Forest combustible moisture content detection method based on LiDAR technology

Abstract

The invention belongs to the technical field of moisture content detection, and discloses a forest combustible moisture content detection method based on LiDAR technology, comprising: acquiring LiDAR scanning data of a target area, extracting point cloud information in the scanning data, the point cloud information comprising signal intensity and space coordinates; calculating the reflectivity and scattering characteristic of each point according to the point cloud information to obtain a reflectivity distribution diagram and a scattering characteristic diagram; reflectivity and scattering characteristic reference values corresponding to the types and the growth stages of the combustibles are obtained from a pre-established database, the reference values are compared with actual measurement values, and a reflectivity deviation value and a scattering characteristic deviation value are calculated; and acquiring rainfall data and transpiration data of the target area from the meteorological database, and calculating a moisture content change trend in a time dimension according to the rainfall data and the transpiration data to obtain a moisture content time change curve.

Description

Technical Field

[0001] The present invention belongs to the technical field of moisture content detection, and particularly relates to a method for detecting the moisture content of forest combustibles based on LiDAR technology. Background Art

[0002] When using LiDAR technology to detect the moisture content of forest combustibles, a key technical problem is faced, that is, how to establish an accurate correlation model between the moisture content of combustibles and LiDAR signals. Due to the complex and variable forest environment and the wide variety of combustibles, the surface reflectivity and scattering characteristics of combustibles of different types and at different growth stages vary significantly. In addition, the spatial distribution of combustibles also shows a high degree of non-uniformity and randomness, further increasing the difficulty of modeling.

[0003] Traditional modeling methods mainly rely on a large number of field samplings and experimental measurements. By manually collecting combustible samples, measuring their moisture content, and synchronously obtaining the corresponding LiDAR data, and then training the correlation model based on these data. However, this method is time-consuming and laborious, and it is difficult to cover a sufficient number of combustible types and growth conditions, resulting in limited generalization ability and applicability of the model.

[0004] Therefore, it is urgent to explore new modeling approaches, make full use of the rich information contained in LiDAR data, and start from two dimensions of time and space to depict the dynamic change law of the moisture content of combustibles. In the time dimension, it is necessary to consider the influence of factors such as precipitation and transpiration on the moisture content of combustibles, which shows obvious seasonal and periodic characteristics; in the space dimension, it is necessary to consider the influence of factors such as terrain and slope aspect on the moisture content, and depict the spatial distribution pattern of the moisture content of combustibles. Only by deeply analyzing the internal relationship between LiDAR signals and moisture content in the dynamic spatio-temporal background can a more accurate and robust correlation model be established, providing a reliable decision-making basis for forest fire early warning and prevention. Summary of the Invention

[0005] To solve the problems existing in the prior art, the present invention provides a method for detecting the moisture content of forest combustibles based on LiDAR technology. By integrating technologies such as LiDAR scanning, combustible classification, meteorological data analysis, and machine learning, it realizes the accurate prediction of the moisture content of combustibles in the target area, provides an important basis for forest fire prevention and management, and has significant practical value and social benefits.

[0006] To achieve the above object, the present invention provides the following solution:

[0007] A method for detecting the moisture content of forest combustibles based on LiDAR technology, the method comprising:

[0008] Obtain the LiDAR scan data of the target area, and extract the point cloud information in the LiDAR scan data, where the point cloud information includes: signal intensity and spatial coordinates;

[0009] According to the point cloud information, calculate the reflectivity and scattering characteristics of each point to obtain a reflectivity distribution map and a scattering characteristic map;

[0010] According to the reflectivity distribution map and the scattering characteristic map, combine with a preset combustible classification model to judge the type of combustibles;

[0011] Obtain the reflectivity and scattering characteristic reference values corresponding to the type of combustibles and the growth stage from a pre-established database, compare the reference values with the actual measured values, and calculate the reflectivity deviation value and the scattering characteristic deviation value;

[0012] Obtain the precipitation data and transpiration data of the target area from the meteorological database, and according to the precipitation data and transpiration data, calculate the change trend of the moisture content in the time dimension to obtain a moisture content time change curve;

[0013] Combine the corrected reflectivity and scattering characteristic values with the moisture content time change curve, input them into a pre-trained machine learning model, and use the random forest algorithm to train the correlation model between the reflectivity, scattering characteristics and moisture content to obtain a trained moisture content prediction model;

[0014] Complete the detection of the moisture content of forest combustibles according to the trained moisture content prediction model.

