Multispectral remote sensing disaster monitoring and early warning system using deep learning

Abstract

The invention discloses a multispectral remote sensing disaster monitoring and early warning system using deep learning, and the system comprises a spectral data collection and preprocessing module which is used for continuously recording image data in different time periods based on surface reflection information to obtain an original data set, preprocessing the original data set to obtain a clear image data set; the mode recognition module is used for carrying out mode classification on the clear image data set to obtain a potential disaster related feature combination; the disaster risk assessment module is used for performing time sequence analysis on the potential disaster related feature combination to obtain a probability assessment value associated with the disaster risk; the disaster risk early warning module is used for constructing a quantitative index system associated with the disaster risk based on the probability evaluation value and determining an early warning reference standard based on the constructed quantitative index system associated with the disaster risk; and the closed-loop monitoring module is used for continuously monitoring the multispectral data acquired in real time based on an early warning reference standard to generate early warning information containing risk levels and outputting the early warning information.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of remote sensing monitoring and disaster early warning, and particularly relates to a multispectral remote sensing disaster monitoring and early warning system using deep learning. BACKGROUND

[0002] Multispectral remote sensing technology plays a crucial role in the field of disaster monitoring and early warning. It captures spectral information of different wavebands on the ground, providing important data support for natural disaster prediction and management. Therefore, it has attracted worldwide attention, especially in the context of climate change and frequent natural disasters, its strategic value is increasingly prominent.

[0003] However, many current disaster monitoring methods have significant shortcomings in data analysis and early warning mechanisms. Traditional methods often rely too much on data at a single time point, lack the ability to continuously track dynamic changes, and more importantly, existing technologies perform poorly in dealing with data interference in complex environments, such as cloud cover, surface cover change and other factors often leading to misjudgment or omission, making it difficult to meet the needs of high-precision early warning.

[0004] Under this background, the core challenges in this field gradually emerge. First, the dynamic change characteristics of multispectral data require the system to accurately capture the subtle differences in spectral reflectance at different time periods, and such differences are often disturbed by various external factors, increasing the difficulty of identification. The deeper problem is how to associate these differences with potential disaster risks to form a scientific and reasonable early warning standard. If the problems of difference identification and risk association cannot be effectively solved, the timeliness and accuracy of the early warning system will be greatly reduced, making it difficult to provide effective hints before disasters occur.

[0005] Therefore, how to accurately identify abnormal change patterns in multispectral data through deep learning technology and establish a scientific and reasonable early warning threshold system based on spectral difference has become a key problem in building an efficient early warning system for disasters. SUMMARY

[0006] To solve the above technical problems, the present application proposes a multispectral remote sensing disaster monitoring and early warning system using deep learning to solve the problems existing in the prior art.

[0007] To achieve the above purpose, the present application provides a multispectral remote sensing disaster monitoring and early warning system using deep learning, comprising:

[0008] A spectral data acquisition and preprocessing module is used to obtain ground reflectance information from multiple wavebands, continuously record image data at different time periods based on the ground reflectance information to obtain an original data set, and preprocess the original data set to obtain a clear image data set.

[0009] a pattern recognition module configured to perform pattern classification on the clear image dataset to obtain a potential disaster-related feature combination;

[0010] a disaster risk assessment module configured to perform time series analysis on the potential disaster-related feature combination to obtain a probability evaluation value associated with disaster risk;

[0011] a disaster risk early warning module configured to construct a quantitative index system associated with disaster risk based on the probability evaluation value, and determine a warning reference standard based on the constructed quantitative index system associated with disaster risk;

[0012] a closed-loop monitoring module configured to continuously monitor the multi-spectral data obtained in real time based on the warning reference standard to generate early warning information output containing risk levels.

[0013] Optionally, the multi-spectral data acquisition and preprocessing module comprises a multi-spectral data acquisition unit and a data preprocessing unit.

[0014] The multi-spectral data acquisition unit is configured to obtain ground surface reflection information from multiple wave bands, and continuously record image data of different time periods to form an original dataset containing dynamic change characteristics.

[0015] The data preprocessing unit is configured to perform denoising and correction operations on each frame of image of the original dataset using image preprocessing techniques to generate an optimized clear image dataset.

