Forest and grass fire slash recovery prediction method based on historical location slice information

Abstract

A forest and grass fire slash recovery prediction method based on historical bit slice information is characterized by comprising the following steps: step 1, constructing a forest and grass fire slash recovery prediction system; 2, an image acquisition module acquires a satellite remote sensing image of a target forest land area; 3, performing image correction operation on the satellite remote sensing image by an image correction module to obtain a standard image; 4, the EVI, T and MVHI of the standard image are obtained through calculation by a multi-modal remote sensing data fusion calculation module; 5, performing normalization processing and data enhancement on the EVI, the T and the MVHI by a preprocessing module; 6, a machine learning model in the dynamic time analysis module predicts first recovery time through a machine learning algorithm; an ARIMA time sequence model in the dynamic time analysis module predicts and obtains second recovery time according to the MVHI; and 7, performing weighted average fusion on the first recovery time and the second recovery time by the dynamic time analysis module, and outputting final recovery time. The method has the effect of accurately evaluating and predicting the recovery time of forest grass vegetation after the fire.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring and post-disaster recovery, and in particular to a forest and grass fire scar recovery prediction method based on historical bitmap information. Background Art

[0002] Forest fires pose a serious threat to the ecological environment and human society. Their suddenness and destructiveness make effective fire monitoring and recovery assessment particularly important. However, existing technologies have significant shortcomings in predicting post-fire recovery. Traditional recovery time prediction methods rely on limited field data and manual analysis, making it difficult to provide comprehensive and real-time assessments. Recovery assessments based on simple vegetation indices (such as NDVI) cannot accurately reflect the true health of vegetation during dynamic fire recovery. In addition, single remote sensing data sources, such as optical imagery or thermal infrared data, have certain limitations in monitoring fire recovery and cannot fully capture the complex dynamic changes in vegetation recovery. Although existing time series analysis methods (such as ARIMA models) can provide basic predictions of data trends, when combining multiple data sources, it is difficult to fully consider the spatial variability in the recovery process. Although spatial interpolation methods can process spatial data, they have limited adaptability to the timeliness and dynamic changes in recovery time prediction.

[0003] Disadvantages of existing technologies: Traditional methods are insufficient in terms of accuracy and real-time performance in data fusion, dynamic monitoring, and recovery time prediction. Innovative technical solutions are urgently needed to improve the scientificity and practicality of fire recovery assessment. Summary of the Invention

[0004] The present invention provides a forest and grass fire scar restoration prediction method based on historical position slice information, which can accurately evaluate and predict the restoration time of forest and grass vegetation after fire.

[0005] To achieve the above-mentioned purpose, the present invention provides a method for predicting forest and grass fire scar restoration based on historical bitmap information, the key of which is to include the following steps:

[0006] Step 1: Constructing a forest and grass fire scar restoration prediction system, wherein the forest and grass fire scar restoration prediction system is provided with an image acquisition module, an image correction module, a multimodal remote sensing data fusion calculation module, a preprocessing module, and a dynamic time analysis module connected in sequence;

[0007] The dynamic time analysis module is provided with a machine learning model and an ARIMA time series model;

[0008] Step 2: The image acquisition module acquires satellite remote sensing images of the target forest area before and after the fire with time series characteristics, and transmits the satellite remote sensing images to the image correction module;

[0009] Step 3: The image correction module performs an image correction operation on the satellite remote sensing image to obtain a standard image, and transmits the standard image to the multimodal remote sensing data fusion calculation module;

[0010] Step 4: The multimodal remote sensing data fusion calculation module calculates the enhanced vegetation index (EVI) based on the optical data in the standard image; the multimodal remote sensing data fusion calculation module calculates the surface temperature (T) based on the thermal infrared data in the standard image; the multimodal remote sensing data fusion calculation module calculates the comprehensive vegetation health index (MVHI) based on the enhanced vegetation index (EVI) and the surface temperature (T), and then transmits the enhanced vegetation index (EVI), the surface temperature (T) and the comprehensive vegetation health index (MVHI) to the preprocessing module;

[0011] Step 5: The preprocessing module performs normalization and data enhancement on the enhanced vegetation index EVI, surface temperature T and comprehensive vegetation health index MVHI, and then passes the preprocessed data to the dynamic time analysis module;

[0012] Step 6: The machine learning model in the dynamic time analysis module uses a machine learning algorithm to predict the vegetation recovery time based on the pre-processed enhanced vegetation index EVI, surface temperature T and comprehensive vegetation health index MVHI to obtain the first recovery time RT ML ;

[0013] The ARIMA time series model in the dynamic time analysis module analyzes the time series trend of the comprehensive vegetation health index MVHI according to the MVHI, and then calculates the time series trend of the MVHI according to the time series of the MVHI {MVHI t}, predict the future MVHI value {MVHI t+k}, and finally get the second recovery time RT ARIMA ;

[0014] Step 7: The dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion and output the final recovery time RT.

