A three-dimensional monitoring and early warning method, system, terminal and medium for forestry disasters

Abstract

The present invention discloses a three-dimensional monitoring and early warning method, system, terminal and medium for forestry disasters, and belongs to the technical field of forestry disaster monitoring. In view of the problems of low efficiency, insufficient precision and poor practicality of existing forestry disaster monitoring, the present invention proposes a solution integrating high-resolution remote sensing images and artificial intelligence technology. The method first uses satellites and drones to obtain multi-perspective, multi-level high-resolution images of the target forest; then, through intelligent preprocessing, multi-dimensional feature extraction, refined similarity analysis and other steps, it automatically identifies abnormal forest areas and distinguishes their disaster types; finally, it generates a disaster area distribution map, and extracts key disaster information to serve disaster early warning and forestry management decision-making. The present invention significantly improves the timeliness, accuracy and practicality of forestry disaster monitoring, and provides new ideas and methods for precise and intelligent forestry disaster prevention and control.

Description

Technical Field

[0001] The present invention relates to the field of forestry disaster monitoring, and in particular to a three-dimensional monitoring and early warning method, system, terminal and medium for forestry disasters. Background Art

[0002] Forestry plays an important role in maintaining ecological balance and protecting biodiversity. However, due to factors such as climate change and human activities, various forest disasters occur frequently, posing huge challenges to the sustainable development of forestry. Traditional forest disaster monitoring mainly relies on manual field surveys and conventional remote sensing image interpretation, which has problems such as low efficiency, high cost, and poor timeliness, and it is difficult to meet the needs of precise prevention and control. In recent years, with the rapid development of high-resolution remote sensing, artificial intelligence and other technologies, it has provided a new opportunity for the realization of intelligent monitoring and early warning of forestry disasters.

[0003] The core of high-resolution remote sensing technology is to obtain remote sensing image data with a spatial resolution of meters or even sub-meters. Compared with traditional medium- and low-resolution satellite images (such as the Landsat series, with a spatial resolution of 30 meters), high-resolution images (such as WorldView-3, with a spatial resolution better than 0.5 meters) can provide richer detailed information such as the texture and structure of objects, providing a data basis for refined disaster identification. In addition, high-resolution images usually have higher spectral resolution and can provide more spectral channels, which helps to distinguish different types of objects.

[0004] In terms of data analysis, artificial intelligence technologies such as deep learning provide new ideas for intelligent interpretation of remote sensing images. Traditional image interpretation mainly relies on manual visual interpretation, which has problems such as strong subjectivity and low efficiency. Deep learning can automatically learn multi-scale and hierarchical features in images by building a multi-layer neural network, and establish a mapping relationship between features and different types of objects, thus achieving end-to-end image classification and target detection.

[0005] However, the application of intelligent analysis of high-resolution remote sensing data in the field of forestry disaster monitoring is not mature and systematic enough. Existing research mainly focuses on the detection of a single data source and a single type of disaster, and pays insufficient attention to the preprocessing and noise elimination of remote sensing data, which affects the reliability of monitoring. In addition, the integration and utilization of data from different sources and different spatiotemporal scales needs to be strengthened. There is still room for improvement in the refined extraction, intuitive expression and business application of disaster information. Comprehensively utilizing high-resolution remote sensing big data and artificial intelligence algorithms to build a practical three-dimensional monitoring and early warning solution for forestry disasters is of great significance to improving the level of intelligence and refinement in forestry disaster prevention and control. Summary of the invention

[0006] Aiming at the problems of low efficiency, insufficient precision and poor practicability of existing forestry disaster monitoring methods, the present invention proposes a three-dimensional monitoring and early warning method, system, terminal and medium for forestry disasters.

[0007] Traditional forestry disaster monitoring mainly relies on manual field surveys and visual interpretation of satellite images, which has the disadvantages of untimely response, heavy workload, and strong subjectivity, making it difficult to meet the current needs of fine disaster management. In response to the above technical problems, the present invention innovatively proposes the following solutions:

[0008] In a first aspect, the present invention provides a three-dimensional monitoring and early warning method for forestry disasters, comprising:

[0009] Step S01 collects high-resolution remote sensing image data of the target forest;

[0010] Step S02 preprocesses the remote sensing image data to obtain non-dispersive remote sensing image data, and obtains the confidence probability corresponding to the non-dispersive remote sensing image data;

[0011] Step S03 determines the remote sensing image data collected in the target forest area as the first non-dispersive remote sensing image data, calculates the corresponding characteristic factors between the first non-dispersive remote sensing image data according to the confidence probability, extracts abnormal key characteristic indicators for different disaster types, and uses the extracted abnormal features of each dimension to form a first characteristic vector ; Determine the historical remote sensing image data of the target forest area as the second non-dispersive remote sensing image data, and extract the disaster key features of each dimension from the second non-dispersive remote sensing image data to form a second feature vector ;

[0012] Step S04 designs a similarity classification function based on the first feature vector and the second feature vector: ,in, is the kernel function parameter, which is used to control the similarity decay rate. The larger the value, the faster the similarity decays with increasing distance; is the feature weight matrix, which is used to balance the importance of different features; represents the difference of eigenvectors, represents transpose;

[0013] The similarity classification function is optimized through the classification learning algorithm, so that it can accurately measure whether the target forest area has experienced the same type of disaster as in the history of the target forest area: ,in is the optimal kernel function parameter, is the optimal feature weight matrix;

[0014] Step S05 performs disaster area detection on the first non-dissipative remote sensing image data, represents the area to be detected in the window as a feature vector, uses the similarity classification function to determine whether the area belongs to a certain type of known disaster, and calculates its disaster degree value;

[0015] Step S06: outlining the boundaries of the identified disaster area, generating a mapping vector boundary and a disaster area distribution map, and calculating geometric parameters of the disaster area;

[0016] Step S07 converts the monitoring results of the disaster area distribution map, vector boundaries, and geometric parameters into information products and services that can be used for business analysis and decision support.