[0016] Optionally, the data preprocessing unit comprises a first preprocessing subunit, a second preprocessing subunit, a third preprocessing subunit, a fourth preprocessing subunit, and a fifth preprocessing subunit.

[0017] The first preprocessing subunit is configured to perform preliminary denoising processing on each frame of image to eliminate external interference and obtain a denoised first image dataset.

[0018] The second preprocessing subunit is configured to perform geometric and radiation correction on the denoised first image dataset to generate a corrected second image dataset.

[0019] The third preprocessing subunit is configured to perform secondary denoising processing on the corrected second image dataset using a contrast threshold range method to obtain an optimized third image dataset.

[0020] The fourth preprocessing subunit is configured to group the optimized third image dataset in time sequence to obtain a grouped fourth image dataset, and extract image change rules of the grouped fourth image dataset in each time period to obtain a change feature set.

[0021] The fifth preprocessing subunit is configured to perform abnormal segment classification on the change feature set by using a random forest algorithm to obtain an optimized clear image data set.

[0022] Optionally, the pattern recognition module comprises a spectral feature extraction unit and a pattern classification unit.

[0023] The spectral feature extraction unit is configured to extract spectral reflection characteristics of the clear image data set, construct a multi-dimensional feature matrix containing band differences, and analyze change rules of spectral reflection differences in different time periods based on the multi-dimensional feature matrix to obtain a feature distribution result reflecting changes in the ground surface.

[0024] The pattern classification unit is configured to perform pattern classification on the spectral reflection differences based on the feature distribution result by using a deep learning model to determine potential disaster-related feature combinations.

[0025] Optionally, the pattern classification unit comprises a data standardization and preliminary extraction subunit, an abnormal fluctuation separation and labeling subunit, a feature fusion and clustering subunit, and a disaster feature recognition and output subunit.

[0026] The data standardization and preliminary extraction subunit is configured to perform standardization processing on spectral reflection difference data in the feature distribution result to obtain a normalized feature basis, and perform preliminary extraction on spectral reflection difference patterns based on the normalized feature basis by using a convolutional neural network model to determine difference pattern distribution characteristics.

[0027] The abnormal fluctuation separation and labeling subunit is configured to separate abnormal fluctuation parts in the difference pattern distribution characteristics by using a feature screening method to obtain a separated abnormal fluctuation set, and determine fluctuation amplitudes in the abnormal fluctuation set according to a preset threshold range to obtain a labeled fluctuation data set.

[0028] The feature fusion and clustering subunit is configured to perform fusion processing on the labeled fluctuation data set and environmental background data to determine fused background fluctuation characteristics, and perform grouping processing on the background fluctuation characteristics by using a clustering tool based on the fused background fluctuation characteristics to determine feature combination distribution related to disaster risks.

[0029] The disaster feature recognition and output subunit is configured to perform statistical analysis on the feature combination distribution related to disaster risks and output potential disaster-related feature combinations.

[0030] Optionally, the disaster risk assessment module comprises a feature analysis and time series analysis unit and a risk probability evaluation unit.

[0031] The feature analysis and time series analysis unit is configured to weight the potential disaster-related feature combinations to obtain a weighted feature distribution.

[0032] The risk probability evaluation unit is configured to calculate a probability evaluation value associated with a disaster risk based on the weighted feature distribution, and determine a potential disaster risk level of the current ground surface change.

[0033] Optionally, the risk probability evaluation unit calculates an expression of the disaster risk probability as follows:

[0034] P=CNN(X☉W(△X))

[0035] In the formula, W(△X) is a dynamic weight matrix, CNN(·) is a trained convolutional neural network, P is a disaster risk probability, and X is original multi-spectral time series data.

[0036] Optionally, the disaster risk early warning module comprises a quantitative index construction unit, a dynamic early warning threshold setting unit, and an early warning information generation and output unit.

[0037] The quantitative index construction unit is configured to construct a quantitative index system associated with the disaster risk according to the probability evaluation value.

[0038] The dynamic early warning threshold setting unit is configured to set a dynamically adjusted early warning threshold system for different types of disaster features in combination with application scenarios and risk levels.

[0039] The early warning information generation and output unit is configured to analyze real-time monitoring data based on the early warning threshold system, trigger an early warning signal when the analysis result conforms to an abnormal change pattern, and generate early warning information containing a risk level for output.

