Early-stage intelligent identification method and device for large-area potential landslide

Abstract

The invention relates to a method and a device for early intelligent identification of a large-area potential landslide, and belongs to the technical field of image target identification and geological disaster monitoring, and the method comprises the steps: collecting historical data of a landslide area, obtaining surface deformation rate and gradient data, and forming a sample data set; a potential landslide identification model composed of a data input layer, a feature coding network and an identification result output layer is constructed, the feature coding network is composed of nine cascaded feature learning modules, and each feature learning module is composed of a network hiding layer and a data normalization layer; training and verifying the potential landslide identification model to obtain a final potential landslide identification model; obtaining actual measurement data of the monitoring area; and inputting measured data into the final potential landslide identification model, and outputting a potential landslide identification result of the monitoring area. According to the method, potential landslide candidate points and non-hidden-danger points in a large-area range are distinguished, and reliable technical support is provided for landslide disaster prevention and control.

Description

Technical Field

[0001] The invention relates to a method and a device for early intelligent identification of potential landslides in a large area, belonging to the technical field of image target identification and geological disaster monitoring. Background Art

[0002] Landslides, as a serious geological disaster, have a huge impact on the ecological environment and social economy. Therefore, early identification of landslides is crucial for the prevention, control and management of landslide disasters. Mountain deformation monitoring is the most direct and effective way to identify, warn, and make decisions about landslide disasters. By monitoring slope deformation, its stability can be evaluated and its movement state can be reflected. Traditional landslide monitoring methods include levels, GPS observation stations, crack meters, etc. However, traditional methods can only meet the monitoring of proven landslide points within a local area, and are inefficient, easily restricted by external conditions, and cannot accurately reflect the overall change characteristics and trends of regional-scale landslide morphology. They are especially powerless for remote landslides in remote areas and high altitudes.

[0003] Satellite synthetic aperture radar interferometry (InSAR) technology can overcome the above difficulties. The surface phase changes observed by InSAR data can be used to invert the sliding deformation characteristics of the slope body, which is an effective means of early identification of potential landslides in large areas. Permanent scatterer InSAR (PS-InSAR), small baseline set InSAR (SBAS-InSAR), multi-primary image coherent target small baseline InSAR (MCTSB-InSAR) or full scatterer InSAR (FS-InSAR) and other technologies are used to invert the surface deformation information of the differential interferometric phase. The results can intuitively reflect the deformation rate of objects and terrain in the observation scene. However, satellite SAR images cover a wide range, and the InSAR monitoring results contain not only the surface deformation information of the mountain, but also interference information such as flat ground subsidence and infrastructure deformation. It is difficult to distinguish them using the threshold method. In order to eliminate this interference information, people usually use GIS software to overlay InSAR monitoring results and optical images for manual interpretation and delineation of landslide areas, or use the delineated landslide area vector combined with optical images to further make a training sample set, and then train a deep learning image segmentation network to achieve the purpose of intelligent identification of landslide areas. This method is only applicable to the identification of landslides after disasters, and it is difficult to be applied to the accurate identification of potential landslides before disasters, because there are no obvious signs of landslides on optical images before disasters.

[0004] In summary, although InSAR technology is an effective means of observing surface deformation in large areas, the identification of potential landslides from InSAR monitoring results still requires manual interpretation, which is a time-consuming and laborious process. In order to reduce the cost of manual interpretation, the present invention provides a method and device for early intelligent identification of potential landslides in large areas. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a method and device for early intelligent identification of potential landslides in a large area, which can realize automatic distinction between potential landslide candidate points and non-hazardous points within a large area, and efficiently and accurately identify potential landslides in a large area.

[0006] The technical solution adopted by the present invention to solve the technical problem is: In a first aspect, an embodiment of the present invention provides a method for early intelligent identification of potential landslides in a large area, comprising the following steps: Step S1, collecting historical data of the landslide area, the historical data includes SAR satellite time series data and digital elevation model DEM, obtaining the surface deformation rate based on FS-InSAR technology, using the digital elevation model DEM to invert the slope data, and combining the location information of the registered landslide for interpretation, to form a sample data set including a training sample set and a verification sample set; Step S2, constructing a potential landslide identification model, the potential landslide identification model consists of a data input layer, a feature encoding network and an identification result output layer, the data input layer is used to receive a data combination (x1, x2) including a surface deformation rate x1 and a slope x2 as input data; the feature encoding network is composed of 9 cascaded feature learning modules, each feature learning module is composed of a network hidden layer and a data normalization layer, the network hidden layer is composed of m neurons, each neuron adaptively learns the fusion features of the input data; the data normalization layer adopts a one-dimensional batch normalization operation with learnable parameters; Step S3, training and verifying the potential landslide identification model based on the training sample set and the verification sample set to obtain a final potential landslide identification model; Step S4, collecting SAR satellite time series data and digital elevation model DEM of the monitoring area, obtaining the surface deformation rate based on FS-InSAR technology, and calculating the slope data using the digital elevation model DEM to form measured data; Step S5, inputting the measured data into the final potential landslide identification model, and outputting the potential landslide identification result of the monitoring area.

