Geological disaster monitoring and early warning method based on artificial intelligence

Abstract

The present invention relates to the field of geological disaster monitoring, and in particular to a geological disaster monitoring and early warning method based on artificial intelligence. The method comprises the steps of: obtaining a historical mountain video; obtaining a pre-constructed geological disaster detection network, wherein each convolution layer in the geological disaster detection network comprises a plurality of convolution kernel groups, and each convolution kernel group comprises a horizontal convolution kernel and a vertical convolution kernel; inputting a preset number of frames of historical mountain images in the historical mountain video into the geological disaster detection network, performing convolution processing using the horizontal convolution kernel to obtain a historical mountain image after horizontal convolution, and performing convolution processing using the vertical convolution kernel on all the historical mountain images after horizontal convolution obtained in the convolution kernel group to obtain a historical mountain image after vertical convolution; constructing a first loss function; and performing network training control using the first loss function to realize geological disaster monitoring. Accurate monitoring of geological disasters can be achieved by adjusting the network structure and the loss function.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster monitoring, and in particular to a geological disaster monitoring and early warning method based on artificial intelligence. Background Art

[0002] When a geological disaster occurs, if it is not discovered in time, it will affect people's life safety; if it is discovered in time and people are warned, they will have enough time to respond, which will greatly reduce the impact of geological disasters. Landslides are a common geological disaster, which poses a great threat to people's life safety when they occur. Therefore, it is necessary to monitor the mountain conditions where landslides often occur in real time to achieve early detection of landslide disasters.

[0003] When a landslide occurs, the stones and soil on the mountain will change. This feature can be collected through image information, so landslide detection can be performed through machine vision. As a common technology for machine vision, convolutional neural networks have high detection accuracy for single images. However, convolutional neural networks have poor description capabilities for the temporal variation characteristics of image information. At the same time, landslides are mainly reflected in the information on the changes in stones and soil on the mountain, so conventional convolutional neural networks have relatively poor detection accuracy for landslides. How to make convolutional neural networks learn the temporal variation characteristics of image information to achieve accurate landslide detection has become the research focus of this invention.

[0004] A patent application document with publication number CN117765688A discloses a remote sensing monitoring natural disaster data processing system. The method in the application document realizes natural disaster detection by comparing natural condition data with a preset disaster threshold value. The method in the application document does not involve the use of convolutional neural networks for landslide early warning. Therefore, the method in the application document cannot solve the problem in the present invention well. Summary of the invention

[0005] In order to solve the problem of how to let the convolutional neural network learn the temporal variation characteristics of image information, the present invention provides a geological disaster monitoring and early warning method based on artificial intelligence, which includes the following steps:

[0006] The mountain videos collected in history are obtained and recorded as historical mountain videos;

[0007] Obtaining a pre-constructed geological disaster detection network, wherein each convolution layer in the geological disaster detection network comprises a plurality of convolution kernel groups, and each convolution kernel group comprises a horizontal convolution kernel and a vertical convolution kernel;

[0008] Input a preset number of frames of historical mountain images in the historical mountain video into the geological disaster detection network, use the horizontal convolution kernel to perform convolution processing in the historical mountain image to obtain the historical mountain image after horizontal convolution, and use the vertical convolution kernel to perform inter-image convolution processing on all the historical mountain images after horizontal convolution obtained in the convolution kernel group to obtain the historical mountain image after vertical convolution; construct a first loss function , the sequence of all pixels at a position in the historical mountain image after all horizontal convolutions obtained by each convolution kernel group in each convolution layer is recorded as the analysis sequence of the position, Indicates Convolutional layer The sequence of autocorrelation values ​​of the analysis sequence at all positions of the convolution kernel group is composed of, Indicates Convolutional layer The sequence of all pixels in the historical mountain image after the longitudinal convolution of the convolution kernel group, represents the calculation of correlation, , Represent the number of convolution kernel groups and the number of convolution layers in each convolution layer respectively;

[0009] The first loss function is used to perform network training control to achieve geological disaster monitoring.

[0010] The present invention controls the geological disaster detection network to learn the time series variation characteristics of mountain image information by designing a first loss function, thereby improving the accuracy of the geological disaster detection network in monitoring landslides.