[0040] Compared with the prior art, the present application has the following advantages and technical effects:

[0041] The multispectral remote sensing disaster monitoring and early warning system using deep learning of the present application can obtain ground surface reflection information from multiple wave bands and continuously record it, while denoising and correcting the original data to generate clear image data sets, effectively eliminating the influence of external interference factors, through the multispectral data acquisition and preprocessing module. The pattern recognition module uses a deep learning model to classify the spectral reflection differences and accurately identify potential disaster-related feature combinations, especially in complex environments. The disaster risk assessment module obtains probability evaluation values related to disaster risk by time series analysis of abnormal change patterns combined with historical data, providing a scientific basis for early warning. The disaster risk early warning module constructs a quantitative index system based on the probability evaluation values and dynamically adjusts the early warning threshold to ensure the accuracy and adaptability of the early warning. The closed-loop monitoring module automatically adjusts the data acquisition frequency according to the early warning information and focuses on monitoring high-risk areas, forming a closed-loop mechanism to capture risk changes in a timely manner. Overall, the present application significantly improves the accuracy and timeliness of disaster monitoring and early warning, providing strong technical support for disaster prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application, and of the description of these embodiments, are used to explain the application and are not intended to limit the application. In the drawings:

[0043] Figure 1 The structure diagram of the multispectral remote sensing disaster monitoring and early warning system using deep learning of the present application embodiment. DETAILED DESCRIPTION

[0044] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described here can be executed in a different order.

[0046] As Figure 1 shown, the present embodiment provides a multispectral remote sensing disaster monitoring and early warning system using deep learning, which includes a multispectral data acquisition and preprocessing module, a pattern recognition module, a disaster risk assessment module, a disaster risk early warning module, and a closed-loop monitoring module.

[0047] The multispectral data acquisition and preprocessing module is used for collecting original data sets by a multispectral remote sensing image acquisition system, and performing denoising and correction operations on the original data sets to eliminate the influence of external interference factors such as cloud cover, and generate an optimized clear image data set.

[0048] The multispectral data acquisition and preprocessing module includes a multispectral data acquisition unit and a data preprocessing unit. The multispectral data acquisition unit acquires ground surface reflection information from multiple wavebands through a multispectral remote sensing image acquisition system, continuously records image data for different time periods, and forms an original data set containing dynamic change characteristics. The data preprocessing unit uses image preprocessing techniques to perform denoising and correction operations on each frame of image according to the time series image set, eliminates the influence of external interference factors such as cloud cover, generates an optimized clear image data set, and determines clean data that can be used for subsequent analysis.

[0049] Further, the data preprocessing unit includes a first preprocessing subunit, a second preprocessing subunit, a third preprocessing subunit, a fourth preprocessing subunit, and a fifth preprocessing subunit. The first preprocessing subunit is used for preliminary denoising of each frame of image to eliminate external interference and obtain a denoised first image data set. The second preprocessing subunit is used for geometric and radiometric correction of the denoised first image data set to generate a corrected second image data set. The third preprocessing subunit uses a contrast threshold range method to perform secondary denoising on the corrected second image data set to obtain an optimized third image data set. The fourth preprocessing subunit is used for grouping the optimized third image data set in time sequence to obtain a grouped fourth image data set, and extracting the image change rule of the grouped fourth image data set in each time period to obtain a change characteristic set. The fifth preprocessing subunit is used for classifying abnormal segments of the change characteristic set using a random forest algorithm to obtain an optimized clear image data set.

[0050] As a specific embodiment of the present embodiment, by processing the original data set, the data of each frame of image is obtained, the image preprocessing technology is used for preliminary denoising processing of each frame of image, the influence of external interference is eliminated, and a first denoised image data set is obtained. According to the first denoised image data set, for the interference factors such as cloud cover, a correction processing method is used for geometric and radiation correction of the image, and a second corrected image data set is generated. For the second corrected image data set, the feature information of the clear image is obtained, and by comparing the preset threshold range, if the sharpness of a certain frame of image is lower than the threshold, secondary denoising processing is performed, and a third optimized image data set is obtained. According to the third optimized image data set, for the characteristics of time sequence, a segmented extraction method is used to group the image data in time sequence, and a fourth grouped image data set is determined. Through the fourth grouped image data set, for the dynamic characteristics of each group of images, a time series analysis tool is used to extract the image change rule in each time period, and a change characteristic set is obtained. For the change characteristic set, combined with the properties of clean data, if the change characteristics of a certain time period deviate from the preset range, it is marked as an abnormal segment, and a fifth marked image data set is generated. According to the fifth marked image data set, a random forest algorithm is used for classification processing of the abnormal segment, the distribution mode of the abnormal segment is judged, and a final analysis image data set is determined.

