Landslide identification method and device based on multi-model decision-level fusion

Abstract

The present invention provides a landslide identification method and device with multi-model decision-level fusion, the method comprising: obtaining a remote sensing image to be identified; inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model; wherein the landslide identification model comprises a decision maker and multiple learners, the multiple learners are used to respectively obtain the predicted categories of the remote sensing image to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified. The present invention realizes the intelligent extraction of landslide information in the drawdown zone, improves the work efficiency of landslide identification and the practical level of intelligent landslide identification.

Description

Technical Field

[0001] The invention relates to the technical field of landslide remote sensing identification, and in particular to a landslide identification method and device with multi-model decision-level fusion. Background Art

[0002] Remote sensing technology is an important technology in landslide identification. In view of the shortcomings of traditional satellite remote sensing and aerial remote sensing methods, such as difficulty in timely acquisition of remote sensing images during the low water level period, and difficulty in obtaining vertical elevation information of steep terrain by vertical downward remote sensing, a ship-borne remote sensing technology has emerged in recent years. That is, using a ship platform to obtain remote sensing images of the drawdown zone of the reservoir area from a horizontal perspective, using ship-borne images to build a three-dimensional model of the drawdown zone, and then using manual visual interpretation methods to identify landslides on the three-dimensional model. Manual visual interpretation methods have the disadvantages of being time-consuming, labor-intensive, and inefficient. How to automatically and efficiently identify landslides from remote sensing images has always been a hot and difficult issue in landslide remote sensing research.

[0003] With the rapid development of deep learning technology for remote sensing data, the use of deep learning methods for automatic identification of landslides has become the mainstream research direction in landslide remote sensing. Existing deep learning landslide identification methods mainly use remote sensing images with a vertical downward perspective. For shipborne images of the drawdown zone of the reservoir area, if they are processed into vertical orthophotos, there will be serious information loss in areas with steep terrain, resulting in missed landslides. With the development of three-dimensional point cloud deep learning technology, especially the introduction of deep learning models such as PointNet, it has become a theoretically feasible solution to process shipborne images to generate three-dimensional point clouds and then directly identify landslides from the point clouds. However, from the perspective of practical application, processing all shipborne data into three-dimensional point clouds and then using deep learning algorithms to identify landslides on the point clouds is a huge workload and has low practicality. Summary of the invention

[0004] The present invention provides a landslide identification method and device with multi-model decision-level fusion, which are used to solve the defect that the accuracy and practicability of landslide identification in the prior art cannot be achieved at the same time, and improve the automation level and practicability of landslide identification.

[0005] The present invention provides a landslide identification method based on multi-model decision-level fusion, comprising:

[0006] Acquire remote sensing images to be identified;

[0007] Inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0008] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0009] According to a multi-model decision-level fusion landslide identification method provided by the present invention, the remote sensing image to be identified includes a remote sensing image of a drawdown zone in a reservoir area acquired from a horizontal perspective.

[0010] According to a multi-model decision-level fusion landslide identification method provided by the present invention, the multiple learners are obtained by respectively sending the pre-processed sample data sets to different neural network models for training.

[0011] According to a landslide identification method based on multi-model decision-level fusion provided by the present invention, the data preprocessing includes:

[0012] Sample data set partitioning, normalization, encoding conversion, data augmentation, positive and negative sample concatenation, and increasing randomness.

[0013] According to a multi-model decision-level fusion landslide identification method provided by the present invention, the sample data set division includes: using a five-fold cross-validation method to divide the sample data set into 5 piles, each training selects one pile as a test set, and the other four piles as training sets.

[0014] According to a landslide identification method of multi-model decision-level fusion provided by the present invention, the neural network model includes:

[0015] VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit.

[0016] The present invention also provides a landslide identification device with multi-model decision-level fusion, comprising:

[0017] An acquisition module is used to obtain remote sensing images to be identified;

[0018] An identification module, used for inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0019] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0020] According to a multi-model decision-level fusion landslide identification device provided by the present invention, the remote sensing image to be identified includes a remote sensing image of a drawdown zone in a reservoir area acquired from a horizontal perspective.