[0011] Preferably, the obtaining of a pre-built geological disaster detection network includes:

[0012] Constructing a geological disaster detection network comprising a plurality of convolutional layers, a plurality of downsampling layers and a plurality of fully connected layers, wherein the convolutional layers comprise a plurality of convolutional kernel groups and a preset number of longitudinal convolutional kernels;

[0013] Each convolution kernel group contains a preset number of horizontal convolution kernels and one vertical convolution kernel.

[0014] The present invention adjusts the network structure of the traditional convolutional neural network so that the adjusted network can learn the temporal variation characteristics of mountain image information, thereby improving the accuracy of the geological disaster detection network in monitoring landslides.

[0015] Preferably, a method for obtaining another loss function of the geological disaster detection network includes:

[0016] The image flows into the first convolution layer, and the horizontal convolution kernel of the convolution kernel group is used to obtain the historical mountain image after horizontal convolution. The vertical convolution kernel outside the convolution kernel group is used to perform inter-image convolution processing on all the historical mountain images after horizontal convolution corresponding to a frame of historical mountain image to obtain a comprehensive convolution image, which is downsampled and input into the next convolution layer; after completing the processing of each convolution layer, the comprehensive convolution image of the last convolution layer is obtained, and the sequence of downsampled images of the comprehensive convolution image of the last convolution layer and the sequence of downsampled images of the historical mountain images after the vertical convolution of each convolution layer are spliced ​​together and input into the fully connected layer, and the mean square error loss function is calculated using the output data and labels of the fully connected layer, which is recorded as the second loss function.

[0017] When designing the second loss function, the present invention not only utilizes the information learned in each image, but also utilizes the temporal information between images. The loss function designed in this way can enable the network to learn more comprehensive information, thereby improving the accuracy of the network.

[0018] Preferably, the using the first loss function to perform network training control includes:

[0019] The historical mountain video is divided into several image groups containing a preset number of frame images, the image groups are input into the geological disaster detection network, the corresponding loss values ​​are calculated using the first loss function and the second loss function, the network parameters in the geological disaster detection network are updated using the gradient descent method according to the corresponding loss values, the image groups are input into the geological disaster detection network in turn until the loss values ​​converge, and the training of the geological disaster detection network is completed.

[0020] The present invention utilizes the first loss function and the second loss function to train the geological disaster detection network, so as to control the geological disaster detection network to learn more information about landslides, thereby improving the accuracy of the geological disaster detection network in monitoring landslides.

[0021] Preferably, the performing convolution processing in the historical mountain image includes:

[0022] The transverse convolution kernel slides on the historical mountain image, and after each slide, the transverse convolution kernel is used to perform convolution processing on the pixels in the historical mountain image masked by the transverse convolution kernel.

[0023] Preferably, the performing of inter-image convolution processing includes:

[0024] An arbitrary pixel position is taken as the target position, and the pixel sequence at the target position in a preset number of frames of historical mountain images is recorded as the pixel sequence of the target position, and the pixel sequence of each position is obtained; the vertical convolution kernel in the convolution kernel group is used to perform convolution processing on the pixel sequence of each position.

[0025] The present invention allows the network to learn the change information of the image in time sequence by performing convolution processing between images, thereby improving the network's ability to describe landslide phenomena and improving the network's accuracy in monitoring geological disasters.

[0026] Preferably, the method of realizing geological disaster monitoring includes:

[0027] A preset number of newly collected mountain image frames are input into the trained geological disaster detection network to obtain the geological disaster detection result. If the geological disaster detection result is that there is a landslide, an alarm is issued; if the geological disaster detection result is that there is no landslide, no alarm is issued.

[0028] Preferably, the method for obtaining the autocorrelation value includes:

[0029] Perform autocorrelation detection on the sequence to obtain the autocorrelation value.

[0030] The present invention has the following technical effects:

[0031] The present invention controls the geological disaster detection network to learn the time series variation characteristics of mountain image information by designing a first loss function, thereby improving the description ability of the geological disaster detection network on landslide phenomena, thereby improving the accuracy of the geological disaster detection network on landslide monitoring;

[0032] Furthermore, the present invention adjusts the network structure of the traditional convolutional neural network so that the adjusted network can learn the temporal variation characteristics of mountain image information, thereby improving the accuracy of the geological disaster detection network in monitoring landslides. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0034] Figure 1 It is a flow chart of a method in a geological disaster monitoring and early warning method based on artificial intelligence according to an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of the geological disaster detection network provided by the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0038] The embodiment of the present invention discloses a geological disaster monitoring and early warning method based on artificial intelligence, referring to Figure 1 , comprising steps S1 to S4:

[0039] S1: Obtain the mountain videos collected in the past and record them as historical mountain videos.