[0015] Preferably, obtaining the LiDAR scan data of the target area and extracting the point cloud information in the scan data includes:

[0016] Obtain the LiDAR scan data of the target area, preprocess the LiDAR scan data and extract the point cloud data;

[0017] According to the spatial range of the scan area, divide the point cloud data in space and divide the point cloud into multiple sub-regions;

[0018] According to the point cloud data of each sub-region, extract the three-dimensional spatial coordinates of the point cloud and the corresponding signal intensity value to obtain the point cloud coordinates and intensity information of the sub-region;

[0019] Merge the point cloud coordinates and intensity information of each sub-region to obtain the complete point cloud data of the target area, including the three-dimensional coordinates and signal intensity of the points.

[0020] Preferably, according to the point cloud information, calculating the reflectivity and scattering characteristics of each point to obtain a reflectivity distribution map and a scattering characteristic map includes:

[0021] Obtain the point cloud information, and for each point in the point cloud information, calculate the reflectivity and scattering characteristic value of the point;

[0022] Generate a reflectivity distribution map according to the reflectivity value of each point calculated. Different colors or grayscales in the reflectivity distribution map represent different ranges of reflectivity values;

[0023] Generate a scattering characteristic map according to the scattering characteristic value of each point calculated. Different colors or grayscales in the scattering characteristic map represent different ranges of scattering characteristic values;

[0024] Perform image segmentation processing on the reflectivity distribution map to obtain the distribution positions and ranges of different reflectivity regions;

[0025] Perform image segmentation processing on the scattering characteristic map to obtain the distribution positions and ranges of different scattering characteristic regions;

[0026] Perform image registration on the reflectivity distribution map and the scattering characteristic map so that the positions of corresponding points in the two images coincide;

[0027] Fuse the registered reflectivity distribution map and scattering characteristic map to generate a fused point cloud attribute map. Different colors or grayscales represent different combinations of reflectivity and scattering characteristics.

[0028] Preferably, according to the reflectivity distribution map and the scattering characteristic map, combined with a preset combustible classification model, the types of combustibles are judged to include:

[0029] Obtain the reflectivity distribution map, and for each point in the reflectivity distribution map, extract the reflectivity value of the point;

[0030] According to the preset combustible type classification model, judge whether the reflectivity value of the point is within the preset range;

[0031] If the reflectivity value of the point is within the preset range, determine the type of combustible corresponding to the point through the combustible type classification model;

[0032] If the reflectivity value of the point exceeds the preset range, mark the point as an unknown type;

[0033] Perform classification judgment on the type of combustible for each point in the reflectivity distribution map;

[0034] Construct a combustible type classification model according to the corresponding relationship between the reflectivity value and the type of combustible;

[0035] Use the reflectivity distribution map after classification judgment as the combustible type distribution map for subsequent identification and positioning of combustibles.

[0036] Preferably, obtaining the reflectivity and scattering characteristic reference values ​​corresponding to the type and growth stage of the combustible from a pre-established database, comparing the reference values ​​with the actual measured values, and calculating the reflectivity deviation value and the scattering characteristic deviation value includes:

[0037] Obtaining information on the type of combustibles and the growth stage of the combustibles, and obtaining reference reflectivity and reference scattering characteristic values ​​of the corresponding type and growth stage from a pre-established database;

[0038] Obtain image information of the combustible area, obtain spectral data of each pixel point in the area according to the image information, and calculate the actual reflectivity and actual scattering characteristic value of each pixel point through the spectral data;

[0039] According to the actual reflectivity and reference reflectivity of each pixel, the reflectivity deviation value of each pixel is calculated. If the reflectivity deviation value is greater than a preset threshold, it is marked;

[0040] Calculate the scattering characteristic deviation value of each pixel point according to the actual scattering characteristic value and the reference scattering characteristic value of each pixel point, and mark it if the scattering characteristic deviation value is greater than a preset threshold value;

[0041] A support vector machine algorithm is used to train according to the marked reflectivity deviation value and the marked scattering characteristic deviation value, so as to obtain a trained reflectivity deviation and scattering characteristic deviation classification model;

[0042] Using a K-nearest neighbor algorithm, according to a reflectivity deviation and scattering characteristic deviation classification model, the unlabeled reflectivity deviation values ​​and the unlabeled scattering characteristic deviation values ​​are classified to determine the classification results of the reflectivity deviation values ​​and the scattering characteristic deviation values;

[0043] The random forest algorithm is used to judge whether the classification results of the reflectivity deviation value and the scattering characteristic deviation value meet the preset conditions according to the classification results of the reflectivity deviation value and the scattering characteristic deviation value, and to obtain the accuracy of the classification results of the reflectivity deviation value and the scattering characteristic deviation value.