[0007] As a possible implementation of this embodiment, the output feature of the hidden layer of the network is expressed as: , In the formula, is the ReLU activation function; Is the connection l The weight of the i-th neuron in the hidden layer and the j-th neuron in the l+1-th hidden layer, For the l+1 The bias of the jth neuron in the hidden layer; j, i represent the neuron indexes of the current hidden layer and the previous hidden layer, respectively, j = 1, 2…m l+1 , m l+1 For the l+1 The number of neurons in the hidden layer, i=1, 2…m l ;when l =1, 2…9, For the l Hidden layer i The output features of neurons, m l For the l The number of neurons in the hidden layer; when l =0, Represents the input data x of the data input layer, namely the surface deformation rate and slope. At this time, m l is the dimension of the input data, i.e. m l =2.

[0008] As a possible implementation of this embodiment, the data normalization layer performs one-dimensional batch normalization processing on the data features output by the hidden layer, and its output value is expressed as: , , In the formula, is the output value of batch normalization for the jth feature of the kth sample, γ j and β j is the parameter to be learned; k represents the sample index, k =1,2,…, K ; μ j Indicates j Features about the number of samples K The average value of ; Indicates j The variance of the features, ; ε is a small constant that prevents the denominator from being zero.

[0009] As a possible implementation of this embodiment, the recognition result output layer performs one-dimensional batch normalization processing on the output features of the last hidden layer, performs linear weighted summation in the output layer, and uses the Sigmoid activation function to calculate the probability of the summation result. The calculated value y of the output layer is expressed as: , In the formula, is the Sigmoid activation function, is the weight matrix of the output layer, M L is the number of neurons in the 9th hidden layer; b (L) is the bias vector of the 9th hidden layer.

[0010] As a possible implementation of this embodiment, the sample data set is generated by using the surface deformation rate and slope data of the corresponding survey area generated by FS-InSAR technology, combined with an existing registered landslide hazard point database for discrimination. The sample data set consists of a certain number of positive samples (landslide hazard points) and negative samples (non-landslide hazard points). The label of the positive sample is represented by 1, and the label of the negative sample is represented by 0.

[0011] As a possible implementation of this embodiment, the training step of the potential landslide identification model includes: Set model hyperparameters and initialize network parameters for each neuron in each hidden layer; The potential landslide identification model is trained by using a small batch sample supervised learning mechanism, and the network parameters are updated iteratively; Among them, the model hyperparameters include the number of hidden layers and the number of neurons in each hidden layer. The number of hidden layers is set to 9, and the number of neurons in each hidden layer is 2048, 1024, 512, 256, 128, 64, 32, 16 and 8 respectively; the network parameters refer to weights and biases. The weights are initialized to random numbers that obey a normal distribution with a mean of 0 and a variance of , and the biases are initialized to 0.

[0012] As a possible implementation of this embodiment, the specific process of training using a small batch sample supervised learning mechanism includes: Set the training termination conditions, including the total number of training cycles, the error threshold and tolerance for verification recognition accuracy, the initial learning rate and the number of samples in the mini-batch input; The training sample set is used in batches in each cycle, and the model is iteratively trained using the Adam optimizer and the learning rate adaptive update strategy; The binary cross entropy function is used as the target loss function to calculate the error between the potential landslide identified by the model and the sample label after each training cycle. Using the validation sample set, calculate the validation accuracy of the model in identifying potential landslides after each training cycle; Determine whether the training cycle has reached the preset total number of cycles or the error threshold of the verification accuracy. If so, save the trained model; otherwise, continue training.

[0013] As a possible implementation method of this embodiment, the learning rate adaptive update strategy is to determine whether the verification accuracy of the model in each training cycle is improved compared with the previous training cycle. If not, the count is increased by 1, otherwise the count is set to 0; if the count reaches a preset tolerance, the learning rate is updated.

[0014] As a possible implementation of this embodiment, the measured data is a vector composed of InSAR surface deformation rate and slope data, and the potential landslide identification result is obtained by using a trained model to identify the measured data and output a result of 0 or 1 (1 indicates a potential landslide, and 0 indicates a non-potential landslide).