[0015] Through this design, the multimodal remote sensing data fusion calculation module generates a comprehensive vegetation health index (MVHI) by fusing optical and thermal infrared data. Optical data is used to calculate the Enhanced Vegetation Index (EVI), while thermal infrared data is used to calculate the surface temperature (T), which assesses the impact of environmental conditions on vegetation recovery. These data are then combined to calculate the MVHI, improving the accuracy of vegetation health assessments.

[0016] The dynamic temporal analysis module uses historical remote sensing imagery to analyze vegetation changes before and after fires, accurately identifying and quantifying areas of post-fire vegetation damage and the extent of recovery. By processing and analyzing multi-temporal data, it ensures highly accurate recovery time predictions.

[0017] The time series model predicts vegetation recovery time (RT) by analyzing post-fire trends in the MVHI. Vegetation is considered recovered when the MVHI reaches or exceeds a specific threshold. The MVHI provides contextual information on vegetation health, enhancing the model's predictive capabilities. The dynamic time analysis module dynamically adjusts prediction parameters, flexibly adapting to different fire recovery scenarios and improving the accuracy of RT predictions.

[0018] Preferably, in step 3, the image correction operation includes checking and processing missing values, outliers and noise in the satellite remote sensing image, and performing atmospheric correction and image correction on the image.

[0019] First, interpolation, mean filling and other methods are used to process missing values ​​in satellite remote sensing images, remove outliers and perform noise filtering; then atmospheric correction is performed to remove the influence of the atmosphere on satellite remote sensing images, and radiation correction is performed on the thermal infrared data in satellite remote sensing images to ensure the accuracy and consistency of the data.

[0020] Preferably, in step 4, the calculation expression of the enhanced vegetation index EVI is:

[0021]

[0022] Where G is the gain factor; C1 and C2 are correction coefficients; L is the adjustment factor; RED is the reflectance of the red light band; NIR is the reflectance of the near-infrared band; BLUE is the reflectance of the blue light band;

[0023] The calculation expression of the surface temperature is:

[0024]

[0025] Where T is the surface temperature; Lλ is the thermal infrared radiation brightness; ε is the surface emissivity; σ is the Stefan-Boltzmann constant, σ = 5.67×10 -8 W / m 2 ·K 4 ;

[0026] The calculation expression of the comprehensive vegetation health index MVHI is:

[0027] MVHI=a×EVI+b×T

[0028] Among them, a and b are weight coefficients.

[0029] By integrating multi-source data, the assessment accuracy of vegetation health status is improved, providing reliable data support for recovery time prediction.

[0030] Preferably, in step 6, the machine learning model uses a random forest algorithm to predict the recovery time, and the model expression is:

[0031]

[0032] Where N is the number of trees, N = 100; f i ([MVHI_t,EVI_t,T_t]) is the predicted value of the i-th decision tree.

[0033] The first recovery time predicted by the machine learning model is a fixed value based on current features, which tends to give a one-time recovery time estimate. It is more suitable for short-term predictions or modeling of complex feature relationships, but may ignore the dynamic changes of time series.

[0034] Preferably, the machine learning model is optimized by minimizing the mean square error (MSE) loss function, which is expressed as follows:

[0035]

[0036] Where m is the number of training samples, m is a positive integer; RT true,j is the true recovery time of the jth sample; RT ML,j is the predicted recovery time of the jth sample.

[0037] As a preference: in step 6, the ARIMA time series model first performs a t} Perform first-order difference to ensure stationarity, and the difference form is:

[0038] ΔMVHI t =MVHI t -MVHI t-1

[0039] The model equation of the ARIMA time series model is:

[0040] ΔMVHI t =φ1ΔMVHI t-1 +∈ t -θ1∈ t-1

[0041] Where t is the current time; ΔMVHI t (floating point number) is the first-order difference value; φ1 is the autoregressive coefficient; θ1 is the moving average coefficient; ∈ tis white noise; φ1 (autoregressive coefficient, floating point number, typical value: 0.5) is the coefficient of the AR(1) part; θ1 (moving average coefficient, floating point number, typical value: 0.3) is the coefficient of the MA(1) part; ∈ t (white noise, floating point number, mean 0, standard deviation estimated by data) is the random error term.