[0040] Specifically, the mountain videos collected in history are obtained and recorded as historical mountain videos. In this embodiment, the mountain videos are collected by aerial photography. Other embodiments may use other methods to collect videos, and this embodiment does not specifically limit this.

[0041] The historical mountain video is divided into several image groups containing a preset number of frame images. Whether there is a landslide in the image group is manually determined. The label of the image group with a landslide is set to 1, and the label of the image group without a landslide is set to 0.

[0042] S2: Obtain a pre-constructed geological disaster detection network, wherein each convolution layer in the geological disaster detection network includes a plurality of convolution kernel groups, and each convolution kernel group includes a horizontal convolution kernel and a vertical convolution kernel.

[0043] It should be noted that in order to improve the detection accuracy of landslide disasters, the geological disaster detection network needs to learn the temporal change characteristics of mountain image information; in order to enable the geological disaster detection network to learn the temporal change characteristics of mountain image information, the network structure of the conventional convolutional neural network needs to be adjusted.

[0044] Preferably, as an example, obtaining a pre-built geological disaster detection network includes:

[0045] Build a geological disaster detection network, Figure 2The structure of the geological disaster detection network is shown in the schematic diagram of the geological disaster detection network. As shown in the figure, the first N1 layers of the geological disaster detection network are interleaved convolution layers and downsampling layers, and the N1+1 to N2 layers of the geological disaster detection network are fully connected layers. Each convolution layer includes Q convolution kernel groups and a preset number of longitudinal convolution kernels. Each convolution kernel group includes a preset number of transverse convolution kernels and one longitudinal convolution kernel. N1 and N2 represent the preset first number of layers and the preset second number of layers, respectively, and Q represents the preset number of groups. This embodiment is described by taking N1 as 16, N2 as 18, Q as 5, and the preset number as 20 as an example. Other embodiments may take other values, and this embodiment is not specifically limited.

[0046] It should be supplemented that, in the present invention, the convolution kernel that performs convolution processing within an image is referred to as a horizontal convolution kernel, and the convolution kernel that performs convolution processing between images is referred to as a vertical convolution kernel.

[0047] It should be noted that the description here mainly highlights the differences from the traditional convolutional neural network, and does not focus on the similarities. For example, each convolutional layer of the traditional neural network will use an activation function for activation processing after convolution, and the geological disaster detection network in the present invention is the same, and will also perform an activation operation. This is the same as the traditional convolutional neural network, so it will not be repeated.

[0048] It should be further explained that the traditional convolutional neural network only inputs one image at a time, so the traditional neural network can only learn the information of one image. In order to enable the geological disaster detection network to learn the changing characteristics of multiple mountain image information, the present invention adjusts the network structure of the traditional convolutional neural network, wherein by setting a convolution kernel group, each convolution kernel group contains not only a horizontal convolution kernel but also a vertical convolution kernel, wherein the vertical convolution kernel is used to learn the changing information of the mountain image in the frame sequence. The traditional convolutional neural network only has a horizontal convolution kernel to learn the information within a single image. Therefore, the geological disaster detection network obtained by this adjustment can not only learn the information within a single image, but also learn the changing information of the mountain image in the frame sequence.

[0049] S3: Input a preset number of frames of historical mountain images in the historical mountain video into the geological disaster detection network, use the horizontal convolution kernel to perform convolution processing in the historical mountain image to obtain the historical mountain image after horizontal convolution, and use the vertical convolution kernel to perform inter-image convolution processing on all the historical mountain images after horizontal convolution obtained in the convolution kernel group to obtain the historical mountain image after vertical convolution; construct a first loss function.

[0050] Preferably, as an example, after a preset number of frames of historical mountain images in the historical mountain video are input into the geological disaster detection network, the data flow process is as follows:

[0051] After a preset number of frames of historical mountain images are input into the geological disaster detection network, the preset number of frames of historical mountain images first flow into the first convolutional layer.

[0052] The horizontal convolution kernel in the convolution kernel group of the first convolution layer performs convolution processing on the historical mountain image to obtain the historical mountain image after horizontal convolution. The vertical convolution layer in the convolution kernel group of the first convolution layer performs inter-image convolution processing on all the historical mountain images after horizontal convolution obtained in the convolution kernel group to which it belongs to obtain the historical mountain image after vertical convolution. The historical mountain image after vertical convolution will flow into the first loss function calculation link. At the same time, the historical mountain image after vertical convolution will flow into the input link of the fully connected layer after downsampling.