[0017] Furthermore, the preprocessing of the remote sensing image data includes:

[0018] Performing radiometric calibration, geometric correction, image matching and image enhancement on the remote sensing image data;

[0019] The confidence probability corresponding to the non-dispersed remote sensing image data is calculated using a similarity evaluation algorithm based on local contrast and gradient direction histogram.

[0020] Furthermore, the extracting of feature vectors from anomaly key features and disaster key features includes: extracting anomaly key feature indicators and disaster key feature indicators from spectral features, texture features, time features and spatial features, wherein the spectral features include hotspot anomaly index and smoke index, and the spatial features include vegetation damage index and landscape fragmentation.

[0021] Furthermore, the first feature vector The calculation formula is:

[0022]

[0023]

[0024]

[0025] in, Indicates The characteristic value of the abnormal key feature, Indicates The number of features extracted from the abnormal key features; For the The weight coefficient of each feature; For the The feature extraction function takes image as input. and confidence , the output is the quantized value of the feature; is a feature parameter set.

[0026] Furthermore, the disaster area detection is performed on the first non-dissipative remote sensing image data, including:

[0027] a. Divide the first non-dispersive remote sensing image data into sliding windows according to the preset step size and window size to obtain a number of sub-areas, and the total number of sub-areas D is: , where H and 𝑊 represent the height and width of the first non-dispersive remote sensing image data, ℎ and 𝑤 represent the height and width of the sliding window, and s represents the step size. Indicates rounding down;

[0028] b. Calculate the similarity score between each sub-region and the known disaster area , assuming that the first eigenvectors of the r known abnormal feature regions are , then Similarity score of each sub-region for: ,in, Indicates the sub-area number currently being detected. represents the total number of known disaster samples, Indicates The score that is most similar to the sub-region among all known disaster samples, The sub-region to be detected The characteristic vector of is the characteristic vector of the known r-th disaster sample area, Indicates calculating the similarity between two feature vectors;

[0029] c. Score the similarity Converted into disaster probability value ;

[0030] d. The disaster probability value of the sub-region Merge into a complete disaster area distribution map.

[0031] Furthermore, the similarity score Converted into disaster probability value The calculation formula is: ,in, and is a parameter that controls the shape of the probability distribution.

[0032] Furthermore, outlining the boundaries of the identified disaster area includes:

[0033] a. Perform threshold segmentation on the disaster area distribution map to obtain the first non-dispersive remote sensing image segmentation map of the disaster area and the non-disaster area;

[0034] b. Performing a connected domain analysis on the first non-dispersive remote sensing image segmentation map to mark the disaster areas that are not connected to each other;

[0035] c. Filter out false positive areas based on geometric properties such as the area of ​​the connected domain and the aspect ratio of the circumscribed rectangle;

[0036] d. Extract the coordinates of the boundary pixels of the connected domains of each disaster area;

[0037] e. Calculate the geometric parameters of each disaster area.

[0038] In a second aspect, the present invention provides a three-dimensional monitoring and early warning system for forestry disasters, comprising:

[0039] a. Data acquisition module, used to collect high-resolution remote sensing image data of the target forest;

[0040] b. a data preprocessing module, for preprocessing the remote sensing image data to obtain non-dispersed remote sensing image data and obtain the confidence probability corresponding to the non-dispersed remote sensing image data;

[0041] c. A feature extraction module, which is used to calculate the corresponding feature factors between the first non-dispersive remote sensing image data according to the confidence probability, extract abnormal key feature indicators for different disaster types, and use the extracted abnormal features of each dimension to form a first feature vector; and extract the disaster key features of each dimension from the second non-dispersive remote sensing image data to form a second feature vector;

[0042] d. a data classification module, for designing a similarity classification function based on the first feature vector and the second feature vector, and optimizing the function through a classification learning algorithm;

[0043] e. A disaster area detection module, which is used to detect disaster areas on the first non-dissipative remote sensing image data to determine whether the area to be tested belongs to a certain type of known disaster;

[0044] f. A disaster boundary delineation module, which is used to delineate the boundaries of the identified disaster areas and generate mapping vector boundaries and disaster area distribution maps;

[0045] g. Analysis and decision-making module, which is used to convert monitoring results into information products and services that can be used for business analysis and decision support.

[0046] In a third aspect, the present invention provides a terminal device, comprising:

[0047] a. Processor;

[0048] b. a memory for storing computer programs;

[0049] Wherein, when the processor executes the computer program, any one of the three-dimensional monitoring and early warning methods for forestry disasters provided in the specification of the present invention is implemented.

[0050] In a fourth aspect, the present invention proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements any one of the three-dimensional monitoring and early warning methods for forestry disasters provided in the specification of the present invention.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. Three-dimensional data collection. Combining high-resolution satellite remote sensing and drone aerial photography, forest image data is acquired from multiple angles and levels, providing rich key information such as ground texture and structure for subsequent analysis.

[0053] 2. Intelligent processing flow: Through a series of algorithms such as data preprocessing, feature extraction, and similarity analysis, abnormal areas in forest images can be automatically identified and judged whether they are specific types of disasters, reducing the subjectivity and uncertainty of manual interpretation.

[0054] 3. Multidimensional feature extraction. From the perspectives of abnormal features and disaster features, key indicators that can sensitively reflect different types of disasters are extracted and quantified through feature vectors, providing more comprehensive and accurate criteria for disaster identification.

[0055] 4. Refinement of similarity analysis. Design a similarity measurement function to accurately measure the similarity between the area to be tested and the known disaster area by optimizing the feature subspace and distance measurement, thereby improving the reliability of disaster identification.

