrix is... Let be the covariance matrix. As a preset positive number, It is the identity matrix. This constitutes the disturbance.
8. A fire detection system based on satellite data, characterized in that, The system includes: a first data acquisition module, an encoding module, a calculation module, a second data acquisition module, an anomaly calculation module, and a classification module; The first data acquisition module is used to acquire multiple satellite time-series data of areas without fire points; The encoding module is used to input the satellite time series data into a preset encoder to obtain the potential spatial vector of each satellite time series data. The calculation module is used to obtain the mean vector between the latent space vectors, and to obtain the covariance matrix between the latent space vectors based on the mean vector and the latent space vectors; The second data acquisition module is used to acquire the target satellite time series data of the target point to be detected, and to acquire the target potential spatial vector of the target satellite time series data using the preset encoder; The anomaly calculation module is used to obtain the latent spatial anomaly of the target satellite time series data based on the target latent spatial vector, the mean vector and the covariance matrix. The classification module is used to input the potential spatial anomaly degree into a preset classifier to obtain the fire detection result of the target point to be detected; the fire detection result includes fire points and no fire points.
9. A fire detection device based on satellite data, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the processor performs the steps of the method as described in any one of claims 1-7.