[0044] Preferably, the precipitation data and transpiration data of the target area are obtained from the meteorological database, and the moisture content variation trend in the time dimension is calculated according to the precipitation data and transpiration data to obtain the moisture content time variation curve, which includes:

[0045] According to the geographical location information of the target area, the historical precipitation data and evaporation data of the area are obtained from the meteorological database;

[0046] Preprocess the acquired precipitation data and evaporation data to remove outliers and missing values;

[0047] Using the time series analysis method, calculate the precipitation and transpiration at each time point to obtain the precipitation and transpiration data sequences in the time dimension;

[0048] According to the precipitation and transpiration data sequences, calculate the water content at each time point to obtain the water content data sequence in the time dimension;

[0049] Conduct trend analysis on the water content data sequence, using the moving average or exponential smoothing method to determine the change trend of the water content;

[0050] According to the change trend of the water content, generate the time-varying curve of the water content to visually display the dynamic change process of the water content.

[0051] Preferably, combine the corrected reflectivity and scattering characteristic values with the time-varying curve of the water content, input them into a pre-trained machine learning model, train the correlation model between the reflectivity, scattering characteristics and water content, and obtain the trained water content prediction model, including:

[0052] Obtain the original data of the reflectivity and scattering characteristics, and correct the original data according to the preset correction rules to obtain the corrected reflectivity and scattering characteristic values;

[0053] Obtain the curve data of the water content changing with time, correlate the curve data with the corrected reflectivity and scattering characteristic values to form a training data set;

[0054] According to the input values and output values of the training data set, train the correlation model to obtain the trained water content prediction model.

[0055] Preferably, according to the trained water content prediction model, complete the detection of the water content of forest combustibles, including:

[0056] According to the input values of the reflectivity and scattering characteristics to be measured, predict through the trained water content prediction model to obtain the corresponding water content prediction value;

[0057] Judge whether the water content prediction value exceeds the preset threshold range. If it exceeds, trigger the warning mechanism to prompt that the water content is abnormal;

[0058] According to the deviation between the water content prediction value and the actual water content, retrain the water content prediction model to continuously optimize the prediction accuracy of the model;

[0059] Correlate the water content prediction result with the time curve to generate the time-varying curve of the water content prediction, providing data support for the subsequent water content trend analysis.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] The present invention discloses a method for detecting the moisture content of forest combustibles based on LiDAR technology. This method obtains the LiDAR point cloud data of the target area, extracts the reflectivity and scattering characteristic information, and determines the type of combustibles by combining with a preset combustible classification model. Subsequently, the actual measured value is compared with the reference value in the database to calculate the deviation value. At the same time, the meteorological data is used to calculate the time variation trend of the moisture content. Finally, the corrected reflectivity, scattering characteristic values and the moisture content change curve are input into a pre-trained machine learning model, and a moisture content prediction model is established using the random forest algorithm. By integrating technologies such as LiDAR scanning, combustible classification, meteorological data analysis and machine learning, the present invention realizes the accurate prediction of the moisture content of combustibles in the target area, provides an important basis for forest fire prevention and management, and has significant practical value and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the present invention, the following briefly introduces the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 Schematic flow chart of a method for detecting the moisture content of forest combustibles based on LiDAR technology according to an embodiment of the present invention;

[0064] Figure 2 Schematic diagram of the modeling steps of the moisture content prediction model according to an embodiment of the present invention;

[0065] Figure 3 Schematic diagram of the LSTM unit structure according to an embodiment of the present invention;

[0066] Figure 4 Schematic diagram of a combustible moisture content prediction model based on informer according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0068] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0069] Embodiment 1

[0070] As shown Figure 1 in the figure, an embodiment of the present invention provides a method for detecting the moisture content of forest combustibles based on LiDAR technology, and the method includes:

[0071] Obtain LiDAR scan data of a target area, and extract point cloud information from the LiDAR scan data, wherein the point cloud information includes: signal intensity and spatial coordinates;

[0072] According to the point cloud information, calculate the reflectivity and scattering characteristics of each point to obtain a reflectivity distribution map and a scattering characteristics map;

[0073] According to the reflectivity distribution map and the scattering characteristics map, combine a preset combustible classification model to determine the type of combustibles;

[0074] Obtain the reflectivity and scattering characteristics reference values corresponding to the type of combustibles and the growth stage from a pre-established database, compare the reference values with the actual measured values, and calculate the reflectivity deviation value and the scattering characteristics deviation value;

[0075] Obtain the precipitation data and transpiration data of the target area from the meteorological database, and calculate the moisture content change trend in the time dimension according to the precipitation data and the transpiration data to obtain a moisture content time change curve;

[0076] Combine the corrected reflectivity and scattering characteristic values with the moisture content time change curve, input them into a pre-trained machine learning model, and use the random forest algorithm to train the correlation model between the reflectivity, scattering characteristics and moisture content to obtain a trained moisture content prediction model;

[0077] Complete the detection of the moisture content of forest combustibles according to the trained moisture content prediction model.