[0015] In a second aspect, an embodiment of the present invention provides a device for early intelligent identification of potential landslides in a large area, comprising: The data set construction module is used to collect historical data of the landslide area, including SAR satellite time series data and digital elevation model DEM, obtain the surface deformation rate based on FS-InSAR technology, invert the slope data using the digital elevation model DEM, and interpret the location information of the registered landslides to form a sample data set including a training sample set and a verification sample set; A model building module is used to build a potential landslide identification model, the potential landslide identification model is composed of a data input layer, a feature encoding network and an identification result output layer, the data input layer is used to receive a data combination (x1, x2) including a surface deformation rate x1 and a slope x2 as input data; the feature encoding network is composed of 9 cascaded feature learning modules, each feature learning module is composed of a network hidden layer and a data normalization layer, the network hidden layer is composed of m neurons, each neuron adaptively learns the fusion features of the input data; the data normalization layer adopts a one-dimensional batch normalization operation with learnable parameters; A model training module, used for training and verifying the potential landslide identification model based on the training sample set and the verification sample set to obtain a final potential landslide identification model; The measured data acquisition module is used to collect SAR satellite time series data and digital elevation model DEM in the monitoring area, obtain the surface deformation rate based on FS-InSAR technology, calculate the slope data using the digital elevation model DEM, and form measured data; The model identification module is used to input the measured data into the final potential landslide identification model and output the potential landslide identification result of the monitoring area.

[0016] The beneficial effects of the technical solution of the embodiment of the present invention are as follows: The present invention constructs a lightweight potential landslide identification model LSNet, which comprehensively utilizes the InSAR surface deformation rate and slope information, autonomously learns the multi-dimensional difference characteristics of potential landslide candidate points and other interference targets (ground subsidence, buildings, infrastructure and other deformations), and realizes the automatic distinction between potential landslide candidate points and non-hazardous points in a large area, efficiently and accurately identifies potential landslides in a large area, reduces the cost of manual interpretation, and improves the prevention and control and management efficiency of landslide disasters. The present invention is suitable for the early identification of potential landslides in a large area, and provides reliable technical support for the prevention and control of landslide disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of a method for early intelligent identification of potential landslides in a large area according to an exemplary embodiment; Figure 2 is a schematic diagram of a device structure for early intelligent identification of potential landslides in a large area according to an exemplary embodiment; Figure 3 is a network structure diagram of a potential landslide identification model according to an exemplary embodiment; Figure 4 It is a specific implementation flow chart of the present invention for early intelligent identification of potential landslides in a large area. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the technical features of the solution of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0019] like Figure 1 As shown, a method for early intelligent identification of potential landslides in a large area provided by an embodiment of the present invention includes the following steps: Step S1, collecting historical data of the landslide area, the historical data includes SAR satellite time series data and digital elevation model DEM, obtaining the surface deformation rate based on FS-InSAR technology, using the digital elevation model DEM to invert the slope data, and combining the location information of the registered landslide for interpretation, to form a sample data set including a training sample set and a verification sample set; Step S2, constructing a potential landslide identification model, the potential landslide identification model consists of a data input layer, a feature encoding network and an identification result output layer, the data input layer is used to receive a data combination (x1, x2) including a surface deformation rate x1 and a slope x2 as input data; the feature encoding network is composed of 9 cascaded feature learning modules, each feature learning module is composed of a network hidden layer and a data normalization layer, the network hidden layer is composed of m neurons, each neuron adaptively learns the fusion features of the input data; the data normalization layer adopts a one-dimensional batch normalization operation with learnable parameters; Step S3, training and verifying the potential landslide identification model based on the training sample set and the verification sample set to obtain a final potential landslide identification model; Step S4, collecting SAR satellite time series data and digital elevation model DEM of the monitoring area, obtaining the surface deformation rate based on FS-InSAR technology, and calculating the slope data using the digital elevation model DEM to form measured data; Step S5, inputting the measured data into the final potential landslide identification model, and outputting the potential landslide identification result of the monitoring area.

[0020] The present invention comprehensively utilizes the InSAR surface deformation rate and slope information to improve the accuracy of potential landslide identification; by constructing a lightweight potential landslide identification model, the automatic distinction between potential landslide candidate points and non-hazardous points in a large area is realized, the accuracy and efficiency of potential landslide identification are improved, and the cost of manual interpretation is reduced; a one-dimensional Batch normalization operation is added between each hidden layer, the fusion characteristics of the input data can be adaptively learned, and the stability and generalization ability of the potential landslide identification model are improved; a small batch sample supervised learning mechanism and an adaptive learning rate adjustment strategy are adopted during the training process to improve the convergence speed and recognition accuracy of the model.