[0056] 5. Accurate description of disaster areas. Delineate the boundaries and extract attributes of the identified disaster areas, generate vector boundaries and distribution maps, and calculate geometric parameters such as the affected area and fragmentation, to present the disaster situation intuitively and quantitatively.

[0057] 6. Business-oriented monitoring results: Timely release disaster warnings, generate thematic maps of disaster areas, conduct disaster change detection, establish a data sharing platform, and transform monitoring results into information products that support forestry management and scientific research decision-making.

[0058] In summary, the present invention has built a full-process, integrated forestry disaster monitoring and early warning solution from data acquisition, disaster identification to information services. It gives full play to the spatial information advantages of high-resolution remote sensing and the data analysis advantages of artificial intelligence, significantly improving the efficiency, accuracy and practicality of forestry disaster monitoring, and providing new ideas and methods for accurate disaster prevention and intelligent management and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art 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 paying creative labor.

[0060] Figure 1 : Overall steps of the three-dimensional monitoring and early warning method for forestry disasters.

[0061] Figure 2 : Structural diagram of the three-dimensional monitoring and early warning system for forestry disasters.

[0062] Figure 3 : Schematic diagram of the structure of the terminal equipment. DETAILED DESCRIPTION

[0063] Figure 1 This is a general step diagram of the three-dimensional monitoring and early warning method for forestry disasters of the present invention. Figure 1 As shown, the three-dimensional monitoring and early warning method for forestry disasters includes steps S01 to S07.

[0064] Step S01, through field survey of the target forest area, high-resolution satellite remote sensing images and drone aerial images are used to collect high-resolution remote sensing image data of the target forest.

[0065] For example, field investigation and data collection are used to obtain high-resolution remote sensing image data of the monitoring area to provide a data source for subsequent analysis. The main methods include field investigation and remote sensing data acquisition. Through field investigation, the background information such as the topography, vegetation conditions, and disaster types of the monitoring area can be understood to provide prior knowledge for the interpretation and analysis of remote sensing data.

[0066] For example, high-resolution satellite remote sensing images (such as WorldView-3 satellite images, with a spatial resolution better than 0.5 meters) and drone aerial images (such as DJI Phantom 4 RTK drones, with a spatial resolution better than 0.1 meters) are used to collect high-resolution remote sensing image data of the target forest. During the data collection process, it is necessary to optimize the selection and parameter setting of satellite remote sensing images and drone aerial images. For satellite remote sensing images, select clear and cloudless images with good atmospheric transparency, and select appropriate spectral bands (such as near-infrared and short-wave infrared bands for forest fires) based on the seasonal characteristics of the disaster. For drone aerial images, the route and altitude are reasonably planned according to the actual situation of the disaster area. Usually, a relative flight altitude of 50-100 meters is selected, and the lens angle is perpendicular to the ground.

[0067] The present invention does not impose any specific restrictions on the number of drones and the type of high-resolution satellite remote sensing image data, and users can choose according to actual needs.

[0068] Step S02, preprocessing the collected remote sensing image data to obtain non-dispersive remote sensing image data, and obtaining the confidence probability corresponding to the non-dispersive remote sensing image data.

[0069] The original remote sensing image data is collected through remote sensing sensors. Depending on the specific application requirements and sensor types, the collected data may include: optical images in the visible light (0.4-0.7μm), near infrared (0.7-1.1μm) and mid-infrared (1.1-3μm) bands; hyperspectral images (usually containing dozens to hundreds of bands) containing the reflectivity or radiance curves of surface materials at different wavelengths; and infrared images that reflect the thermal radiation characteristics of ground objects (such as thermal infrared bands, 8-14μm), etc. The original remote sensing image data is preprocessed, mainly including radiation calibration, geometric correction, image matching, image enhancement and other steps. The preprocessing step is to eliminate noise, distortion and other influencing factors in the image, improve data quality and availability, and obtain non-dispersive remote sensing image data.

[0070] Exemplarily, a rule or a data statistical method is used to calculate the confidence of the preprocessed non-dispersed remote sensing data, such as a similarity evaluation algorithm based on local contrast and gradient direction histograms, to calculate the confidence probability corresponding to the non-dispersed remote sensing image data. The image data is divided into a series of local windows, and a feature vector based on the similarity of local contrast and gradient direction histograms is constructed for each window. A similarity threshold is set, and the proportion of all windows exceeding the threshold is counted as the confidence probability of the entire image, with a value range of [0, 1]. The higher the confidence probability, the better the non-dispersed degree of the image and the higher the credibility. The preprocessed non-dispersed remote sensing image data and its confidence probability are output together as input data for subsequent applications such as remote sensing image feature extraction. At the same time, the confidence probability can be used to guide parameter selection, algorithm optimization and uncertainty analysis in subsequent processing, in order to obtain more robust and reliable results.

[0071] Step S03, according to the confidence probability, calculate the corresponding characteristic factors between the first non-dispersive remote sensing image data, and for different disaster types, extract abnormal key characteristic indicators that can describe this type of disaster from the perspectives of spectrum, texture, time, space, etc., and use the extracted abnormal features of each dimension to form a first feature vector; the key disaster features of each dimension extracted from the second non-dispersive remote sensing image data constitute a second feature vector.

[0072] It should be noted that the number of remote sensing images collected for the target forest area is the first non-dispersed remote sensing image data; historical remote sensing image data related to the target forest area are collected, and these data may include disaster and non-disaster data obtained from various surveys, monitoring or research on the forest previously conducted, and these data contain disaster key features similar to abnormal key features in the remote sensing sensor data. The historical remote sensing image data of the target forest area is determined to be the second non-dispersed remote sensing data.

[0073] For different types of disasters, we extract abnormal key characteristic indicators that can sensitively identify such disasters from the perspectives of spectrum, texture, time, and space. Taking forest fire as an example, we extract spectral features such as hot spot anomaly index and smoke index, spatial features such as vegetation damage index and landscape fragmentation, as well as mutation and periodic features in time series. Finally, we summarize the abnormal key features of each dimension in the first non-dissipative remote sensing image data into the first feature vector; similarly, through feature extraction, we summarize the disaster key features of each dimension in the second non-dissipative remote sensing image data into the second feature vector.