[0051] Further, in processing the time sequence image set, for the data acquisition of each frame of image, the multispectral sensor captures the ground information in a certain time period, assuming that data is collected once a day, and a time sequence data set is formed for 30 consecutive days. In the preliminary denoising processing, the mean filter method can be used, and for the noise points in the image, such as pixel abnormalities caused by device jitter, a 3x3 filter window is set, the center pixel value is replaced by the average value of the surrounding pixels, so as to smooth the image and reduce the influence of external interference.

[0052] When correcting the interference factors such as cloud cover, the spatial position of the image is adjusted through geometric correction to ensure that the images of different time periods are aligned to the same geographic coordinate system; at the same time, the radiation correction can adjust the brightness value of the image to a consistent range by referring to the standard reflectivity data, such as correcting the reflectivity of a certain band from the original 0.2-0.8 interval to 0.1-0.7 interval, so as to eliminate the influence of atmospheric scattering.

[0053] In the sharpness threshold comparison and secondary denoising processing, the sharpness threshold is set to 0.75, if the sharpness of a certain frame of image is only 0.6, the secondary processing is performed through the median filter method, the median value of the pixels in the 5x5 window is selected to replace the abnormal points, so as to improve the visibility of image details

[0054] For segmentation extraction according to time series characteristics, 30 days of image data are divided into 6 groups according to every 5 days, which is convenient for subsequent analysis of short-term change trend of images in each group. In the extraction of dynamic characteristics, the change of reflection value of each group of images is analyzed by time series analysis tool. For example, the reflection value of a certain group of images in a certain wave band gradually rises from 0.3 to 0.5 from the first day to the fifth day, indicating that the ground surface may have experienced vegetation growth or water increase. For the abnormal marking of change characteristics, if the change amplitude of the reflection value of a certain time period exceeds the preset range, such as the normal fluctuation range is 0.1, and the change of a certain period reaches 0.3, it is marked as an abnormal period. In the classification processing of the abnormal period, the random forest algorithm is used to analyze the characteristics of the abnormal period in multiple dimensions, such as combining the reflection value, time period and wave band information, to judge whether the abnormality is caused by seasonal change or sudden event.

[0055] The pattern recognition module classifies the spectral reflection difference patterns and identifies the abnormal change patterns. The pattern recognition module includes a spectral feature extraction unit and a pattern classification unit. The spectral feature extraction unit extracts the spectral reflection difference features of each wave band for the clear image data set, constructs a multi-dimensional feature matrix, analyzes the change rule of the spectral reflection difference in different time periods, and obtains the feature distribution result reflecting the ground surface change; the pattern classification unit is used to classify the spectral reflection difference patterns based on the feature distribution result, use the deep learning model to classify the spectral reflection difference patterns, and identify the abnormal change patterns, especially the abnormal fluctuations under complex environmental influence, and judge the potential disaster related feature combination.

[0056] As a specific implementation of the embodiment, for a clear image data set, a spectral analysis tool is used to extract the spectral reflection characteristics of each band, a multi-dimensional feature matrix containing band differences is constructed, and a preliminary spectral reflection feature set is obtained. According to the preliminary spectral reflection feature set, for the difference changes in the time period, a time series segmentation method is used to group process the feature matrix, and the spectral reflection change trend in different time periods is determined. For the spectral reflection change trend in different time periods, the related feature information of the ground surface change is obtained, and by comparing the preset threshold range, if the change trend of a time period exceeds the threshold range, it is marked as an abnormal change period, and a marked feature distribution set is obtained. According to the marked feature distribution set, for the distribution characteristics of the abnormal change period, a random forest algorithm is used to classify the abnormal period, judge the distribution mode of the abnormal change period, and obtain the classified feature distribution result. For the classified feature distribution result, the dynamic characteristics of the ground surface change are obtained, the change law in each time period is extracted through a time series analysis tool, and the feature distribution mode of the ground surface change is determined. According to the feature distribution mode of the ground surface change, for the overall characteristics of the image data set, a data integration method is used to summarize the feature distribution, and a final ground surface change feature set is obtained. For the final ground surface change feature set, the distribution characteristics related to the band difference are obtained, and by comparing and analyzing the distribution set in different time periods, the long-term trend characteristics of the ground surface change are judged.