[0042] ARIMA time series model is used to calculate the time series {MVHI t}Fitting parameters φ1 and θ1, and then predicting the future MVHI value {MVHI t+k}, the expression is:

[0043]

[0044] Among them, MVHI t+k The predicted vegetation health index for the kth day in the future (time t+k) indicates the state of vegetation on the kth day in the future; MVHI t is the vegetation health index at the current time t, which is the known historical data and serves as the starting point for prediction; ΔMVHI t is the change in vegetation health index at the current time t, that is, the increment of MVHI from time t-1 to time t;

[0045] By predicting {MVHI t+k}, get the time point when MVHI first reaches or exceeds the MVHI threshold:

[0046] RT ARIMA =min{k|MVHI t+k ≥0.8}

[0047] Where k is the number of days to be predicted.

[0048] The second recovery time predicted by the ARIMA time series model is derived from the MVHI time series trend. Based on the time point when the MVHI reaches the threshold, it is more suitable for capturing the dynamic process of vegetation recovery. The prediction results may be adjusted as the time series is updated, which is suitable for long-term trend analysis.

[0049] As a preference: in step 7, the dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion, the expression is as follows:

[0050] RT=w1×RT ML +w2×RT ARIMA

[0051] Among them, w1 and w2 are weight coefficients, satisfying w1 + w2 = 1. The initial value of w1 is 0.4, and the initial value of w2 is 0.6. The weights are determined through verification of historical data, and the initial values reflect that the time series trend of ARIMA is more important (w2 > w1).

[0052] By combining the first recovery time predicted by the machine learning model and the MVHI change trend and the second recovery time analyzed by the ARIMA time series model, the dynamic time analysis module reduces the limitations of a single model. For example, machine learning may ignore the time trend, and ARIMA may be insufficient in modeling non-linear relationships, effectively improving the robustness of the prediction.

[0053] The dynamic time analysis module uses the comprehensive vegetation health index MVHI as background information. The dynamic time analysis module can track the change of MVHI over time and judge the recovery process of the vegetation health status. When MVHI reaches or exceeds a specific threshold, the model confirms that the vegetation has recovered and provides a more accurate recovery time point in combination with the time series trend.

[0054] By integrating the prediction results of machine learning and the time series analysis of ARIMA, the dynamic time analysis module can capture the long-term trend and short-term fluctuations of vegetation recovery, and comprehensively consider the dynamic changes of historical data and real-time data. This makes the prediction results not only based on the static feature MVHI value but also consider the dynamic evolution in the time dimension.

[0055] Preferably: The dynamic time analysis module uses the MVHI time series {MVHI t} to evaluate the consistency of the prediction. If the MVHI trend does not match the prediction result, for example, MVHI t decreases recently, but the value of the final recovery time RT is less than the recovery time threshold threshold, then adjust the weights;

[0056] The dynamic time analysis module first calculates the recent trend of MVHI, and the expression is as follows:

[0057]

[0058] where slope represents the recent trend of MVHI;

[0059] If slope < 0 and RT < threshold, it is considered that the prediction is too optimistic, and adjust the weights to rely more on the ARIMA result because ARIMA is better at capturing the time trend. Then increase w2 and correspondingly decrease w1, and then recalculate the final recovery time RT; if the trend is consistent, keep the initial weights unchanged.

[0060] The Dynamic Time Analysis module combines machine learning and time series models to flexibly adapt to different fire recovery scenarios. Leveraging the vegetation health context provided by the MVHI, the module improves the accuracy and adaptability of recovery time and time predictions, supporting scientific recovery decision-making.