[0074] For example, the role of this step is to identify and characterize the typical features of different types of forestry disasters from remote sensing images, and provide a basis for automatic disaster identification. The main methods include spectral feature analysis, texture feature analysis, time series analysis, etc. For forest fires, extract high temperature abnormal signals in hotspot areas, such as sudden increases in near-infrared and short-wave infrared reflectivity, and output hotspot abnormality index and smoke index ; For forest pests and diseases, extract the spectral response anomaly of the target forest area, such as the decrease of green light band reflectivity and the increase of red light band reflectivity, and output the pest anomaly index (PAI); for forest windfall, extract the sudden change of canopy structure and texture, such as the increase of shadow area and the increase of texture roughness, and output the entropy ; For forest degradation, extract the time trend characteristics of vegetation coverage and growth status, such as the downward trend of the vegetation index curve, and output the vegetation index , Vegetation Damage Index and landscape fragmentation . Feature parameter set θⱼ∈[HAI, SI, ENT, ΔNDVI, PAI , VDI, LFI] .

[0075] The input data of this step is the first non-dissipative remote sensing image data and its corresponding confidence probability . Assume that the number of abnormal key feature types to be extracted is , then the output of this step is a The first eigenvector of the abnormal key feature of dimension , each dimension of the vector corresponds to the eigenvalue of an abnormal key feature type. The first eigenvector It is expressed as:

[0076]

[0077]

[0078]

[0079] in, Indicates The characteristic value of the abnormal key feature, Indicates The number of features extracted from the abnormal key features; For the The weight coefficient of each feature; For the The feature extraction function takes image as input. and confidence , the output is the quantized value of the feature; is the characteristic parameter set;

[0080] When extracting features, it is necessary to weight the features according to the actual situation. Adjustments can generally be made by using expert experience to assign weights, or by optimizing weights through methods such as cross-validation. The second process of extracting key features of disasters in each dimension from non-dissipative remote sensing image data is similar, and the second feature vector is output. .

[0081] Step S04, based on the first and second feature vectors obtained in step S03, a similarity classification function is designed. The first and second non-dispersive remote sensing image data are used as input, and the function is optimized by a classification learning algorithm so that it can accurately measure whether the target forest area has the same type of disaster as that in the history of the target forest area.

[0082] Exemplarily, two types of data need to be prepared when performing similarity classification learning. The first type is the first and second feature vectors, which contain descriptions of various abnormal features and characteristic descriptions of disasters, such as the type, intensity, and impact range of abnormal features and disaster features; the second type is paired sample data, namely the first non-dissipative remote sensing image data and the second non-dissipative remote sensing image data, which are used to train the similarity classification model. By optimizing the feature representation and distance measurement function, the distance between samples of the same type is made smaller than the distance between classes, thereby improving the classification and clustering performance.

[0083] For example, the first category: the input data of this step is the first feature vector , where the feature vector of each point can be expressed as In addition, the second eigenvector ,in is the similarity label of the sample, indicating whether the sample belongs to the same type of disaster. The output of this step is a similarity classification function , used to determine any two first and second eigenvectors and The similarity between:

[0084]

[0085] in, is the kernel function parameter, which is used to control the similarity decay rate. The larger the value, the faster the similarity decays with increasing distance; is the feature weight matrix, which is used to balance the importance of different features; represents the difference of eigenvectors, Represents transposition; this function can be used to map the Euclidean distance in the high-dimensional feature space to [0,1] interval to achieve quantitative expression of similarity.

[0086] The output is the similarity value of the two feature vectors, ranging from [0 ， 1] The closer the similarity value is to 1, the more similar the two feature vectors are, that is, the closer the current forest state is to the historical disaster characteristics; the closer the similarity value is to 0, the greater the difference between the two feature vectors, that is, the current state is significantly different from the historical disaster characteristics. Furthermore, in order to obtain the optimal similarity classification function, it is necessary to use a learning algorithm to adjust the parameters. and matrix The optimization process is based on the labeled training sample set. Conduct, among which, and Respectively For the feature vector of the sample, is a label, 1 means it belongs to the same type of disaster, and 0 means it does not belong to the same type of disaster. The optimization goal is to minimize the following loss function: , where the first term is the cross entropy loss, which is used to measure the difference between the predicted value and the true label; the second term is the regularization term, which is used to prevent overfitting. is the regularization coefficient, Represents the feature weight matrix The Frobenius norm of .

[0087] For example, the specific optimization process adopts an alternating iterative approach: first fix , using gradient descent to optimize ; then fix ,optimization . The update formula is: ,in, is the learning rate, is the loss function The partial derivative of . The update uses the projected gradient descent method to ensure Maintain symmetric positive definiteness: ,in, It means projecting the matrix onto a symmetric positive definite cone. The iteration process continues until the loss function converges or the maximum number of iterations is reached. After the optimization is completed, the optimal kernel function parameters are obtained. and the feature weight matrix After optimization, the latest similarity classification function is obtained:

[0088]

[0089] The first eigenvector of the area to be tested and the second eigenvector in the historical disaster database Perform similarity calculation to determine whether a disaster has occurred in the test area by calculating the similarity between the test sample and all historical samples.

[0090] The output of the similarity classification function will be used for subsequent disaster area detection. (usually 0.8), it is considered that the same type of disaster may have occurred in the area. Can be adjusted according to specific application requirements, The larger the value, the stricter the judgment standard.