[0078] In this embodiment, obtaining the LiDAR scan data of the target area and extracting the point cloud information from the scan data includes:

[0079] Obtain the LiDAR scan data of the target area, preprocess the LiDAR scan data and extract the point cloud data;

[0080] According to the spatial range of the scan area, perform spatial division on the point cloud data, and divide the point cloud into multiple sub-regions;

[0081] According to the point cloud data of each sub-region, extract the three-dimensional spatial coordinates of the point cloud and the corresponding signal intensity value to obtain the point cloud coordinates and intensity information of the sub-region;

[0082] Merge the point cloud coordinates and intensity information of each sub-region to obtain the complete point cloud data of the target area, including the three-dimensional coordinates and signal intensity of the points.

[0083] Specifically, the spatial partitioning of point cloud data includes: Spatial analysis is a process of revealing the spatial patterns contained in spatial data and the relationships between elements, and it is the most distinctive content in GIS. To solve the problems of weak timeliness and fineness of data in current spatial analysis, the present invention first introduces LiDAR point cloud data in the spatial analysis process, aiming to generate high-precision DEM and DSM data through data classification and extraction; then extracts combustible element data related to spatial analysis from vector DLG data, and generates raster data of combustible elements after rasterization; then uses the DEM data and DSM data generated from LiDAR point cloud data, combines with the rasterized combustible data, and through normalization and other processes, generates a spatial analysis raster map under a single influencing factor; finally, according to the influence weights of various data on spatial analysis, through raster overlay analysis, generates a spatial analysis raster map with multiple influencing factors, improving the accuracy and efficiency of complex spatial analysis algorithms.

[0084] In this embodiment, calculating the reflectivity and scattering characteristics of each point according to the point cloud information to obtain a reflectivity distribution map and a scattering characteristic map includes:

[0085] Obtain the point cloud information, and for each point in the point cloud information, calculate the reflectivity and scattering characteristic values of this point;

[0086] According to the reflectivity values of each point calculated, generate a reflectivity distribution map, and different colors or grayscales in the reflectivity distribution map represent different reflectivity value ranges;

[0087] According to the scattering characteristic values of each point calculated, generate a scattering characteristic map, and different colors or grayscales in the scattering characteristic map represent different scattering characteristic value ranges;

[0088] Perform image segmentation processing on the reflectivity distribution map to obtain the distribution positions and ranges of different reflectivity regions;

[0089] Perform image segmentation processing on the scattering characteristic map to obtain the distribution positions and ranges of different scattering characteristic regions;

[0090] Perform image registration on the reflectivity distribution map and the scattering characteristic map to make the positions of corresponding points in the two images coincide;

[0091] Fuse the registered reflectivity distribution map and scattering characteristic map to generate a fused point cloud attribute map, and different colors or grayscales represent different combinations of reflectivity and scattering characteristics.

[0092] In this embodiment, judging the type of combustible according to the reflectivity distribution map and the scattering characteristic map in combination with a preset combustible classification model includes:

[0093] Obtain the reflectivity distribution map, and for each point in the reflectivity distribution map, extract the reflectivity value of that point;

[0094] According to the preset combustible type classification model, determine whether the reflectivity value of this point is within the preset range;

[0095] If the reflectivity value of this point is within the preset range, determine the combustible type corresponding to this point through the combustible type classification model;

[0096] If the reflectivity value of this point exceeds the preset range, mark this point as an unknown type;

[0097] Perform classification judgment on the combustible type for each point in the reflectivity distribution map;

[0098] Construct a combustible type classification model according to the corresponding relationship between the reflectivity value and the combustible type;

[0099] Use the reflectivity distribution map after classification judgment as the combustible type distribution map for subsequent combustible identification and positioning.

[0100] Specifically, the preset combustible type classification model is established based on a large amount of experimental data. For example, the reflectivity of wood may be between 0.5 and 0.8, while that of plastic may be between 0.3 and 0.6. When encountering a point with a reflectivity of 0.7, the system will judge that it falls within the preset range of wood. Subsequently, the combustible type classification model will further analyze and may determine it as pine. If a point with a reflectivity of 0.2 is encountered, which exceeds the preset range of common combustibles, the system will mark it as an unknown type, which helps to identify potential anomalies or new combustibles.