[0061] Beneficial Effects of the Invention: By integrating multimodal remote sensing data with time series analysis techniques, the present invention achieves accurate predictions of forest and grassland fire recovery times, overcoming the shortcomings of traditional methods in recovery assessment and enhancing the scientific nature of data processing and prediction. Through technological innovation, the present invention provides an effective solution for post-disaster recovery and management of forest and grassland fires, optimizes fire recovery plans, provides a scientific basis and directional guidance for the smooth implementation of post-disaster recovery work, and enhances the intelligent level of fire recovery management. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and specific examples. The following examples or drawings are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0064] like Figure 1 A method for predicting forest and grass fire scar restoration based on historical bitmap information is shown, comprising the following steps:

[0065] Step 1: Constructing a forest and grass fire scar restoration prediction system, wherein the forest and grass fire scar restoration prediction system is provided with an image acquisition module, an image correction module, a multimodal remote sensing data fusion calculation module, a preprocessing module, and a dynamic time analysis module connected in sequence;

[0066] The dynamic time analysis module is provided with a machine learning model and an ARIMA time series model;

[0067] Step 2: The image acquisition module acquires satellite remote sensing images of the target forest area before and after the fire with time series characteristics, and transmits the satellite remote sensing images to the image correction module;

[0068] Step 3: The image correction module performs an image correction operation on the satellite remote sensing image to obtain a standard image, and transmits the standard image to the multimodal remote sensing data fusion calculation module;

[0069] Step 4: The multimodal remote sensing data fusion calculation module calculates the enhanced vegetation index (EVI) based on the optical data in the standard image; the multimodal remote sensing data fusion calculation module calculates the surface temperature (T) based on the thermal infrared data in the standard image; the multimodal remote sensing data fusion calculation module calculates the comprehensive vegetation health index (MVHI) based on the enhanced vegetation index (EVI) and the surface temperature (T), and then transmits the enhanced vegetation index (EVI), the surface temperature (T) and the comprehensive vegetation health index (MVHI) to the preprocessing module;

[0070] Step 5: The preprocessing module performs normalization and data enhancement on the enhanced vegetation index EVI, surface temperature T and comprehensive vegetation health index MVHI, and then passes the preprocessed data to the dynamic time analysis module;

[0071] Step 6: The machine learning model in the dynamic time analysis module uses a machine learning algorithm to predict the vegetation recovery time based on the pre-processed enhanced vegetation index EVI, surface temperature T and comprehensive vegetation health index MVHI to obtain the first recovery time RT ML ;

[0072] The ARIMA time series model in the dynamic time analysis module analyzes the time series trend of the comprehensive vegetation health index MVHI according to the MVHI, and then calculates the time series trend of the MVHI according to the time series of the MVHI {MVHI t}, predict the future MVHI value {MVHI t+k}, and finally get the second recovery time RT ARIMA ;

[0073] Step 7: The dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion and output the final recovery time RT.

[0074] The normalization process is to scale the data to the range of [0, 1] to avoid dimensional differences. The expression is as follows:

[0075]

[0076] Among them, x is the original value, x min and x max are the minimum and maximum values ​​of the corresponding variables, respectively.

[0077] The data enhancement is to generate new samples by adding Gaussian noise (mean 0, standard deviation 0.01).

[0078] This step ensures that the data is suitable for model input and reduces noise and bias.

[0079] In step 3, the image correction operation includes checking and processing missing values, outliers and noise in the satellite remote sensing image, and performing atmospheric correction and image correction on the image.

[0080] First, interpolation, mean filling and other methods are used to process missing values ​​in satellite remote sensing images, remove outliers and perform noise filtering; then atmospheric correction is performed to remove the influence of the atmosphere on satellite remote sensing images, and radiation correction is performed on the thermal infrared data in satellite remote sensing images to ensure the accuracy and consistency of the data.

[0081] In step 4, the calculation expression of the enhanced vegetation index EVI is:

[0082]

[0083] Where G is the gain factor; C1 and C2 are correction coefficients; L is the adjustment factor; RED is the reflectance of the red light band; NIR is the reflectance of the near-infrared band; BLUE is the reflectance of the blue light band;

[0084] The calculation expression of the surface temperature is:

[0085]

[0086] Where T is the surface temperature; Lλ is the thermal infrared radiation brightness; v is the surface emissivity; σ is the Stefan-Boltzmann constant, σ = 5.67×10 -8 W / m 2 ·K 4 ;

[0087] The calculation expression of the comprehensive vegetation health index MVHI is:

[0088] MVHI=a×EVI+b×T

[0089] Among them, a and b are weight coefficients.

[0090] In step 6, the machine learning model uses a random forest algorithm (number of trees: 100, maximum depth: 10) to predict the recovery time, and the model expression is:

[0091]

[0092] Where N is the number of trees, N = 100; f i ([MVHI_t,EVI_t,T_t]) is the predicted value of the i-th decision tree.