[0091] The similarity classification method proposed in the present invention has the following characteristics: first, it makes full use of the prior knowledge of known disaster samples to improve the accuracy of recognition; second, it adopts kernel function mapping to handle the nonlinear relationship of features; third, through parameter optimization, feature weights are adaptively learned to improve the generalization ability of the model; this method provides an effective technical means for the intelligent identification of forestry disasters.

[0092] Exemplarily, a metric learning algorithm is used to learn an ideal feature subspace and distance metric function. The optimization goal of the algorithm is to make the distance between the first and second feature vectors of the same type in the feature subspace smaller than the distance between classes, thereby improving the accuracy of the similarity metric. After the learning is completed, an optimized distance metric function is output.

[0093] Exemplarily, the second category: In the learned feature subspace, an appropriate distance measurement method is selected for measuring the similarity between the first and second non-dispersed remote sensing image data, such as Mahalanobis distance, Harmanton distance, etc. Considering the correlation between features, the Euclidean distance between the first and second non-dispersed remote sensing image data in the feature subspace is calculated and multiplied by the optimized Mahalanobis matrix. The Mahalanobis distance between the first and second non-dispersed remote sensing image data of the same type will be smaller than the inter-class distance, reflecting their similarity in the feature subspace.

[0094] Furthermore, in this step, the output of the similarity classification learning includes two parts: one is the optimized disaster feature vector representation, and the other is the similarity classification function, which includes the optimized distance measurement matrix and can be directly used for the similarity calculation between the newly collected first and second non-dissipative remote sensing image data.

[0095] Step S05, performing disaster area detection on the first non-dissipative remote sensing image data, representing the area to be detected in the window as a feature vector, using the learned similarity classification function to determine whether the area belongs to a certain type of known disaster, and calculating its disaster severity value.

[0096] Exemplarily, the input data of the disaster area detection step includes two parts: one is the similarity classification function result output by the above step, and the other is the first non-dispersive remote sensing image data. The output of this step is a target forest disaster area distribution map, which indicates the probability that each pixel belongs to the disaster area. The sliding window is divided on the first non-dispersive remote sensing image data according to a certain step length and window size, and the first non-dispersive remote sensing image data is divided into several sub-areas. The total number of sub-areas D is: , where H and 𝑊 represent the height and width of the first non-dispersive remote sensing image data, ℎ and 𝑤 represent the height and width of the sliding window, and s represents the step size. Indicates rounding down.

[0097] Furthermore, the similarity score between each sub-region and the known disaster region is calculated based on the first feature vector obtained in step S03 and the similarity classification function trained in step S04. . Assume that the first eigenvectors of the r known abnormal feature regions are , then The similarity score of each sub-region is: ,in, Indicates the sub-area number currently being detected. represents the total number of known disaster samples, Indicates The score that is most similar to the sub-region among all known disaster samples, The sub-region to be detected The characteristic vector of is the characteristic vector of the known r-th disaster sample area, Indicates the calculation of the similarity between two feature vectors. Further, the similarity score is converted into a disaster probability value , indicating the possibility that the sub-region belongs to a disaster area. The specific disaster probability value is obtained according to the following formula function: ,in, and is a parameter that controls the shape of the probability distribution. The value will make the function The change near θ is steeper, so that the probability value approaches 0 or 1 faster; Decided When the value is 0.5, the output of the function is Adjusted similarity score The threshold value can be adjusted according to the actual situation.

[0098] Furthermore, the disaster probability value of the sub-region Merge into a complete disaster area distribution map For the pixel point on the first non-dispersive remote sensing image data , whose disaster probability is the average of the disaster probabilities of all sub-areas covering the point: ,in, Covering point The number of sub-regions.

[0099] Step S06: Using the identified disaster area, finely outline its boundaries through visual mapping learning to generate mapping vector boundaries and disaster area distribution maps. Calculate the geometric parameters of the disaster area such as area and perimeter.

[0100] The input data of the disaster area boundary delineation step is the disaster area distribution map output by the above step S05 ,The output of this step includes two parts: one is the vector coordinate sequence of the disaster area boundary; the other is the geometric parameters of the disaster area, such as area and perimeter.

[0101] Furthermore, the disaster area distribution map Perform simplified image analysis to obtain the first non-dispersive remote sensing image segmentation map of the disaster area and non-disaster area . Non-dissipative remote sensing image threshold , can be based on The probability distribution of is determined adaptively, and the pixels with probability greater than 90% quantile are taken as disaster areas: ,

[0102] Furthermore, the first non-dispersive remote sensing image segmentation map Connected domain analysis is performed to mark the unconnected disaster areas. For marking adjacent disaster area pixels as the same area, an appropriate region growing algorithm can be selected, for example, a depth first search (DFS) or breadth first search (BFS) can be used. The connected domain is .

[0103] Furthermore, possible false positive areas are filtered out based on the area of ​​the connected domain, the aspect ratio of the circumscribed rectangle, and other geometric properties. , the aspect ratio of the enclosing rectangle is greater than the threshold The area of ​​can be regarded as noise:

[0104]

[0105] Among them, the function , , Represents operations for calculating region area, bounding rectangle, and aspect ratio, respectively.

[0106] Furthermore, for each disaster area connected domain after screening, the coordinates of its boundary pixel points can be extracted by selecting an appropriate boundary extraction algorithm, such as Moore neighborhood tracking algorithm, Suzuki85 boundary tracking algorithm, etc., to extract a sequence of boundary pixel points: boundary v =[(x1,y1),(x2,y2),…,( x Z v , y Z v )] in, ( , ) are the pixel boundary point coordinates, For the The number of disaster area boundary points.

[0107] Furthermore, the geometric parameters of each disaster area are calculated, such as area, perimeter, circumscribed rectangle, etc. Taking area as an example, the polygon area formula can be used for calculation: ,in, is the coordinate of the simplified boundary point of the z-th disaster area, is the number of simplified boundary points.