[0021] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the landslide identification method of multi-model decision-level fusion as described above are implemented.

[0022] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the landslide identification method of multi-model decision-level fusion as described in any one of the above are implemented.

[0023] The present invention provides a multi-model decision-level fusion landslide identification method and device, which identifies images containing landslides from a large number of original shipborne images based on a deep learning target recognition method, and then roughly locates the landslide according to the original position data corresponding to the image. This technical process not only avoids the information loss caused by orthophotos, but also greatly reduces the workload of three-dimensional modeling, which is more in line with actual application needs and has a high practical level. At the same time, the present invention realizes the intelligent extraction of landslide information in the drawdown zone and improves the work efficiency of landslide identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a flow chart of a landslide identification method of model decision-level fusion provided by an embodiment of the present invention;

[0026] Figure 2 It is a flow chart of a method for training a landslide identification model provided by an embodiment of the present invention;

[0027] Figure 3 is a schematic diagram of the structure of a landslide identification device with decision-level fusion provided by an embodiment of the present invention;

[0028] Figure 4 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] Among the existing intelligent landslide identification methods, the use of remote sensing orthophotos with a vertical downward perspective has the problem of missing information, and the point cloud deep learning method has a huge amount of calculation and contains a large amount of invalid calculations. The practicality of these two methods for identifying landslides in the drawdown zone is relatively low. In the process of acquiring landslide data, the shooting distances of the landslide bodies in the samples vary greatly, the proportions of the landslide areas in the entire sample images vary greatly, the sample size is small, and the amount of positive and negative sample data is unbalanced. In response to this series of problems, the present invention has formulated a series of targeted method adjustments, which can greatly improve the accuracy of the recognition algorithm. An embodiment of the present invention proposes a landslide identification method with multi-model decision-level fusion, such as Figure 1 As shown, the following steps are included:

[0031] Step 101: Obtain a remote sensing image to be identified;

[0032] Step 102, inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0033] It should be noted that the categories include belonging to a landslide or not belonging to a landslide.

[0034] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0035] It should be noted that the learner is a neural network model trained by a sample data set.

[0036] The landslide identification method of multi-model decision-level fusion proposed in the embodiment of the present invention identifies images containing landslides from a large number of original shipborne images based on a deep learning target recognition method, and then roughly locates the landslide according to the original position data corresponding to the images. This technical process not only avoids the information loss caused by orthophotos, but also greatly reduces the workload of three-dimensional modeling. It is more in line with actual application needs and has high practicality.

[0037] In at least one embodiment of the present invention, the remote sensing image to be identified includes a remote sensing image of the drawdown zone of the reservoir area acquired from a horizontal perspective.

[0038] It should be noted that, in the embodiments of the present invention, the drawdown zone refers to a special area in rivers, lakes, and reservoirs where the flooded land periodically emerges from the water surface due to seasonal fluctuations in water levels, becoming a wetland.

[0039] In at least one embodiment of the present invention, the remote sensing image of the drawdown zone of the reservoir area obtained from the horizontal viewing angle is automatically cropped to ensure that the proportion of the bank slope area in each sample in the entire sample image is basically the same. Preferably, the bank slope area is automatically extracted using a region growing algorithm based on the image pixel values ​​in the remote sensing image sample, and the sample is automatically cropped.

[0040] In the process of sample processing of the remote sensing image to be identified in the embodiment of the present invention, due to the constraints of the ship's route and the winding river bank when shooting horizontally, the shooting distance of the landslide body in the sample photos varies greatly, and the proportion of the landslide area in the whole sample image varies greatly. Therefore, the embodiment of the present invention adopts the measure of cutting samples according to the proportion of the landslide area to ensure that the proportion of the landslide area in each sample in the whole sample image is basically the same.

[0041] In at least one embodiment of the present invention, the multiple learners are obtained by respectively sending the preprocessed sample data sets to different neural network models for training.