[0101] Performing classification judgment on the combustible type for each point in the reflectivity distribution map includes: First, use the successive projections algorithm (SPA) for basic band screening, compare and analyze the classification accuracies of two deep learning models, namely the one-dimensional convolutional neural network (1DCNN) and the long short-term memory artificial neural network (LSTM), under the conditions of the original spectrum, characteristic bands, and partial characteristic bands, and explore the information-bearing capacity of the characteristic bands for the original spectrum; Then, for the misclassification problem, adopt an advanced band screening method to retrain the misclassified samples of various combustibles under the condition of the basic variable combination, and so on in a cycle until the classification accuracy does not increase significantly, and study the spectral characteristics and misclassification rules of the misclassified samples; Finally, compare the classification accuracies of different methods.

[0102] Specifically, the constructed combustible type classification model includes: (1) Primary combustible types. Referring to the IGBP land cover classification system, the primary combustible types are divided into 3 types, namely forest, shrub, and herb. (2) Secondary combustible types. The primary combustible types are further subdivided into secondary types. Among them, according to leaf shape, forests are divided into coniferous forest, broad-leaved forest, and mixed forest types; shrubs are divided into forested shrubbery and shrubbery types; herbs are divided into woody vegetation-grass mosaic and grass types. (3) Tertiary combustible types. NTC can reflect the distribution of ground combustibles. Taking NTC (non-forest cover percentage) as the classification basis, the secondary combustible types are further divided into tertiary types. Referring to the Prometheus combustible classification system, in the coniferous forest, broad-leaved forest, and mixed forest types, when NTC ≤ 30%, it is classified as the low shrub and grass coverage type, and when NTC > 30%, it is classified as the high shrub and grass coverage type; in the forested shrubbery and shrubbery types, when NTC ≤ 60%, it is classified as the low shrub and grass coverage type, and when NTC > 60%, it is classified as the high shrub and grass coverage type; for the woody vegetation-grass mosaic and grass types, the high and low coverage types are divided by NTC > 50% or 50% ≥ NTC > 10%.

[0103] In this embodiment, the reference reflectance and scattering characteristic reference values corresponding to the combustible type and growth stage are obtained from a pre-established database, and the reference values are compared with the actual measurement values. Calculating the reflectance deviation value and the scattering characteristic deviation value includes:

[0104] Obtain the combustible type information, and at the same time obtain the combustible growth stage information, and obtain the reference reflectance and reference scattering characteristic values corresponding to the corresponding type and corresponding growth stage from the pre-established database;

[0105] Obtain the combustible area image information, obtain the spectral data of each pixel point in the area according to the image information, and calculate the actual reflectance and actual scattering characteristic values of each pixel point through the spectral data;

[0106] According to the actual reflectance and reference reflectance of each pixel point, calculate the reflectance deviation value of each pixel point. If the reflectance deviation value is greater than the preset threshold, mark it;

[0107] According to the actual scattering characteristic value and reference scattering characteristic value of each pixel point, calculate the scattering characteristic deviation value of each pixel point. If the scattering characteristic deviation value is greater than the preset threshold, mark it;

[0108] Adopt the support vector machine algorithm, and train according to the marked reflectance deviation value and the marked scattering characteristic deviation value to obtain the trained reflectance deviation and scattering characteristic deviation classification model;

[0109] Using a K-nearest neighbor algorithm, according to a reflectivity deviation and scattering characteristic deviation classification model, the unlabeled reflectivity deviation values ​​and the unlabeled scattering characteristic deviation values ​​are classified to determine the classification results of the reflectivity deviation values ​​and the scattering characteristic deviation values;

[0110] The random forest algorithm is used to judge whether the classification results of the reflectivity deviation value and the scattering characteristic deviation value meet the preset conditions according to the classification results of the reflectivity deviation value and the scattering characteristic deviation value, and to obtain the accuracy of the classification results of the reflectivity deviation value and the scattering characteristic deviation value.

[0111] Specifically, the pre-established database stores reference reflectance and scattering characteristic values ​​of various combustibles at different growth stages. Taking pine trees as an example, the reference reflectance at the seedling stage may be in the range of 0.3 to 0.4, while that at the mature stage may be in the range of 0.2 to 0.3. These reference values ​​provide a benchmark for subsequent deviation calculations.

[0112] In this embodiment, precipitation data and transpiration data of the target area are obtained from the meteorological database, and the moisture content variation trend in the time dimension is calculated according to the precipitation data and transpiration data, and the moisture content time variation curve is obtained, including:

[0113] According to the geographical location information of the target area, the historical precipitation data and evaporation data of the area are obtained from the meteorological database;

[0114] Preprocess the acquired precipitation data and evaporation data to remove outliers and missing values;

[0115] The time series analysis method is used to calculate the precipitation and evaporation at each time point to obtain the precipitation and evaporation data series in the time dimension;

[0116] According to the precipitation and transpiration data series, the moisture content at each time point is calculated to obtain the moisture content data series in the time dimension;

[0117] Carry out trend analysis on the moisture content data series and use moving average or exponential smoothing method to determine the changing trend of moisture content;

[0118] According to the changing trend of moisture content, a moisture content time variation curve is generated to intuitively display the dynamic change process of moisture content.