[0093] The machine learning model is optimized by minimizing the mean square error (MSE) loss function, which is expressed as follows:

[0094]

[0095] Where m is the number of training samples, m is a positive integer; RT true,j is the true recovery time of the jth sample; RT ML,j is the predicted recovery time of the jth sample.

[0096] In step 6, the ARIMA time series model first performs a t} Perform first-order difference to ensure stationarity, and the difference form is:

[0097] ΔMVHI t =MVHI t -MVHI t-1

[0098] The model equation of the ARIMA time series model is:

[0099] ΔMVHI t =φ1ΔMVHI t-1 +∈ t -θ1∈ t-1

[0100] Where t is the current time; ΔMVHI t (floating point number) is the first-order difference value; φ1 (autoregressive coefficient, floating point number, typical value: 0.5) is the coefficient of the AR(1) part; θ1 (moving average coefficient, floating point number, typical value: 0.3) is the coefficient of the MA(1) part; ∈ t (white noise, floating point number, mean 0, standard deviation estimated by data) is the random error term.

[0101] ARIMA time series model is used to calculate the time series {MVHI t}Fitting parameters φ1 and θ1, and then predicting the future MVHI value {MVHI t+k}, the expression is:

[0102]

[0103] Among them, MVHI t+k The predicted vegetation health index for the kth day in the future (time t+k) indicates the state of vegetation on the kth day in the future; MVHI t is the vegetation health index at the current time t, which is the known historical data and serves as the starting point for prediction; ΔMVHI t is the change in vegetation health index at the current time t, that is, the increment of MVHI from time t-1 to time t;

[0104] By predicting {MVHI t+k}, get the time point when MVHI first reaches or exceeds the MVHI threshold:

[0105] RT ARIMA =min{k|MVHI t+k ≥0.8}

[0106] Where k is the number of days to be predicted.

[0107] The parameters of the ARIMA time series model are set as follows: number of autoregressive terms p = 1, number of differences d = 1, and number of moving average terms q = 1.

[0108] In step 7, the dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion, the expression is as follows:

[0109] RT=w1×RT ML +w2×RT ARIMA

[0110] Among them, w1 and w2 are weight coefficients, satisfying w1+w2=1, the initial value of w1 is 0.4, and the initial value of w2 is 0.6; the weights are verified and determined by historical data, and the initial values ​​reflect the time series trend of ARIMA more importantly (w2>w1).

[0111] The dynamic time analysis module uses MVHI as a core indicator, serving as both an input feature for machine learning models and as a time series analysis target for ARIMA. It uses both the dynamic changes in MVHI (time series trends) and its static characteristics (current health) to determine vegetation recovery status. Recovery is confirmed when the MVHI reaches a threshold, providing a more accurate RT.

[0112] The dynamic time analysis module uses the MVHI time series {MVHI t}Evaluate the consistency of the forecast. If the MVHI trend does not match the forecast, MVHI t If it decreases in the near future but the final recovery time RT is less than the recovery time threshold, the weight is adjusted;

[0113] The dynamic time analysis module first calculates the recent trend of MVHI, which is expressed as follows:

[0114]

[0115] Among them, slope represents the recent trend of MVHI;

[0116] If slope < 0 and RT < threshold, it is considered that the prediction is overly optimistic. Adjust the weights to rely more on the ARIMA results because ARIMA is better at capturing time trends. Then increase w2 and correspondingly decrease w1, and then recalculate the final recovery time RT. If the trends are consistent, keep the initial weights unchanged.

[0117] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting forest and grass fire scar restoration based on historical bitmap information, characterized in that: It includes the following steps: Step 1: Construct a forest and grassland fire-affected area restoration prediction system, which is equipped with an image acquisition module, an image correction module, a multi-modal remote sensing data fusion calculation module, a preprocessing module, and a dynamic time analysis module connected in sequence; The dynamic time analysis module is equipped with a machine learning model and an ARIMA time series model; Step 2: The image acquisition module acquires satellite remote sensing images before and after the fire with time series characteristics in the target forest area, and transmits the satellite remote sensing images to the image correction module; Step 3: The image correction module performs image correction operations on the satellite remote sensing images to obtain standard images, and transmits them to the multi-modal remote sensing data fusion calculation module; Step 4: The multi-modal remote sensing data fusion calculation module calculates the enhanced vegetation index EVI based on the optical data in the standard image; the multi-modal remote sensing data fusion calculation module calculates the surface temperature T based on the thermal infrared data in the standard image; the multi-modal remote sensing data fusion calculation module calculates the comprehensive vegetation health index MVHI based on the enhanced vegetation index EVI and the surface temperature T, and then transmits the enhanced vegetation index EVI, the surface temperature T, and the comprehensive vegetation health index MVHI to the preprocessing module; Step 5: The preprocessing module performs normalization processing and data enhancement on the enhanced vegetation index EVI, the surface temperature T, and the comprehensive vegetation health index MVHI, and then transmits the preprocessed data to the dynamic time analysis module; Step 6: The machine learning model in the dynamic time analysis module uses a machine learning algorithm to predict the vegetation recovery time based on the pre-processed enhanced vegetation index EVI, surface temperature T and comprehensive vegetation health index MVHI to obtain the first recovery time RT ML ; The ARIMA time series model in the dynamic time analysis module analyzes the time series trend of the comprehensive vegetation health index MVHI according to the MVHI, and then calculates the time series trend of the MVHI according to the time series of the MVHI {MVHI t }, predict the future MVHI value {MVHI t+k }, and finally get the second recovery time RT ARIMA ; Step 7: The dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion and output the final recovery time RT.

2. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 is characterized by: In Step 3, the image correction operation includes checking and processing missing values, outliers, and noises in the satellite remote sensing images, and performing atmospheric correction and image correction on the images.

3. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 is characterized by: In Step 4, the calculation expression of the enhanced vegetation index EVI is: where G is the gain factor; C1 and C2 are correction coefficients; L is the adjustment factor; RED is the reflectance of the red light band; NIR is the reflectance of the near-infrared band; BLUE is the reflectance of the blue light band; The calculation expression of the surface temperature is: Where T is the surface temperature; Lλ is the thermal infrared radiation brightness; ε is the surface emissivity; σ is the Stefan-Boltzmann constant, σ = 5.67×10 -8 W / m 2 ·K 4 ; The calculation expression of the comprehensive vegetation health index MVHI is: MVHI = a×EVI + b×T where a and b are weight coefficients.

4. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 is characterized by: In Step 6, the machine learning model uses the random forest algorithm to predict the recovery time, and the model expression is: Where N is the number of trees; f i ([MVHI_t,EVI_t,T_t]) is the predicted value of the i-th decision tree.

5. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 or 6, characterized in that: The machine learning model is optimized by minimizing the mean square error MSE loss function, and the expression is as follows: Where m is the number of training samples; RT true,j is the true recovery time of the jth sample; RT ML,j is the predicted recovery time of the jth sample.

6. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 is characterized by: In step 6, the ARIMA time series model first performs a t } Perform first-order difference, the difference form is: ΔMVHI t =MVI t -MVI t-1 The model equation of the ARIMA time series model is: ΔMVHI t =φ1ΔMVHI t-1 +∈ t -θ1∈ t-1 Where t is the current time; ΔMVHI t is the first-order difference value; φ1 is the autoregressive coefficient; θ1 is the moving average coefficient; ∈ t is white noise; ARIMA time series model is used to calculate the time series {MVHI t }Fitting parameters φ1 and θ1, and then predicting the future MVHI value {MVHI t+k }, the expression is: Among them, MVHI t+k MVHI is the predicted vegetation health index for the kth day in the future (time t+k); t is the vegetation health index at the current time t; ΔMVHI t is the change in vegetation health index at the current time t; By predicting {MVHI t+k }, get the time point when MVHI first reaches or exceeds the MVHI threshold: RT ARIMA =min{k∣MVHI t+k ≥0.8} where k is the number of days predicted.

7. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 1 is characterized by: In step 7, the dynamic time analysis module analyzes the first recovery time RT ML and the second recovery time RT ARIMA Perform weighted average fusion, the expression is as follows: RT=w1×RT ML +w2×RT ARIMA where w1 and w2 are weight coefficients, satisfying w1 + w2 = 1.

8. The method for predicting forest and grass fire scar restoration based on historical bitmap information according to claim 7 is characterized by: The dynamic time analysis module uses the MVHI time series {MVHI t }Evaluate the consistency of the forecast. If the MVHI trend does not match the forecast, MVHI t If it decreases in the near future but the final recovery time RT is less than the recovery time threshold, the weight is adjusted; The dynamic time analysis module first calculates the recent trend of MVHI, and the expression is as follows: where slope represents the recent trend of MVHI; If slope < 0 and RT < threshold; then increase w2, correspondingly decrease w1, and then recalculate the final recovery time RT; if the trends are the same, keep the initial weights unchanged.

Images