[0042] In at least one embodiment of the present invention, the decision maker also includes a fully connected layer, which uses a voting method to perform decision-level fusion. The voting method is suitable for classification tasks and votes on the prediction results of multiple learners, that is, the minority obeys the majority.

[0043] In at least one embodiment of the present invention, the data preprocessing includes:

[0044] Sample data set partitioning, normalization, encoding conversion, data augmentation, positive and negative sample concatenation, and increasing randomness.

[0045] It should be noted that the sample data set division specifically refers to:

[0046] For the division of training set, validation set and test set, after completing the data set division, generate npy files of positive and negative samples (.npy files are binary files dedicated to numpy) and npy files corresponding to sample labels.

[0047] In at least one embodiment of the present invention, the sample data set division includes: using a five-fold cross-validation method to divide the sample data set into 5 piles, and dividing the positive samples and negative samples into 5 piles respectively; selecting one pile as a test set and the other four piles as training sets for each training; repeating the previous step five times, and selecting a different training set each time.

[0048] It should be noted that the positive samples and negative samples include remote sensing images belonging to landslides and remote sensing images not belonging to landslides, respectively. Since the amount of data in the data set in the embodiment of the present invention is small, in order to reduce the under-learning state that may occur in model training, in addition to data enhancement, a five-fold cross validation is selected. In the cross validation, each sample data can be used as both training data and test data, which effectively solves the problem of small data volume.

[0049] The normalization specifically refers to:

[0050] The maximum and minimum value normalization processing method, the calculation formula is shown in formula (1):

[0051]

[0052] In the formula, x i Represents the pixel value of the image, min(x) and max(x) represent the minimum and maximum values ​​of the image pixels respectively. Image normalization makes the pixel value fall in the interval [0,1].

[0053] Normalization is to eliminate the impact of the dimensions between indicators to solve the comparability between data indicators. After the original data is processed by data standardization, each indicator is at the same order of magnitude for comprehensive comparative evaluation.

[0054] The encoding conversion refers to one-hot encoding conversion, and one-hot encoding is One-Hot encoding, and its method is to use N-bit state register to encode N states, each state has its independent register bit, and at any time, only one of them is valid. One-Hot encoding is the representation of categorical variables as binary vectors. This first requires mapping the categorical values ​​to integer values. Then, each integer value is represented as a binary vector, and except for the index of the integer, it is all zero value and is marked as 1. One-hot encoding solves the problem that the classifier in the embodiment of the present invention is not good at processing attribute data, and to a certain extent also plays a role in expanding features.

[0055] Data enhancement is a general term for methods of expanding data. Data enhancement can increase the samples of the training set, effectively alleviate the overfitting of the model, and also give the model stronger generalization ability. Image data enhancement usually has the following methods: horizontal flipping, vertical flipping, translation, rotation, brightness adjustment, and noise addition.

[0056] In order to solve the imbalance problem of positive and negative samples, it is necessary to enhance the data of positive samples. After the positive samples are divided into five piles, the four piles of data to be used for model training are enhanced by adding noise, flipping, rotating, etc. In order to avoid falsely high test results, the one pile of data used for testing is not processed.

[0057] The positive and negative sample splicing specifically refers to splicing the training data that has completed data enhancement in the positive sample with the training data in the negative sample to obtain the training set of each fold. Similarly, the test data in the positive sample and the test data in the negative sample are spliced ​​to obtain the test set of each fold. Since the amount of positive and negative sample data in the data set of the embodiment of the present invention is unbalanced, in order to reduce the sample balance problem, it is necessary to perform data enhancement on the positive samples in the model training data. In addition, we do not process the negative samples. After completing the enhancement of the positive samples, the positive and negative samples are spliced ​​and suffled to generate the final data set.

[0058] The purpose of increasing randomness is to avoid the influence of the order of data input on network training, increase randomness, improve the generalization performance of the network, avoid excessive gradients during weight update due to the appearance of regular data, and avoid overfitting or underfitting of the final model.

[0059] In at least one embodiment of the present invention, the neural network model includes:

[0060] VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit.