[0119] In this embodiment, the corrected reflectivity and scattering characteristic values ​​are combined with the moisture content time variation curve and input into a pre-trained machine learning model to train the association model between the reflectivity, scattering characteristic and moisture content. The trained moisture content prediction model includes:

[0120] Obtain the original data of reflectivity and scattering characteristics, and correct the original data according to the preset correction rules to obtain the corrected reflectivity and scattering characteristic values;

[0121] Obtain the curve data of water content changing with time, associate the curve data with the corrected reflectivity and scattering characteristic values to form a training data set;

[0122] Train the association model according to the input values and output values of the training data set to obtain a trained water content prediction model.

[0123] Specifically, the modeling steps of the water content prediction model are as Figure 2 shown. Divide the field-measured meteorological elements and water content data into training sets and test sets. The training set is used to train the deep learning model and fit the parameters of the direct estimation method equation, and the test set is used to evaluate the prediction accuracy of the model. Separate prediction models for the water content sequences of two types of combustibles are constructed.

[0124] Reconstruct the direct estimation method: First, select the Simard equilibrium water content equation, as shown in formula (1):

[0125]

[0126] In the formula: E is the equilibrium water content, H is the relative humidity, and T is the temperature.

[0127] According to Byram's moisture diffusion equation:

[0128]

[0129] In the formula: d m / d t represents the change value of the litter water content within the t time period; m represents the water content of the combustible; E represents the litter equilibrium water content; τ represents the litter time lag, M t represents the water content at time t, M 0 represents the initial water content.

[0130] Let dt = Δt = 1h, discretize the formula to obtain the discrete i-form calculation equation for the litter water content:

[0131] M t = λ 2 M t-1 + λ(1 - λ)E t-1 +(1 - λ)E t (3)

[0132] In the formula: E t-1 and E t respectively represent the equilibrium water content values at times t and t - 1; M t and Mt-1 They represent the water content values at times t and t - 1 respectively; λ is the parameter to be estimated. Substitute into the above formula using the fitted Simard equilibrium water content equation. Use the SPSS nonlinear regression method to fit the parameters of the model to obtain the value of parameter λ in formula (4).

[0133] M t+j = λ 2 M t+j-1 + λ(1 - λ)E t+j-1 +(1 - λ)E t+j (4)

[0134] If predicting the water content sequence with a time step of i hours in the future, that is, predicting the water content within the interval [t, t + i]. Let j = 0, 1, …, i, and recursively calculate to obtain M t , M t+1 , … M t+i .

[0135] The prediction of the water content sequence is actually a time series prediction problem. Under the rolling prediction setting with a fixed window size, the input of the model at time point t:

[0136]

[0137] In the formula: X t is the input of the model. L x represents the length of the current input sequence, represents the i-th in the window, the input vector at time point t, and each input is a vector of d x dimensions.

[0138] The output is the predicted corresponding sequence:

[0139]

[0140] In the formula: Y t is the output of the model. L y represents the length of the current output sequence, represents the i-th in the window, the output vector at time point t, and each output is a vector of d y dimensions.

[0141] In a standard recurrent neural network (RNN), as time goes by, the vanishing gradient problem makes it difficult for the network to learn long-term dependencies in the time series. While the long short-term memory network (LSTM) solves the vanishing gradient problem through a unique design structure, effectively maintaining long-term information, enabling the model to retain information in a very long sequence and make effective predictions. Such as Figure 3As shown in the figure, the LSTM unit mainly includes an input gate, a forget gate, an output gate, and a cell state that maintains the internal state.

[0142] In the LSTM for predicting long sequences, since the output at time t depends on the output at time t-1, the longer the predicted sequence, the worse the prediction speed and effect. At the same time, since the LSTM model uses the backpropagation algorithm to calculate the loss function to optimize the model parameters, it is difficult for the model to converge. Aiming at the problems of poor prediction effect and slow convergence of the LSTM model in long sequence prediction, an Informer model is tried to be applied to predict the moisture content.

[0143] As Figure 4 shown in the figure, the Informer model mainly consists of five parts: input, encoder, decoder, fully connected layer, and output. The model is built using some core components of the Transformer model (such as the encoder and decoder). It uses the multi-head sparse self-attention mechanism to calculate the degree of closeness between the current time point and the previous time points in parallel, assigns larger weights to the features of important time points, reduces the model complexity, and avoids serious information loss. At the same time, convolutional pooling operations are added between adjacent attention blocks to downsample the features, reduce the length of the input sequence, and greatly improve the speed.