[0021] As a possible implementation of this embodiment, the output feature of the network hidden layer in the feature learning module is expressed as: , In the formula, is the ReLU activation function; Is the connection l The weight of the i-th neuron in the hidden layer and the j-th neuron in the l+1-th hidden layer, For the l+1 The bias of the jth neuron in the hidden layer; j, i represent the neuron indexes of the current hidden layer and the previous hidden layer, respectively, j = 1, 2…m l+1 , m l+1 For the l+1 The number of neurons in the hidden layer, i=1, 2…ml ;when l =1, 2…9, For the l Hidden layer i The output features of neurons, m l For the l The number of neurons in the hidden layer; when l =0, Represents the input data x of the data input layer, namely the surface deformation rate and slope. At this time, m l is the dimension of the input data, i.e. m l =2.

[0022] As a possible implementation of this embodiment, the data normalization layer performs one-dimensional batch normalization processing on the data features output by the hidden layer, so that the statistical distribution of the input data of the subsequent hidden layer is more stable, and its output value is expressed as: , , In the formula, is the output value of batch normalization for the jth feature of the kth sample, γ j and β j is the parameter to be learned; k represents the sample index, k =1,2,…, K ; μ j Indicates j Features about the number of samples K The average value of ; Indicates j The variance of the features, ; ε is a small constant that prevents the denominator from being zero.

[0023] As a possible implementation of this embodiment, the recognition result output layer performs one-dimensional batch normalization (1D-Batch normalization) on the output features of the last hidden layer, performs linear weighted summation in the output layer, and uses the Sigmoid activation function to calculate the probability of the summation result, that is, to calculate each group of input data ( x 1, x 2) The probability of being a potential landslide candidate point, the calculated value y of the output layer is expressed as: , In the formula, is the Sigmoid activation function, is the weight matrix of the output layer, M L is the number of neurons in the 9th hidden layer; b (L) is the bias vector of the 9th hidden layer.

[0024] As a possible implementation of this embodiment, the sample data set is generated by using the surface deformation rate and slope data of the corresponding survey area generated by FS-InSAR technology, combined with the existing registered landslide risk point database for discrimination. The training sample set and the verification sample set are divided according to a 7:3 ratio. The training sample set is used to supervise the training of the potential landslide identification model, and the verification sample set is used to evaluate the potential landslide identification accuracy of the potential landslide identification model after each training cycle. The sample data set consists of a certain number of positive samples (landslide risk points) and negative samples (non-landslide risk points). The label of the positive sample is represented by 1, and the label of the negative sample is represented by 0.

[0025] As a possible implementation of this embodiment, the training step of the potential landslide identification model includes: Set model hyperparameters and initialize network parameters for each neuron in each hidden layer; The potential landslide identification model is trained by using a small batch sample supervised learning mechanism, and the network parameters are updated iteratively; Among them, the model hyperparameters include the number of hidden layers and the number of neurons in each hidden layer. The number of hidden layers is set to 9, and the number of neurons in each hidden layer is 2048, 1024, 512, 256, 128, 64, 32, 16 and 8 respectively; the network parameters refer to weights and biases. The weights are initialized to random numbers that obey a normal distribution with a mean of 0 and a variance of , and the biases are initialized to 0.

[0026] As a possible implementation of this embodiment, the specific process of training using a small batch sample supervised learning mechanism includes: Set the training termination conditions, including the total number of training cycles (such as 200), the verification recognition accuracy error threshold (such as 10 -6 ) and tolerance (such as 20), set the initial learning rate lr (default value is 0.001) and the number of samples in the mini-batch input k ; The training sample set is used in batches in each cycle, and the Adam optimizer and the learning rate adaptive update strategy are used to iteratively train the model. The number of batches in each cycle is calculated based on the integer ratio between the number of training samples and the number of small batch input samples, that is, T=[Q1 / k], where Q1 is the number of samples in the training sample set, and [ ] represents the value.

[0027] The binary cross entropy function is used as the target loss function to calculate the error between the potential landslide identified by the model and the sample label after each training cycle. Using the validation sample set, calculate the validation accuracy of the model in identifying potential landslides after each training cycle; Determine whether the training cycle has reached the preset total number of cycles or the error threshold of the verification accuracy. If so, save the trained model; otherwise, continue training.

[0028] As a possible implementation of this embodiment, the learning rate adaptive update strategy is to determine whether the verification accuracy of the model in each training cycle is improved compared with the previous training cycle. If not, the count is increased by 1, otherwise the count is set to 0; if the count reaches a preset tolerance, the learning rate is updated. , continue training the LSNet network model.