[0061] In the embodiment of the present invention, five models, namely, VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit, are trained respectively, and the corresponding accuracy and loss values ​​are obtained, as shown in Table 1. It can be seen that the five models used in the neural network model of the embodiment of the present invention perform well in landslide classification, and the accuracy rates are all over 92%.

[0062] Table 1 Comparison of accuracy of different neural network models

[0063] Acc Loss VGG19 92.59259104728699％ 0.30833032727241516 DenseNet121 93.33333373069763％ 0.20521868765354156 EffiientNetB0 92.59259104728699％ 0.23532849550247192 SEresNext50 93.33333373069763％ 0.42525672912597656 Vit 94.44444179534912％ 0.42417433857917786 average 93.259258270263672％ 0.219661715626716612

[0064] In order to improve the accuracy of image classification, the present invention performs decision-level model fusion, and adopts three types of result fusion methods: result majority voting (the voting method of the present invention), result direct averaging, and result weighted averaging. Result direct averaging improves the prediction ability of the overall model by fusing the average prediction level of each model. It does not emphasize the outstanding advantages of individual models, but can effectively make up for the shortcomings of some models. Finally, result weighted averaging is improved on the basis of result direct averaging. It expresses the contribution of each model to the fusion model result by adding weight parameters. Different weights can obtain different model fusion results, and then obtain different prediction results. Among them, the VIT model performed best with an accuracy rate of 94.44%.

[0065] As shown in Table 2, the three different decision fusions all achieved significant results, among which the result majority voting method adopted in the embodiment of the present invention has high classification accuracy and small loss, and has the best effect.

[0066] Table 2 Comparison of accuracy of different fusion methods

[0067] Acc Loss Result of majority vote 0.9481481481481482 0.8418248418248417 The results are directly averaged 0.9444444444444444 0.829004329004329 Weighted average of results 0.9481481481481482 0.8524808524808525

[0068] In at least one embodiment of the present invention, Figure 2 As shown, the training process of each neural network model is as follows:

[0069] Step a: Acquire shipborne image data;

[0070] Step b: data preprocessing;

[0071] Step c: model pre-training;

[0072] Transfer learning is used to transfer the pre-trained model parameters for training to improve the accuracy of small-scale training sets. The present invention uses the weights trained by ImageNet as a feature extractor and freezes all parameters outside the fully connected layer of the Convolutional Neutral Network (CNN) model.

[0073] Step d: Use the sample data set to train the pre-trained model to obtain the trained model as a learner;

[0074] After the model is preloaded, the training set is put into the model for training, the accuracy and loss value curves are plotted, and finally the accuracy of the model on the test set is calculated. After the model training is completed, the set of model weights with the best results on the validation set in each training is stored.

[0075] Step e: Model fusion.

[0076] The configurations of each learner training of the landslide identification model in the embodiment of the present invention are as follows:

[0077] The embodiment of the present invention adopts the Adam optimizer, wherein the learning rate is set to 0.001.

[0078] The formula of the cross entropy loss function used in the embodiment of the present invention is as follows:

[0079]

[0080] The embodiment of the present invention uses EarlyStopping and ModelCheckpoint as callback functions. EarlyStopping is a callback function used to stop training in advance, which can speed up learning and improve parameter adjustment efficiency. It can stop training when the cross entropy loss function on the training set no longer decreases (that is, the degree of decrease is less than a certain threshold). ModelCheckpoint is used to save the weight file after the model training is completed.

[0081] The present invention trains five models, namely, VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit, that is, for the test set in each fold of the five-fold cross validation, there are five pre-trained models respectively, and the results are added and voted after prediction to obtain the accuracy and loss value of this fold, and finally the five-fold average accuracy is obtained by calculation.