[0144] The calculation of the multi-head sparse probabilistic self-attention mechanism is shown in formula (7), which contains three vectors, namely the Query vector (Q), the Key vector (K), and the Value vector (V).

[0145]

[0146] In the formula: L Q , L K and L V respectively represent the linear transformation layers of the query set, key set, and value set in the Informer module, and d is the input dimension.

[0147] In this embodiment, according to the trained moisture content prediction model, the detection of the moisture content of forest combustibles includes:

[0148] According to the input values of the reflectivity and scattering characteristics to be measured, perform prediction through the trained moisture content prediction model to obtain the corresponding moisture content prediction value;

[0149] Judge whether the moisture content prediction value exceeds the preset threshold range. If it exceeds, trigger the warning mechanism to indicate that the moisture content is abnormal;

[0150] According to the deviation between the moisture content prediction value and the actual moisture content, retrain the moisture content prediction model to continuously optimize the prediction accuracy of the model;

[0151] Associate the water content prediction results with the time curve to generate the time-varying curve of water content prediction, providing data support for subsequent water content trend analysis.

[0152] The embodiments described above are only descriptions of the preferred embodiments of the present invention and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for detecting moisture content of forest combustibles based on LiDAR technology, characterized in that: The method comprises: Acquire LiDAR scanning data of the target area, and extract point cloud information from the LiDAR scanning data, wherein the point cloud information includes: signal strength and spatial coordinates; According to the point cloud information, the reflectivity and scattering characteristics of each point are calculated to obtain a reflectivity distribution map and a scattering characteristic map; Determine the type of combustible material based on the reflectivity distribution map and scattering characteristic map combined with the preset combustible material classification model; Obtaining reflectivity and scattering characteristic reference values ​​corresponding to the type and growth stage of the combustible from a pre-established database, comparing the reference values ​​with actual measured values, and calculating reflectivity deviation values ​​and scattering characteristic deviation values; Obtaining precipitation data and transpiration data of the target area from a meteorological database, and calculating the moisture content variation trend in the time dimension according to the precipitation data and transpiration data to obtain a moisture content time variation curve; The corrected reflectivity and scattering characteristic values ​​are combined with the moisture content time variation curve, and input into a pre-trained machine learning model to train the association model between the reflectivity, scattering characteristic and moisture content, so as to obtain a trained moisture content prediction model; The moisture content of forest fuel is detected based on the trained moisture content prediction model.

2. The method according to claim 1, characterized in that Obtaining LiDAR scanning data of a target area, and extracting point cloud information from the scanning data includes: Obtain LiDAR scan data of the target area, preprocess the LiDAR scan data and extract point cloud data; According to the spatial range of the scanning area, the point cloud data is spatially divided into multiple sub-areas; According to the point cloud data of each sub-area, the three-dimensional spatial coordinates of the point cloud and the corresponding signal strength value are extracted to obtain the point cloud coordinates and strength information of the sub-area; The point cloud coordinates and intensity information of each sub-area are merged to obtain the complete point cloud data of the target area, including the three-dimensional coordinates and signal strength of the points.

3. The method according to claim 1, characterized in that According to the point cloud information, the reflectivity and scattering characteristics of each point are calculated to obtain a reflectivity distribution map and a scattering characteristic map including: Acquire the point cloud information, and for each point in the point cloud information, calculate the reflectivity and scattering characteristic value of the point; A reflectivity distribution graph is generated according to the calculated reflectivity value of each point, and different colors or grayscales in the reflectivity distribution graph represent different reflectivity value ranges; A scattering characteristic diagram is generated according to the calculated scattering characteristic value of each point, wherein different colors or grayscales in the scattering characteristic diagram represent different scattering characteristic value ranges; Perform image segmentation processing on the reflectivity distribution map to obtain the distribution position and range of different reflectivity areas; Perform image segmentation processing on the scattering characteristic map to obtain the distribution position and range of different scattering characteristic areas; Perform image registration on the reflectivity distribution map and the scattering characteristic map so that the positions of corresponding points in the two images coincide; The registered reflectivity distribution map and scattering characteristic map are fused to generate a fused point cloud attribute map. Different colors or grayscales represent different combinations of reflectivity and scattering characteristics.