[0057] As a specific implementation of the embodiment, the mode classification unit includes: a data standardization and preliminary extraction subunit, an abnormal fluctuation separation and marking subunit, a feature fusion and clustering subunit, and a disaster feature recognition and output subunit; wherein the data standardization and preliminary extraction subunit is used to standardize the spectral reflection difference data in the feature distribution result to obtain a normalized feature basis; and based on the normalized feature basis, a convolutional neural network model is used to preliminarily extract the spectral reflection difference mode to determine the difference mode distribution characteristics; the abnormal fluctuation separation and marking subunit is used to separate the abnormal fluctuation part in the difference mode distribution characteristics by using a feature screening method to obtain a separated abnormal fluctuation set; and according to a preset threshold range, the fluctuation amplitude in the abnormal fluctuation set is judged to obtain a marked fluctuation data set; the feature fusion and clustering subunit is used to fuse the marked fluctuation data set with the environmental background data to judge the fused background fluctuation characteristics; and based on the fused background fluctuation characteristics, a clustering tool is used to group process the background fluctuation characteristics to determine the feature combination distribution related to the disaster risk; the disaster feature recognition and output subunit is used to statistically analyze the feature combination distribution related to the disaster risk, and output the potential disaster related feature combination.

[0058] Further, in processing the feature distribution data, the standardization processing of spectral reflectance information is performed by a data preprocessing tool to adjust the reflectance values of different wavebands to a unified dimension range. Assuming that the reflectance value range of a certain waveband in the original data is between 0.1 and 0.9, and that of another waveband is between 0.3 and 1.2, after standardization, all data can be mapped to the interval of 0 to 1, which is convenient for subsequent analysis. In extracting the spectral reflectance and difference mode, when using a convolutional neural network model, the image data is input into the model in time sequence to capture the subtle difference features between wavebands. Assuming that in a certain time period, the reflectance value of the red light waveband suddenly increases by 0.2, and that of the near-infrared waveband decreases by 0.1, the model can identify this abnormal combination through multi-layer convolution operation, and preliminarily judge it as a signal of vegetation cover change. In separating the abnormal fluctuation part, the feature screening method is based on statistical threshold setting, for example, the part with a fluctuation amplitude more than twice the average value is marked as abnormal. Assuming that the reflectance value of a certain area fluctuates by 0.3 in a certain time period, and the average fluctuation is only 0.05, it is separated into an abnormal fluctuation set. In the correlation analysis of abnormal fluctuation and potential disaster, if the fluctuation amplitude exceeds the preset threshold value 0.25, it is marked as high-risk fluctuation. For example, the fluctuation values of a certain area for three consecutive days are 0.28, 0.3 and 0.27, all of which exceed the standard, and are marked as high-risk. In the fusion of fluctuation data set and environmental background data, the data superposition method combines the fluctuation data with the background information such as soil moisture and rainfall. Assuming that the soil moisture data of a high-risk fluctuation area is 80%, which is much higher than the average of 50%, after fusion, it can be inferred that the fluctuation may be related to waterlogging. In the grouping processing of background fluctuation features, the clustering tool can divide the features into three groups according to similarity, corresponding to high, medium and low disaster risks respectively. Assuming that the feature fluctuation value of a certain group is high and the soil moisture exceeds the standard, it is classified into the high-risk group. In the summary of feature combination distribution, the data integration method integrates the high-risk feature sets of different time periods to form a complete disaster-related feature set. Assuming that a certain area has two high-risk fluctuations in three months, after summarizing, the long-term risk trend can be judged.

[0059] The disaster risk assessment module is configured to obtain a probability evaluation value associated with the disaster risk based on the potential disaster-related feature combination; if the disaster-related feature combination determined exceeds a preset threshold range, further time series analysis is performed on the abnormal change mode, and a probability evaluation value associated with the disaster risk is obtained by comparing the current change trend with the historical data.