[0029] As a possible implementation of this embodiment, the measured data is a batch or a group of vectors composed of InSAR surface deformation rate and slope data, and the potential landslide identification result is obtained by using a trained model to identify the measured data and output a result of 0 or 1 (1 indicates a potential landslide, and 0 indicates a non-potential landslide).

[0030] like Figure 2 As shown, an embodiment of the present invention provides a device for early intelligent identification of potential landslides in a large area, comprising: The data set construction module is used to collect historical data of the landslide area, including SAR satellite time series data and digital elevation model DEM, obtain the surface deformation rate based on FS-InSAR technology, invert the slope data using the digital elevation model DEM, and interpret the location information of the registered landslides to form a sample data set including a training sample set and a verification sample set; A model building module is used to build a potential landslide identification model, the potential landslide identification model is composed of a data input layer, a feature encoding network and an identification result output layer, the data input layer is used to receive a data combination (x1, x2) including a surface deformation rate x1 and a slope x2 as input data; the feature encoding network is composed of 9 cascaded feature learning modules, each feature learning module is composed of a network hidden layer and a data normalization layer, the network hidden layer is composed of m neurons, each neuron adaptively learns the fusion features of the input data; the data normalization layer adopts a one-dimensional batch normalization operation with learnable parameters; A model training module, used for training and verifying the potential landslide identification model based on the training sample set and the verification sample set to obtain a final potential landslide identification model; The measured data acquisition module is used to collect SAR satellite time series data and digital elevation model DEM in the monitoring area, obtain the surface deformation rate based on FS-InSAR technology, calculate the slope data using the digital elevation model DEM, and form measured data; The model identification module is used to input the measured data into the final potential landslide identification model and output the potential landslide identification result of the monitoring area.

[0031] The present invention comprehensively utilizes the InSAR deformation rate and slope information to construct a neural network model LSNet for identifying potential landslide candidate points, which can automatically distinguish the potential landslide deformation in the InSAR deformation monitoring results from the deformation of other targets.

[0032] like Figure 3 As shown in Figure 1, the LSNet network is a simple classification network model. It is a fully connected network consisting of 11 network layers, including 1 input layer, 9 hidden layers and 1 output layer. The input layer contains a set of InSAR deformation rate x 1 and slope x 2 data combination, that is ( x 1, x 2). The number of neurons in each hidden layer is 2048, 1024, 512, 256, 128, 64, 32, 16 and 8 respectively. For ease of description, assume that each hidden layer consists of M neurons, each neuron adaptively learns the fusion features of the input data, that is, the excitation of the weighted combination of the input data. If the weight from the input layer to the first hidden layer is defined as W (1) , whose dimensions are M × N (Here N =2, which is the number of input data features), the bias vector is , then for the first hidden layer j Neurons ( j =1,2,…,m l ) output features It is expressed as: (1), In the formula, is the ReLU activation function.

[0033] Multiple neurons can be used to learn multiple features of input data. Through the deep and orderly cascading of multiple hidden layers, higher-level abstract semantic features can be generated, which helps the network capture the deep-level difference features in the input data that can effectively reflect the potential landslide candidate points and other targets. Then, for the first l +1 hidden layer j Neurons (j =1,2,…, m l +1), the output feature is ,Right now: (2), In formula (2), is the ReLU activation function; Is the connection l The weight of the i-th neuron in the hidden layer and the j-th neuron in the l+1-th hidden layer, For the l+1 The bias of the jth neuron in the hidden layer; j, i represent the neuron indexes of the current hidden layer and the previous hidden layer, respectively, j = 1, 2…m l+1 , m l+1 For the l+1 The number of neurons in the hidden layer, i=1, 2…m l ;when l =1, 2…9, For the l Hidden layer i The output features of neurons, m l For the l The number of neurons in the hidden layer; when l =0, Represents the input data x of the data input layer, namely the surface deformation rate and slope. At this time, m l is the dimension of the input data, i.e. m l =2.