[0108] Step S07, further transforming the monitoring results such as the disaster area distribution map, vector boundaries, geometric parameters, etc. obtained in the above steps into information products and services that can be used for business analysis and decision support.

[0109] According to the disaster area distribution map and vector boundaries, quantitative statistics and analysis of the disaster situation are carried out, such as calculating the disaster-stricken area, regional distribution, and areas of disasters of different levels. Take the statistics of the disaster-stricken area as an example: Use the vector data of the disaster area boundary to produce thematic maps of the disaster area distribution, which intuitively display the spatial distribution characteristics of the disaster situation. Thematic maps can be produced using GIS software such as ArcGIS and QGIS, or they can be automatically generated using Python drawing libraries such as Matplotlib and Bokeh. Through comparative analysis of multi-temporal disaster area monitoring results, the dynamic evolution of the disaster situation is revealed, and the scope and extent of the disaster are assessed. Change detection can be carried out from the following aspects:

[0110] For example, changes in the scope of the disaster area: compare and analyze the boundaries of the disaster area in different periods, and calculate the area, location and other information of the newly added and disappeared disaster areas. Changes in the level of disaster: compare and analyze the area of ​​disaster areas of different levels in different periods, and evaluate the trend and speed of the development of the disaster. Changes in related features: analyze the relationship between the scope and level of the disaster area and the characteristics of terrain, vegetation, meteorology, etc., and explore the driving mechanism of the change of the disaster.

[0111] Furthermore, based on the results of disaster monitoring and change detection, early warning information is generated in a timely manner and released through various channels, such as websites, mobile apps, and text messages, to remind relevant departments and the public to take preventive measures.

[0112] For example, the warning information may include: Location and scope of the disaster area: clearly describe the geographical location, administrative affiliation, boundaries of the disaster area and other information. Disaster level and trend: describe the current disaster level and the trend of change in the future. Possible impact and recommended measures: analyze and predict the possible impact of the disaster, and provide prevention and response suggestions.

[0113] Furthermore, a disaster monitoring data sharing platform will be established to provide disaster data query, statistics, analysis and other services for relevant departments, scientific research institutions, and the public, to support scientific research and management decision-making. Data sharing services can be implemented using technical means such as network map services and WebAPI.

[0114] See also Figure 2 , Figure 2A three-dimensional monitoring and early warning system device 200 for forestry disasters provided in an embodiment of the present invention comprises a data acquisition module 201, a data preprocessing module 202, a feature extraction module 203, a data classification module 204, a disaster area detection module 205, a disaster boundary delineation module 206, and an analysis and decision module 207, wherein the data acquisition module 201 collects high-resolution remote sensing image data of various types of disasters in the target forest to provide a data source for subsequent analysis; the data preprocessing module 202 preprocesses the collected remote sensing image data to improve data quality and availability, and calculates the confidence probability of non-dispersed images; the feature extraction module 203 extracts abnormal key features for different disaster types, and extracts the key features of the abnormal key features for different disaster types; The characteristic indicators and disaster key characteristic indicators are summarized to form the first and second characteristic vectors; the data classification module 204 designs a similarity classification function, which is optimized through a classification learning algorithm so that it can accurately measure whether the target forest area has experienced the same type of disaster as in the history of the target forest area; the disaster area detection module 205 uses a similarity classification function to determine whether the area to be tested belongs to a certain type of known disaster, and calculates the disaster severity value; the disaster boundary delineation module 206 finely delineates the identified disaster area boundary, generates a vector boundary and a disaster area distribution map, and calculates geometric morphological parameters; the analysis and decision-making module 207 converts the monitoring results into information products and services that can be used for business analysis and decision support, and supports scientific research and management.

[0115] In some implementations, the data preprocessing module 202 performs preprocessing on the collected remote sensing image data to improve data quality and availability, and calculates the confidence probability of the non-scattered image, and executes:

[0116] The collected remote sensing image data is preprocessed to obtain non-dispersed remote sensing image data and the corresponding confidence probability of the non-dispersed remote sensing image data. The preprocessing steps include radiometric calibration, geometric correction, image matching, image enhancement, etc., which aim to eliminate noise, distortion and other influencing factors in the image and improve data quality and availability. The confidence probability of the preprocessed non-dispersed image data is calculated using the local contrast and gradient direction histogram similarity evaluation algorithm.

[0117] In some implementations, the feature extraction module 203 extracts key feature indicators for different disaster types, aggregates them to form first and second feature vectors, and executes:

[0118] According to the confidence probability, the corresponding characteristic factors between the first non-dispersive remote sensing image data are calculated. For different types of disasters, the abnormal key characteristic indicators of each dimension that can describe the disaster are extracted from the perspectives of spectrum, texture, time, space, etc. The extracted features are used to form the first feature vector; the key disaster features of each dimension extracted from the second non-dispersive remote sensing image data form the second feature vector. The abnormal features of each dimension of the first non-dispersive image of the target area are summarized as the first feature vector, and the disaster features of each dimension of the historical second non-dispersive image of the area are summarized as the second feature vector.

[0119] In some embodiments, the data classification module 204 optimizes the classification learning algorithm in the design similarity classification function so that it can accurately determine whether the target forest area has the same type of disaster as the target forest area in history, and executes:

[0120] Based on the first eigenvector, a similarity classification function is designed. The first and second non-dispersive remote sensing image data are used as input, and the function is optimized through a classification learning algorithm so that it can accurately measure whether the target forest area has experienced the same type of disaster as in the history of the target forest area. The metric learning algorithm is used to learn the feature subspace and distance metric function so that the distance between the first and second eigenvectors of the same type in the feature subspace is less than the distance between classes. The optimized distance metric function is output as the similarity metric.