[0060] Further, the disaster risk assessment module comprises a feature analysis and time series analysis unit and a risk probability evaluation unit; the feature analysis and time series analysis unit is configured to perform weighted processing on the potential disaster-related feature combination to obtain a weighted feature distribution; and the risk probability evaluation unit is configured to calculate a probability evaluation value associated with the disaster risk based on the weighted feature distribution, and determine the potential disaster risk level of the current ground surface change.

[0061] As a specific implementation of the present embodiment, the risk probability evaluation unit calculates the expression of the disaster risk probability as:

[0062] P = CNN(X☉W(△X))

[0063] In the formula, W(ΔX) is a dynamic weight matrix, CNN(·) is a trained convolutional neural network, P is a disaster risk probability, and X is original multispectral time series data.

[0064] As a specific implementation of the present embodiment, for the association of disaster features and feature combinations, the feature combinations are preliminarily classified by a data screening tool to obtain a classified feature subset. According to the classified feature subset, in combination with the characteristics of abnormal changes and change patterns, a data matching method is used to perform hierarchical processing on the change patterns to obtain a hierarchical pattern set. According to the hierarchical pattern set, for the corresponding relationship between the current trend and the time series, the pattern set is time series labeled by a data mapping tool to determine a labeled time series feature group. According to the labeled time series feature group, in combination with the logic of historical data and data comparison, if the change amplitude in the time series feature group exceeds the preset threshold range, the feature group is weighted to obtain a weighted feature distribution. According to the weighted feature distribution, for the association of trend analysis and probability evaluation, if the fluctuation frequency in the feature distribution is higher than the preset frequency threshold, the distribution is deeply analyzed by a convolutional neural network model to obtain an analyzed risk probability value. According to the analyzed risk probability value, in combination with the corresponding relationship between the risk association and the disaster feature, the probability value and the disaster feature are integrated by a data fusion method to judge the integrated risk level distribution. According to the integrated risk level distribution, for the mapping relationship between the risk association and the change pattern, the level distribution is structured stored by a data archiving tool to determine the final risk pattern archive.

[0065] Further, in the study of the correlation between spectral reflectance data and disaster characteristics, for the preliminary classification of feature combinations, the data screening tool groups feature combinations according to different intervals of reflectance intensity. Assuming that a data set contains reflectance values of multiple spectral bands, the reflectance values in the range of 0.3 to 0.5 are classified as low-intensity group, and the reflectance values in the range of 0.5 to 0.8 are classified as medium-intensity group. For hierarchical processing of the characteristics of change patterns, the data matching method is used to layer the patterns according to time span and fluctuation amplitude. Assuming that in a pattern set, the short-term fluctuation span less than 24 hours is classified as instantaneous layer, and the span more than 72 hours is classified as long-term layer. In the time sequence labeling of the pattern set, the data mapping tool is used to correspond the pattern set to specific time points. For example, an abnormal fluctuation is labeled as occurring in the period of 15:00 to 16:00 on a specific date, and combined with environmental background data, it is determined whether it is related to a specific weather event. For weighted processing of the change amplitude of the time sequence feature group, if the fluctuation amplitude of a feature group exceeds the preset threshold value 0.2, its weight can be increased to 1.5 times to highlight its importance. In the deep analysis by the convolutional neural network model, the fluctuation frequency in the feature distribution is taken as the input. Assuming that the fluctuation frequency of a certain distribution is 3 times per hour, which is higher than the preset threshold value 2 times, the model will focus on analyzing its potential risk probability. For the integration of risk probability value and disaster characteristics, the data fusion method is used to associate the features with probability values above 0.75 with historical disaster data to determine whether they may correspond to some environmental anomalies.

[0066] The disaster risk early warning module includes a quantitative index construction unit, a dynamic early warning threshold setting unit, and a early warning information generation and output unit. The quantitative index construction unit is used to construct a quantitative index system related to disaster risk according to the probability evaluation value. The dynamic early warning threshold setting unit is used to set a dynamically adjusted early warning threshold system for different types of disaster characteristics in combination with application scenarios and risk levels. The early warning information generation and output unit is used to analyze real-time monitoring data based on the early warning threshold system, trigger an early warning signal when the analysis result conforms to an abnormal change pattern, and generate early warning information containing risk levels for output.