[0034] In general, the slope value is always greater than 0, and its value range is [0,90]. The values ​​of the surface deformation rate monitored by InSAR are both positive and negative (negative values ​​indicate surface subsidence; positive values ​​indicate surface uplift). It can be seen that there are usually obvious differences in the values ​​of InSAR surface deformation rate and slope. Then, when using the sample set composed of these two types of data for network training, the statistical distribution of the output characteristics of each neuron in the hidden layer usually has obvious differences. The output of the previous hidden layer is also the input of the next hidden layer, which means that the statistical distribution of the input data of the next hidden layer will also have obvious differences, which can easily lead to the instability of the update of model parameters (i.e., weights and biases) during network training, and it is difficult to converge to the global optimal solution. This is because the model is always adapting to different distributions during network training, and it is easy to fall into a local optimal solution. On the other hand, the network model will overfit these inconsistent distribution characteristics, thereby reducing its prediction ability for unseen data, that is, reducing the generalization ability of the network model. In order to solve the above problems, we added a one-dimensional batch normalization operation with learnable parameters between each hidden layer. Batch normalization adaptively normalizes the data features of each hidden layer output, making the statistical distribution of the input data of each hidden layer more stable. k Samples ( x 1, x 2), the hidden layer will produce a dimension of k × m The output features of , then use one-dimensional Batch normalization for each feature j (that is, each feature in the m dimensions) is batch normalized, and its output value It is expressed as: (3), (4), In the formula, γ j and β j is the parameter to be learned; k represents the sample index, k =1,2,…, K ; μ j Indicates j Features about the number of samples K The average value of ; Indicates j The variance of the features, .

[0035] Finally, the output features of the last hidden layer (i.e., the 9th hidden layer) are processed by one-dimensional Batchnormalization, and then linear weighted summation is performed in the output layer, and the Sigmoid activation function is used to calculate the probability of the summation result, that is, the probability of each group of input data ( x 1, x 2) The probability of being a potential landslide candidate point. Assume that the last hidden layer contains M L neurons, whose weight matrix to the output layer (single neuron) is W (L) , whose dimensions are M L × q (Here q =1), the bias is b (L) , then the calculated value of the output layer is y for: (5), In the formula, is the Sigmoid activation function.

[0036] like Figure 4 As shown in the figure, during the specific implementation process, the specific implementation process of early intelligent identification of potential landslides in a large area is as follows.

[0037] 1. Build a sample dataset.

[0038] The sample data set is generated by using the surface deformation rate and slope data of the corresponding survey area generated by FS-InSAR technology, combined with the existing registered landslide risk point database for discrimination. The sample data set consists of a certain number of positive samples (landslide risk points) and negative samples (non-landslide risk points). The label of the positive sample is represented by 1, and the label of the negative sample is represented by 0.

[0039] Based on the FS-InSAR technology, the surface deformation rate is obtained from the SAR satellite remote sensing historical time series data of the landslide area, and the slope data is obtained from the terrain data of the landslide area using the digital elevation model DEM. The sample data set is composed of the surface deformation rate, slope and their corresponding labels. The sample data set is divided into a training sample set and a validation sample set in a ratio of 7:3. The training sample set is used to supervise the training of the potential landslide identification model, and the validation sample set is used to evaluate the potential landslide identification accuracy of the potential landslide identification model after each training cycle.

[0040] 2. Build the LSNet network model.

[0041] The LSNet network model constructed for early intelligent identification of potential landslides in large areas is as follows: Figure 3As shown; the potential landslide identification model consists of a data input layer, a feature encoding network and an identification result output layer, wherein the data input layer is used to receive a data combination including surface deformation rate and slope as input data; the feature encoding network consists of 9 cascaded feature learning modules, each feature learning module consists of a network hidden layer and a data normalization layer, the network hidden layer consists of m neurons, and each neuron adaptively learns the fusion features of the input data; the data normalization layer adopts a one-dimensional batch normalization operation with learnable parameters.

[0042] 3. Train the LSNet network model.

[0043] First, set the LSNet neural network model hyperparameters and initialize the network parameters of each neuron in each hidden layer; LSNet neural network model hyperparameters refer to the number of hidden layers and the number of neurons in each hidden layer. Set the number of hidden layers to 9, and the number of neurons in each hidden layer to 2048, 1024, 512, 256, 128, 64, 32, 16, and 8, respectively. The network parameters refer to the weights and biases in the LSNet neural network. The weight W is initialized to obey the mean of 0 and the variance of A normally distributed random number is used, and the bias b is initialized to 0.

[0044] Then, the LSNet neural network is trained using a small batch sample supervised learning mechanism, and the network parameters are iteratively updated; the specific process includes: 31. Set the training termination condition (i.e. the total number of training cycles (e.g. 200), the verification recognition accuracy error threshold is 10 -6 ), tolerance is 20, network parameter initial learning rate lr The default value is 0.001, the number of samples input in a small batch k ; 32. Using the training sample set, the LSNet neural network is iteratively trained in batches in each cycle using the Adam optimizer and the learning rate adaptive update strategy, and the network parameters in the LSNet neural network are continuously learned and updated; the number of batches in each cycle is calculated based on the integer relationship between the number of training samples and the number of small batch input samples, that is, T=[Q1 / k], where Q1 is the number of training samples and [ ] represents the value; the learning rate adaptive update strategy is to determine whether the verification accuracy of the LSNet network model in each training cycle is improved compared with the previous training cycle. If not, the count is increased by 1; otherwise, the count is set to 0; if the count reaches the preset tolerance, the learning rate is updated , continue training the LSNet network model; 33. A binary cross entropy function is used as the network target loss function to measure the error between the potential landslide identified by the LSNet network model and the sample label after each training cycle. 34. Using the verification sample set, calculate the verification accuracy of the LSNet network model in identifying potential landslides after each training cycle; 35. Determine whether the network training cycle reaches the preset total number of cycles or the error threshold of the verification accuracy. If so, exit and save the trained LSNet network model; otherwise, repeat steps 32-34.