4. The method according to claim 1, characterized in that: According to the reflectivity distribution map and scattering characteristic map, combined with the preset combustible classification model, the types of combustibles include: Obtain a reflectivity distribution map, and for each point in the reflectivity distribution map, extract the reflectivity value of the point; According to the preset combustible material classification model, determine whether the reflectivity value of the point is within the preset range; If the reflectivity value of the point is within the preset range, the combustible type corresponding to the point is determined through the combustible type classification model; If the reflectivity value of the point exceeds the preset range, the point is marked as an unknown type; Classify and judge the type of combustible material for each point in the reflectivity distribution map; According to the corresponding relationship between reflectivity value and combustible type, a combustible type classification model is constructed; The reflectivity distribution map after classification judgment is used as the combustible material type distribution map for subsequent combustible material identification and positioning.

5. The method according to claim 1, characterized in that Obtaining the reflectivity and scattering characteristic reference values ​​corresponding to the combustible type and growth stage from a pre-established database, comparing the reference values ​​with the actual measured values, and calculating the reflectivity deviation value and the scattering characteristic deviation value includes: Obtaining information on the type of combustibles and the growth stage of the combustibles, and obtaining reference reflectivity and reference scattering characteristic values ​​of the corresponding type and growth stage from a pre-established database; Obtain image information of the combustible area, obtain spectral data of each pixel point in the area according to the image information, and calculate the actual reflectivity and actual scattering characteristic value of each pixel point through the spectral data; According to the actual reflectivity and reference reflectivity of each pixel, the reflectivity deviation value of each pixel is calculated. If the reflectivity deviation value is greater than a preset threshold, it is marked; Calculate the scattering characteristic deviation value of each pixel point according to the actual scattering characteristic value and the reference scattering characteristic value of each pixel point, and mark it if the scattering characteristic deviation value is greater than a preset threshold value; A support vector machine algorithm is used to train according to the marked reflectivity deviation value and the marked scattering characteristic deviation value, so as to obtain a trained reflectivity deviation and scattering characteristic deviation classification model; Using a K-nearest neighbor algorithm, according to a reflectivity deviation and scattering characteristic deviation classification model, the unlabeled reflectivity deviation values ​​and the unlabeled scattering characteristic deviation values ​​are classified to determine the classification results of the reflectivity deviation values ​​and the scattering characteristic deviation values; The random forest algorithm is used to judge whether the classification results of the reflectivity deviation value and the scattering characteristic deviation value meet the preset conditions according to the classification results of the reflectivity deviation value and the scattering characteristic deviation value, and to obtain the accuracy of the classification results of the reflectivity deviation value and the scattering characteristic deviation value.

6. The method according to claim 1, characterized in that Obtaining precipitation data and transpiration data of the target area from the meteorological database, calculating the moisture content variation trend in the time dimension based on the precipitation data and transpiration data, and obtaining the moisture content time variation curve includes: According to the geographical location information of the target area, the historical precipitation data and evaporation data of the area are obtained from the meteorological database; Preprocess the acquired precipitation data and evaporation data to remove outliers and missing values; The time series analysis method is used to calculate the precipitation and evaporation at each time point to obtain the precipitation and evaporation data series in the time dimension; According to the precipitation and transpiration data series, the moisture content at each time point is calculated to obtain the moisture content data series in the time dimension; Carry out trend analysis on the moisture content data series and use moving average or exponential smoothing method to determine the changing trend of moisture content; According to the changing trend of moisture content, a moisture content time variation curve is generated to intuitively display the dynamic change process of moisture content.

7. The method according to claim 1, characterized in that The corrected reflectivity and scattering characteristic values ​​are combined with the moisture content time variation curve and input into a pre-trained machine learning model to train the association model between the reflectivity, scattering characteristic and moisture content. The trained moisture content prediction model includes: Acquire raw data of reflectivity and scattering characteristics, and correct the raw data according to a preset correction rule to obtain corrected reflectivity and scattering characteristic values; Obtaining curve data of water content changing with time, and associating the curve data with the corrected reflectivity and scattering characteristic values ​​to form a training data set; According to the input value and output value of the training data set, the associated model is trained to obtain a trained moisture content prediction model.

8. The method according to claim 1, characterized in that According to the trained moisture content prediction model, the moisture content detection of forest fuels includes: According to the reflectivity and scattering characteristic input values ​​to be measured, prediction is performed through the trained moisture content prediction model to obtain the corresponding moisture content prediction value; Determine whether the predicted moisture content value exceeds the preset threshold range. If it exceeds, the early warning mechanism is triggered to indicate that the moisture content is abnormal; According to the deviation between the predicted moisture content and the actual moisture content, the moisture content prediction model is retrained to continuously optimize the prediction accuracy of the model; The moisture content prediction results are associated with the time curve to generate a time variation curve of the moisture content prediction, providing data support for subsequent moisture content trend analysis.

Images