[0067] The closed-loop monitoring module is used to continuously monitor the multi-spectral data obtained in real time based on the early warning reference standard, and generate early warning information containing risk levels for output.

[0068] The real-time acquisition tool obtains a raw data stream from a multi-spectral data source, and a preliminary feature set is obtained by preliminarily extracting key features in the data stream. According to the preliminary feature set, the pre-established early warning standard is used to compare the feature set item by item, and if a certain item of data in the feature set deviates from the preset reference basis range, it is marked as a potential abnormal point to determine an abnormal marker set. According to the abnormal marker set, the data in the marker set is matched in mode by combining the change mode analysis tool, and if the matching result conforms to the pre-defined mode of abnormal change, a trigger signal is generated to obtain a signal trigger record. Through the signal trigger record, the early warning signal generation rule is combined to classify and process the abnormal points in the record to obtain the classified abnormal level distribution. According to the classified abnormal level distribution, the distribution data is corresponded to the preset risk level interval through the data mapping tool according to the risk level division logic to determine the final risk level label. The final risk level label and the related data of the abnormal change are integrated by using the information output module to generate the structured early warning information content and obtain a complete early warning information data set. Through the data transmission tool, the complete early warning information data set is sent to the related monitoring system to complete the information pushing process to obtain a pushing completion state.

[0069] Through the early warning information triggering mechanism, the identification data of the high-risk area is obtained from the monitoring system, and the identification data is compared with the geographic information database by using the pre-established area division logic to obtain the accurate range distribution of the high-risk area. According to the accurate range distribution of the high-risk area, the data acquisition frequency is automatically adjusted, the frequency adjustment instruction is issued to the image acquisition equipment in the range, the acquisition parameters are updated through the equipment control module to determine the acquisition plan after the frequency adjustment. According to the acquisition plan after the frequency adjustment, the image acquisition density is increased, the higher resolution spectral reflectance data is obtained by calling the image sensor interface, the raw images acquired are cleaned through the data preprocessing tool to obtain the processed spectral reflectance data set. The difference data is extracted from the processed spectral reflectance data set, the support vector machine algorithm is used for feature classification of the difference data, and if the classification result shows that the difference data deviates from the preset threshold range, it is marked as a potential risk point to obtain a marked risk point set. According to the marked risk point set, the trend of risk expansion is analyzed, the current risk point set is matched with the historical data through the time series comparison tool, and if the matching result shows that the risk point distribution range continues to expand, a risk expansion signal is generated to determine the risk expansion state. According to the risk expansion state, a closed-loop monitoring instruction is generated, the instruction is issued to the acquisition equipment through the data transmission channel, the key area and density of subsequent image acquisition are automatically adjusted, and an updated monitoring strategy is obtained. The updated monitoring strategy is used to continuously perform data acquisition and spectral reflectance difference analysis, and the results of each adjustment and analysis are recorded through the system log to obtain a complete closed-loop monitoring data stream.

[0070] The above merely provides the preferred embodiments of the present application, and the protection scope of the present application is not limited thereto, and any changes or substitutions within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multispectral remote sensing disaster monitoring and early warning system using deep learning, characterized by: include: A spectral data acquisition and preprocessing module is used to obtain surface reflectance information from multiple bands, continuously record image data of different time periods based on the surface reflectance information to obtain a raw data set, and preprocess the raw data set to obtain a clear image data set; A pattern recognition module, configured to perform pattern classification on the clear image dataset to obtain a potential disaster-related feature combination; A disaster risk assessment module, configured to perform a time series analysis on the potential disaster-related feature combinations to obtain a probability assessment value associated with the disaster risk; A disaster risk warning module, configured to construct a quantitative indicator system associated with disaster risks based on the probability assessment value, and determine a warning reference standard based on the constructed quantitative indicator system associated with disaster risks; The closed-loop monitoring module is used to continuously monitor the multispectral data acquired in real time based on the warning reference standard and generate warning information output including risk level.

2. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 1 is characterized in that: The multispectral data acquisition and preprocessing module includes a multispectral data acquisition unit and a data preprocessing unit; The multispectral data acquisition unit is used to obtain surface reflectance information from multiple bands and continuously record image data of different time periods to form an original data set containing dynamic change characteristics; The data preprocessing unit is used to perform denoising and correction operations on each frame of the original data set using image preprocessing technology to generate an optimized clear image data set.

3. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 2 is characterized in that: The data preprocessing unit includes: a first preprocessing subunit, a second preprocessing subunit, a third preprocessing subunit, a fourth preprocessing subunit, and a fifth preprocessing subunit; The first pre-processing sub-unit is used to perform preliminary denoising on each frame of image to eliminate external interference and obtain a denoised first image data set; The second pre-processing sub-unit is used to perform geometric and radiometric correction on the denoised first image dataset to generate a corrected second image dataset; The third pre-processing sub-unit performs secondary denoising on the corrected second image dataset using a comparison threshold range method to obtain an optimized third image dataset; The fourth preprocessing subunit is used to group the optimized third image dataset in chronological order to obtain a grouped fourth image dataset, and extract image change patterns of the grouped fourth image dataset in each time period to obtain a change feature set; The fifth preprocessing subunit is used to use a random forest algorithm to classify abnormal segments of the change feature set to obtain an optimized clear image data set.

4. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 1, characterized in that: The pattern recognition module includes a spectral feature extraction unit and a pattern classification unit; The spectral feature extraction unit is used to extract the spectral reflectance characteristics of the clear image dataset, construct a multidimensional feature matrix containing band differences, and analyze the changing patterns of spectral reflectance differences in different time periods based on the multidimensional feature matrix to obtain feature distribution results that can reflect surface changes; The pattern classification unit is used to perform pattern classification on spectral reflectance differences based on feature distribution results using a deep learning model to determine potential disaster-related feature combinations.

5. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 4 is characterized in that: The pattern classification unit includes: a data standardization and preliminary extraction subunit, an abnormal fluctuation separation and marking subunit, a feature fusion and clustering subunit, and a disaster feature identification and output subunit; The data standardization and preliminary extraction subunit is used to perform standardization on the spectral reflectance difference data in the feature distribution result to obtain a normalized feature basis; and based on the normalized feature basis, a convolutional neural network model is used to perform preliminary extraction on the spectral reflectance difference pattern to determine the difference pattern distribution characteristics; The abnormal fluctuation separation and labeling subunit is used to separate the abnormal fluctuation part in the difference pattern distribution characteristics using a feature screening method to obtain a separated abnormal fluctuation set; and to judge the fluctuation amplitude in the abnormal fluctuation set based on a preset threshold range to obtain a labeled fluctuation data set; The feature fusion and clustering subunit is used to fuse the labeled fluctuation data set with the environmental background data to determine the fused background fluctuation characteristics; and based on the fused background fluctuation characteristics, a clustering tool is used to group the background fluctuation characteristics to determine the feature combination distribution related to disaster risk; The disaster feature identification and output subunit is used to perform statistical analysis on the distribution of feature combinations related to disaster risks and output potential disaster-related feature combinations.

6. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 1, characterized in that: The disaster risk assessment module includes: a feature analysis and time series analysis unit, and a risk probability assessment unit; The feature analysis and time series analysis unit is used to perform weighted processing on the potential disaster-related feature combination to obtain a weighted feature distribution; The risk probability assessment unit is used to calculate a probability assessment value associated with the disaster risk based on the weighted characteristic distribution, and determine the potential disaster risk level of the current surface change.

7. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 6, characterized in that: The risk probability evaluation unit calculates the disaster risk probability using the following expression: P = CNN(X⊙W(ΔX)), where W(ΔX) is the dynamic weight matrix, CNN(·) is the trained convolutional neural network, P is the disaster risk probability, and X is the original multispectral time series data.

8. The multispectral remote sensing disaster monitoring and early warning system using deep learning according to claim 1, characterized in that: The disaster risk warning module includes: a quantitative indicator construction unit, a dynamic warning threshold setting unit, and a warning information generation and output unit; The quantitative indicator construction unit is used to construct a quantitative indicator system associated with disaster risks based on the probability assessment value; The dynamic warning threshold setting unit is used to set a dynamically adjusted warning threshold system based on different types of disaster characteristics, combined with application scenarios and risk levels; The warning information generation and output unit is used to analyze the real-time monitoring data based on the warning threshold system, trigger a warning signal when the analysis result meets the abnormal change pattern, and generate warning information output including the risk level.

Images