[0121] In some implementations, the disaster area detection module 205 uses the similarity measurement function to determine whether the area to be detected belongs to a certain type of known disaster and calculates the disaster degree value, and executes:

[0122] This module uses a sliding window to segment the first non-dispersive remote sensing image to obtain several sub-regions. Then the feature vector of each sub-region is calculated and compared with the feature vector of the known disaster area using a similarity measurement function to obtain the similarity score of the sub-region belonging to the disaster area. Then the similarity score is converted into a disaster probability value, indicating the probability of the sub-region being a disaster area. Finally, the disaster probability values ​​of all sub-regions are merged to generate a disaster area distribution map of the entire image, in which each pixel value represents the probability of belonging to the disaster area.

[0123] In some implementations, the disaster boundary delineation module 206 performs fine delineation of the identified disaster area boundary, generates a vector boundary and a disaster area distribution map, and calculates geometric morphological parameters, and executes:

[0124] This module takes the disaster area distribution map obtained in the previous step as input, and obtains binary images of disaster areas and non-disaster areas through threshold segmentation. Then, connected domain analysis is used to mark each independent disaster area, and possible false detection areas are filtered out based on morphological parameters such as area and bounding rectangle. Next, the coordinate sequence of the boundary pixel points of each filtered disaster area is extracted to obtain the vectorized disaster area boundary. Finally, the geometric attribute parameters of each disaster area, such as area and perimeter, are calculated.

[0125] In some embodiments, the analysis and decision module 207 converts the monitoring results into information products and services that can be used for business analysis and decision support, supports scientific research and management, and performs:

[0126] The monitoring results such as the disaster area distribution map, vector boundaries, geometric parameters, etc. are further transformed into information products and services that can be used for business analysis and decision support. Quantitative statistical analysis of the disaster situation is carried out, the distribution of disaster areas is produced, multi-temporal disaster change detection is carried out, early warning information is generated and released in a timely manner, and a disaster monitoring data sharing platform is established to provide data query, statistics, analysis and other services for relevant departments to support scientific research and management decision-making.

[0127] In some embodiments, the three-dimensional monitoring and early warning system device 200 for forestry disasters can be applied to terminal devices. It should be noted that those skilled in the art can clearly understand that for the sake of brevity, the specific operation process of the three-dimensional monitoring and early warning system device 200 for forestry disasters described above can refer to the corresponding process of the embodiment of the three-dimensional monitoring and early warning method for forestry disasters in the previous text, and will not be repeated here.

[0128] Please refer to Figure 3 , Figure 3 The following is a schematic block diagram of the structure of a terminal device provided by an embodiment of the present invention. Figure 3 As shown, the terminal device 300 includes a processor 301 and a memory 302 , and the processor 301 and the memory 302 are connected via a bus 303 , which may be, for example, an I2C (Inter-integrated Circuit) bus.

[0129] Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit (CPU), and the processor 301 can 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 gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0130] Specifically, the memory 302 may be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a mobile hard disk.

[0131] Those skilled in the art will understand that Figure 3 The structure shown in is only a partial structural block diagram related to the embodiment of the present invention, and does not constitute a limitation on the terminal device to which the embodiment of the present invention is applied. The specific server may include more or fewer components than those shown in the figure, or combine certain components, or have a different component layout.

[0132] The processor is used to run a computer program stored in the memory, and implement any of the three-dimensional monitoring and early warning methods for forestry disasters provided by the embodiments of the present invention when executing the computer program.

[0133] In one embodiment, the processor is used to run a computer program stored in the memory, and implements the following steps when executing the computer program:

[0134] Collect high-resolution remote sensing image data of the target forest to provide a data source for subsequent analysis.

[0135] The collected remote sensing image data are preprocessed to improve data quality and availability, and the confidence probability of non-escaped images is calculated.

[0136] For different disaster types, abnormal key characteristic indicators are extracted and summarized to form the first and second characteristic vectors.

[0137] A similarity classification function is designed and optimized through the classification learning algorithm so that it can accurately determine whether two areas belong to the same type of disaster.

[0138] The similarity measurement function is used to determine whether the area to be tested belongs to a certain type of known disaster and calculate the disaster degree value.

[0139] The boundaries of the identified disaster areas are carefully outlined to generate vector boundaries and disaster area distribution maps, and the geometric parameters are calculated.

[0140] Transform monitoring results into information products and services that can be used for business analysis and decision support, supporting scientific research and management.

[0141] It should be pointed out that the specific operation process of the terminal device described above can refer to the corresponding process of the embodiment of the three-dimensional monitoring and early warning method for forestry disasters in the previous text, and will not be repeated here.

[0142] The embodiment of the present invention also provides a storage medium for computer-readable storage, the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the three-dimensional monitoring and early warning methods for forestry disasters provided in the description of the embodiment of the present invention. Among them, the above-mentioned storage medium can be an internal storage unit of the terminal device described in the above-mentioned embodiment, such as a hard disk or memory of the terminal device. The storage medium can also be an external storage device of the terminal device, such as a plug-in hard disk equipped on the terminal device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc.

[0143] It will be appreciated by those skilled in the art that all or part of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