[0045] 4. Obtain measured data and input the trained LSNet network model.

[0046] SAR satellite remote sensing time series data and terrain data of the monitoring area are collected, the surface deformation rate is obtained based on the FS-InSAR technology, the slope data is calculated using the digital elevation model DEM, and the measured data of the monitoring area is obtained; the measured data is a batch or a group of vectors composed of InSAR surface deformation rate and slope data.

[0047] 5. Output potential landslide identification results.

[0048] The measured data is input into the trained LSNet neural network model, which processes the measured data and outputs the potential landslide identification result corresponding to each measured data, which is 0 or 1 (1 indicates a potential landslide, and 0 indicates a non-potential landslide). In the specific implementation, the output results can also be displayed to the user in a visual form or stored in a database for subsequent analysis and processing.

[0049] The present invention comprehensively utilizes the InSAR surface deformation rate and slope information to improve the accuracy of potential landslide identification; by constructing a lightweight potential landslide identification model, the automatic distinction between potential landslide candidate points and non-hazardous points in a large area is realized, the accuracy and efficiency of potential landslide identification are improved, and the cost of manual interpretation is reduced; a one-dimensional Batch normalization operation is added between each hidden layer, the fusion characteristics of the input data can be adaptively learned, and the stability and generalization ability of the potential landslide identification model are improved; a small batch sample supervised learning mechanism and an adaptive learning rate adjustment strategy are adopted during the training process to improve the convergence speed and recognition accuracy of the model.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for early intelligent identification of potential landslides in a large area, characterized in that: The steps include: Step S1, collecting historical data of the landslide area, the historical data includes SAR satellite time series data and digital elevation model DEM, obtaining the surface deformation rate based on FS-InSAR technology, using the digital elevation model DEM to invert the slope data, and combining the location information of the registered landslide for interpretation, to form a sample data set including a training sample set and a verification sample set; Step S2, constructing a potential landslide identification model, wherein the potential landslide identification model is composed of a data input layer, a feature encoding network and an identification result output layer, wherein the data input layer is used to receive a data combination including a surface deformation rate and a slope as input data; The feature encoding network is composed of 9 cascaded feature learning modules, each of which is composed of a network hidden layer and a data normalization layer. The network hidden layer is composed of m neurons, and each neuron adaptively learns the fusion features of the input data. The data normalization layer uses a one-dimensional batch normalization operation with learnable parameters; Step S3, training and verifying the potential landslide identification model based on the training sample set and the verification sample set to obtain a final potential landslide identification model; Step S4, collecting SAR satellite time series data and digital elevation model DEM of the monitoring area, obtaining the surface deformation rate based on FS-InSAR technology, and calculating the slope data using the digital elevation model DEM to form measured data; Step S5, inputting the measured data into the final potential landslide identification model, and outputting the potential landslide identification result of the monitoring area.

2. The method for early intelligent identification of potential landslides in a large area according to claim 1 is characterized in that: The output feature of the hidden layer of the network is expressed as: ， In the formula, is the ReLU activation function; Is the connection l The weight of the i-th neuron in the hidden layer and the j-th neuron in the l+1-th hidden layer, For the l+1 The bias of the jth neuron in the hidden layer; j, i represent the neuron indexes of the current hidden layer and the previous hidden layer, respectively, j = 1, 2…m l+1 , m l+1 For the l+1 The number of neurons in the hidden layer, i=1, 2…m l ; when l =1, 2…9, For the l Hidden layer i The output features of neurons, m l For the l The number of neurons in the hidden layer; when l =0, Represents the input data x of the data input layer, namely the surface deformation rate and slope. At this time, m l is the dimension of the input data, i.e. m l =2.

3. The method for early intelligent identification of potential landslides in a large area according to claim 1 is characterized in that: The data normalization layer performs one-dimensional batch normalization processing on the data features output by the hidden layer, and its output value is expressed as: ， ， In the formula, is the output value of batch normalization for the jth feature of the kth sample, γ j and β j is the parameter to be learned; k represents the sample index, k =1,2,…, K ; μ j Indicates j Features about the number of samples K The average value of ; Indicates j The variance of the features, ; ε is a small constant that prevents the denominator from being zero.