[0082] The landslide identification device of the multi-model decision-making level fusion provided by the present invention is described below. The landslide identification device of the multi-model decision-making level fusion described below and the landslide identification method of the multi-model decision-making level fusion described above can be referred to each other. Figure 3 As shown, the landslide identification device with multi-model decision-level fusion in this embodiment includes:

[0083] The acquisition module 301 is used to acquire the remote sensing image to be identified;

[0084] The identification module 302 is used to input the remote sensing image to be identified into the landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0085] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0086] The landslide identification device with multi-model decision-level fusion proposed in the embodiment of the present invention identifies images containing landslides from a large number of original shipborne images based on a deep learning target recognition method, and then roughly locates the landslide according to the original position data corresponding to the image. This technical process not only avoids the information loss caused by orthophotos, but also greatly reduces the workload of three-dimensional modeling, which is more in line with actual application needs and has high practicality.

[0087] In at least one embodiment of the present invention, the remote sensing image to be identified includes a remote sensing image of the drawdown zone of the reservoir area acquired from a horizontal perspective.

[0088] In at least one embodiment of the present invention, the multiple learners are obtained by respectively sending the preprocessed sample data sets to different neural network models for training.

[0089] In at least one embodiment of the present invention, the data preprocessing includes:

[0090] Sample data set partitioning, normalization, encoding conversion, data augmentation, positive and negative sample concatenation, and increasing randomness.

[0091] In at least one embodiment of the present invention, the sample data set division includes: using a five-fold cross-validation method to divide the sample data set into five piles, selecting one pile as a test set for each training and the other four piles as training sets.

[0092] In at least one embodiment of the present invention, the neural network model includes:

[0093] VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit.

[0094] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the landslide identification method of multi-model decision-level fusion, and the method includes:

[0095] Acquire remote sensing images to be identified;

[0096] Inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0097] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0098] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0099] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the landslide identification method of multi-model decision-level fusion provided by the above methods, the method comprising:

[0100] Acquire remote sensing images to be identified;

[0101] Inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0102] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0103] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the landslide identification method of multi-model decision-level fusion provided by the above methods, the method comprising:

[0104] Acquire remote sensing images to be identified;

[0105] Inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model;

[0106] The landslide identification model includes a decision maker and multiple learners, wherein the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified.

[0107] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0109] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A landslide identification method based on multi-model decision-level fusion, characterized in that: include: Acquire remote sensing images to be identified; Inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model; The landslide identification model includes a decision maker and multiple learners, the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified; The remote sensing image to be identified includes a remote sensing image of the drawdown zone of the reservoir area acquired from a horizontal perspective; The multiple learners are obtained by respectively sending the preprocessed sample data sets to different neural network models for training; The data preprocessing includes: Sample data set division, normalization, encoding conversion, data enhancement, positive and negative sample splicing and increasing randomness; The sample data set division includes: using a five-fold cross-validation method to divide the sample data set into five piles, selecting one pile as a test set for each training and the other four piles as training sets.

2. The landslide identification method based on multi-model decision-level fusion according to claim 1 is characterized in that: The neural network model includes: VGG19, DenseNet121, EffiientNetB0, SEresNext50 and Vit.

3. A landslide identification device with multi-model decision-level fusion, characterized in that: include: An acquisition module is used to obtain remote sensing images to be identified; An identification module, used for inputting the remote sensing image to be identified into a landslide identification model to obtain the category of the remote sensing image to be identified output by the landslide identification model; The landslide identification model includes a decision maker and multiple learners, the multiple learners are used to respectively obtain the predicted categories of the remote sensing images to be identified, and the decision maker is used to vote on the predicted categories of the multiple learners to obtain the category of the output remote sensing image to be identified; The remote sensing image to be identified includes a remote sensing image of the drawdown zone of the reservoir area acquired from a horizontal perspective; The multiple learners are obtained by respectively sending the preprocessed sample data sets to different neural network models for training; The data preprocessing includes: Sample data set division, normalization, encoding conversion, data enhancement, positive and negative sample splicing and increasing randomness; The sample data set division includes: using a five-fold cross-validation method to divide the sample data set into five piles, selecting one pile as a test set for each training and the other four piles as training sets.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the landslide identification method of multi-model decision-level fusion as claimed in claim 1 or 2 are implemented.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the landslide identification method of multi-model decision-level fusion as claimed in claim 1 or 2 are implemented.

Images