Claims

1. A three-dimensional monitoring and early warning method for forestry disasters, characterized in that: include: Step S01 collects high-resolution remote sensing image data of the target forest area; Step S02 preprocesses the remote sensing image data to obtain non-dispersive remote sensing image data, and obtains the confidence probability corresponding to the non-dispersive remote sensing image data; Step S03 determines the remote sensing image data collected in the target forest area as the first non-dispersive remote sensing image data, calculates the corresponding characteristic factors between the first non-dispersive remote sensing image data according to the confidence probability, extracts abnormal key characteristic indicators for different disaster types, and uses the extracted abnormal features of each dimension to form a first characteristic vector ; Determine the historical remote sensing image data of the target forest area as the second non-dispersive remote sensing image data, and extract the disaster key features of each dimension from the second non-dispersive remote sensing image data to form a second feature vector ; Step S04: Design a similarity classification function based on the first feature vector and the second feature vector. : ,in, is the kernel function parameter, which is used to control the similarity decay rate. The larger the value, the faster the similarity decays with increasing distance; is the feature weight matrix, which is used to balance the importance of different features; is the first eigenvector No. The characteristic value of the key characteristic of the anomaly; is the second eigenvector No. The characteristic value of the key characteristic of the anomaly; represents the difference of eigenvectors, represents transpose; The similarity classification function is optimized through the classification learning algorithm, so that it can accurately measure whether the target forest area has experienced the same type of disaster as in the history of the target forest area: ,in, is the optimal kernel function parameter, is the optimal feature weight matrix; Step S05 performs disaster area detection on the first non-dissipative remote sensing image data, represents the area to be detected in the window as a feature vector, uses the similarity classification function to determine whether the area belongs to a certain type of known disaster, and calculates its disaster degree value; Step S06: outlining the boundaries of the identified disaster area, generating a mapping vector boundary and a disaster area distribution map, and calculating geometric parameters of the disaster area; Step S07 converts the monitoring results of the disaster area distribution map, vector boundaries, and geometric parameters into information products and services that can be used for business analysis and decision support.

2. The three-dimensional monitoring and early warning method for forestry disasters according to claim 1 is characterized in that: The preprocessing of remote sensing image data includes: Performing radiometric calibration, geometric correction, image matching and image enhancement on the remote sensing image data; The confidence probability corresponding to the non-dispersed remote sensing image data is calculated using a similarity evaluation algorithm based on local contrast and gradient direction histogram.

3. The three-dimensional monitoring and early warning method for forestry disasters according to claim 1 is characterized in that: The extraction of abnormal key characteristic indicators for different disaster types includes: Extract abnormal key feature indicators from spectral features, texture features, time features and spatial features; The spectral characteristics include a hotspot anomaly index and a smoke index, and the spatial characteristics include a vegetation damage index and a landscape fragmentation.

4. The three-dimensional monitoring and early warning method for forestry disasters according to claim 1 is characterized in that: The first eigenvector The calculation formula is: ; ; ;in, represents the first eigenvector No. The characteristic value of the abnormal key feature, Indicates The number of features extracted from the abnormal key features; For the The weight coefficient of each feature; For the The feature extraction function takes image as input. and confidence , the output is the quantized value of the feature; is a feature parameter set.

5. The three-dimensional monitoring and early warning method for forestry disasters according to claim 1 is characterized in that: Disaster area detection is performed on the first non-dissipative remote sensing image data, including: a. Divide the first non-dispersive remote sensing image data into sliding windows according to the preset step size and window size to obtain a number of sub-areas, and the total number of sub-areas D is: , where H and 𝑊 represent the height and width of the first non-dispersive remote sensing image data, ℎ and 𝑤 represent the height and width of the sliding window, and s represents the step size. Indicates rounding down; b. Calculate the similarity score between each sub-region and the known disaster area , assuming that the first eigenvectors of the r known abnormal feature regions are , then Similarity score of each sub-region for: ,in, Indicates the sub-area number currently being detected. represents the total number of known disaster samples, Indicates The score that is most similar to the sub-region among all known disaster samples, The sub-region to be detected The characteristic vector of is the characteristic vector of the known r-th disaster sample area, Indicates calculating the similarity between two feature vectors; c. Score the similarity Converted into disaster probability value ; d. The disaster probability value of the sub-region Merge into a complete disaster area distribution map.

6. The three-dimensional monitoring and early warning method for forestry disasters according to claim 5 is characterized in that: The similarity score Converted into disaster probability value The calculation formula is: ,in, and is a parameter that controls the shape of the probability distribution.

7. The three-dimensional monitoring and early warning method for forestry disasters according to claim 1 is characterized in that: Delineating the boundaries of the identified disaster area includes: a. Perform threshold segmentation on the disaster area distribution map to obtain the first non-dispersive remote sensing image segmentation map of the disaster area and the non-disaster area; b. Performing a connected domain analysis on the first non-dispersive remote sensing image segmentation map to mark the disaster areas that are not connected to each other; c. Filter out false detection areas based on the area of ​​the connected domain and the aspect ratio of the circumscribed rectangle; d. Extract the coordinates of the boundary pixels of the connected domains of each disaster area; e. Calculate the geometric parameters of each disaster area.

8. A three-dimensional monitoring and early warning system for forestry disasters, implementing the three-dimensional monitoring and early warning method for forestry disasters as claimed in any one of claims 1 to 7, characterized in that: include: Data acquisition module, used to collect high-resolution remote sensing image data of the target forest area; A data preprocessing module is used to preprocess the remote sensing image data to obtain non-dispersed remote sensing image data and obtain the confidence probability corresponding to the non-dispersed remote sensing image data; A feature extraction module is used to calculate the corresponding feature factors between the first non-dispersive remote sensing image data according to the confidence probability, extract abnormal key feature indicators for different disaster types, and use the extracted abnormal features of each dimension to form a first feature vector; and extract the disaster key features of each dimension from the second non-dispersive remote sensing image data to form a second feature vector; A data classification module, used for designing a similarity classification function based on the first feature vector and the second feature vector, and optimizing the function through a classification learning algorithm; A disaster area detection module, used to perform disaster area detection on the first non-dissipative remote sensing image data to determine whether the area to be detected belongs to a certain type of known disaster; The disaster boundary delineation module is used to delineate the boundaries of the identified disaster areas and generate mapping vector boundaries and disaster area distribution maps; The analysis and decision-making module is used to convert monitoring results into information products and services that can be used for business analysis and decision support.

9. A terminal device, characterized in that: include: a. Processor; b. a memory for storing computer programs; Wherein, when the processor executes the computer program, the steps of the three-dimensional monitoring and early warning method for forestry disasters described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the three-dimensional monitoring and early warning method for forestry disasters described in any one of claims 1 to 7 are implemented.

Images