4. The method for early intelligent identification of potential landslides in a large area according to claim 1 is characterized in that: The recognition result output layer performs one-dimensional batch normalization on the output features of the last hidden layer, performs linear weighted summation in the output layer, and uses the Sigmoid activation function to calculate the probability of the summation result. The calculated value y of the output layer is expressed as: ， In the formula, is the Sigmoid activation function, is the weight matrix of the output layer, M L is the number of neurons in the 9th hidden layer; b (L) is the bias of the 9th hidden layer.

5. The method for early intelligent identification of potential landslides in a large area according to claim 1 is characterized in that: The sample data set is generated by using the surface deformation rate and slope data of the corresponding survey area generated by FS-InSAR technology, combined with the existing registered landslide hazard point database for discrimination. The sample data set consists of a certain number of positive samples and negative samples, the label of the positive sample is represented by 1, and the label of the negative sample is represented by 0.

6. The method for early intelligent identification of potential landslides in a large area according to claim 1 is characterized in that: The training steps of the potential landslide identification model include: Set model hyperparameters and initialize network parameters for each neuron in each hidden layer; The potential landslide identification model is trained by using a small batch sample supervised learning mechanism, and the network parameters are updated iteratively; Among them, the model hyperparameters include the number of hidden layers and the number of neurons in each hidden layer. The number of hidden layers is set to 9, and the number of neurons in each hidden layer is 2048, 1024, 512, 256, 128, 64, 32, 16 and 8 respectively; the network parameters refer to weights and biases. The weights are initialized to random numbers that obey a normal distribution with a mean of 0 and a variance of , and the biases are initialized to 0.

7. The method for early intelligent identification of potential landslides in a large area according to claim 6 is characterized in that: The specific process of training using a small batch sample supervised learning mechanism includes: Set the training termination conditions, including the total number of training cycles, the error threshold and tolerance for verification recognition accuracy, the initial learning rate and the number of samples in the mini-batch input; The training sample set is used in batches in each cycle, and the model is iteratively trained using the Adam optimizer and the learning rate adaptive update strategy; The binary cross entropy function is used as the target loss function to calculate the error between the potential landslide identified by the model and the sample label after each training cycle. Using the validation sample set, calculate the validation accuracy of the model in identifying potential landslides after each training cycle; Determine whether the training cycle has reached the preset total number of cycles or the error threshold of the verification accuracy. If so, save the trained model; otherwise, continue training.

8. The method for early intelligent identification of potential landslides in a large area according to claim 7 is characterized in that: The learning rate adaptive update strategy is to determine whether the verification accuracy of the model in each training cycle is improved compared with the previous training cycle. If not, the count is increased by 1, otherwise the count is set to 0; if the count reaches a preset tolerance, the learning rate is updated.

9. The method for early intelligent identification of potential landslides in a large area according to any one of claims 1 to 8, characterized in that: The measured data is a vector composed of InSAR surface deformation rate and slope data. The potential landslide identification result is obtained by using a trained model to identify the measured data, and the output result is a potential landslide or a non-potential landslide.

10. A device for early intelligent identification of potential landslides in a large area, characterized in that: include: The data set construction module is used to collect historical data of the landslide area, including SAR satellite time series data and digital elevation model DEM, obtain the surface deformation rate based on FS-InSAR technology, invert the slope data using the digital elevation model DEM, and interpret the location information of the registered landslides to form a sample data set including a training sample set and a verification sample set; A model building module is used to build a potential landslide identification model, wherein the potential landslide identification model is composed of a data input layer, a feature encoding network and an identification result output layer, wherein the data input layer is used to receive a data combination including a surface deformation rate and a slope as input data; The feature encoding network is composed of 9 cascaded feature learning modules, each of which is composed of a network hidden layer and a data normalization layer. The network hidden layer is composed of m neurons, and each neuron adaptively learns the fusion features of the input data. The data normalization layer uses a one-dimensional batch normalization operation with learnable parameters; A model training module is used to train and verify the potential landslide identification model based on the training sample set and the verification sample set to obtain the final potential landslide identification model; The measured data acquisition module is used to collect SAR satellite time series data and digital elevation model DEM in the monitoring area, obtain the surface deformation rate based on FS-InSAR technology, calculate the slope data using the digital elevation model DEM, and form measured data; The model identification module is used to input the measured data into the final potential landslide identification model and output the potential landslide identification results in the monitoring area.

Images