[0053] The longitudinal convolution kernel outside the convolution kernel group performs inter-image convolution processing on all the horizontally convolved historical mountain images corresponding to a frame of historical mountain images to obtain a comprehensive convolution image of the frame of historical mountain images. The comprehensive convolution image will flow into the downsampling layer, and after the downsampling processing by the downsampling layer, it will flow into the next convolution layer. The next convolution layer completes the convolution processing of the convolution layer according to the same processing method as the convolution layer, and so on, completing the convolution processing of each convolution layer and the downsampling processing of each downsampling layer.

[0054] The image output by the last downsampling layer will be converted into a sequence, and the downsampled image of the historical mountain image after the longitudinal convolution of each convolution layer will be converted into a sequence. The sequence composed of the images output by the last downsampling layer will be spliced ​​together with the sequence composed of the downsampled images of the historical mountain images after the longitudinal convolution. The spliced ​​sequence flows into the fully connected layer, and the output data obtained through calculation and analysis by the fully connected layer flows into the second loss function calculation link.

[0055] The above involves the first loss function and the second loss function. The method for determining the first loss function and the second loss function is explained below.

[0056] It should be added that the convolution processing is performed in the historical mountain image, including: the horizontal convolution kernel slides on the historical mountain image, and after each slide, the horizontal convolution kernel is used to perform convolution processing on the pixels in the historical mountain image masked by the horizontal convolution kernel.

[0057] It needs to be further supplemented that the inter-image convolution processing includes:

[0058] An arbitrary pixel position is taken as the target position, and the sequence composed of pixels at the target position in a preset number of frames of historical mountain images is recorded as the pixel sequence of the target position, and the pixel sequence of each position is obtained; the vertical convolution kernel in the convolution kernel group is used to perform convolution processing on the pixel sequence of each position.

[0059] S30: construct a first loss function.

[0060] It should be noted that in order to improve the detection accuracy of landslide disasters, the geological disaster detection network needs to learn the temporal variation characteristics of mountain image information. Therefore, a loss needs to be designed to control the learning direction of the geological disaster detection network.

[0061] Preferably, as an example, constructing a first loss function includes:

[0062]

[0063] Among them, the sequence composed of all pixels at a position in the historical mountain image after all horizontal convolutions obtained by each convolution kernel group in each convolution layer is recorded as the analysis sequence of the position. Indicates Convolutional layer The sequence is composed of the autocorrelation values ​​of the analysis sequence of all positions of the convolution kernel group, which reflects the characteristic information of the change of the mountain image in time series. Indicates Convolutional layer The sequence of all pixels in the historical mountain images after the longitudinal convolution of the convolution kernel groups. This value reflects the information between the mountain images learned by the geological disaster detection network. represents the calculation of correlation, It reflects whether the information between mountain images learned by the geological disaster detection network can describe the changing characteristics of mountain images in time series. The larger the value, the stronger the ability of the information between mountain images learned by the geological disaster detection network to describe the changing characteristics of mountain images in time series. Therefore, the learning effect of the geological disaster detection network is better, and its loss value should be smaller. Represents the number of convolution kernel groups in each convolution layer, Indicates the number of convolutional layers. Represents the first loss function.

[0064] It should be added that the present invention uses the maximum information coefficient between the sequences to represent the correlation between the two sequences. The calculation method of the maximum information coefficient is a prior art and will not be described in detail here.

[0065] It should be noted that, in this embodiment, the autocorrelation value of the sequence is obtained through autocorrelation detection. The method of obtaining the autocorrelation value of the sequence through autocorrelation detection is a prior art and will not be described in detail here.

[0066] S31: Construct a second loss function.

[0067] Preferably, as an example, constructing the second loss function includes:

[0068] The mean square error loss function is calculated using the output data and labels of the fully connected layer, which is recorded as the second loss function.

[0069] It should be noted that the mean square error loss function is an existing loss function, and its specific calculation method will not be repeated here.

[0070] It should be further explained that the second loss function is used to control the accuracy of landslide detection by the geological disaster detection network.

[0071] S4: Using the first loss function to perform network training control to achieve geological disaster monitoring.

[0072] S40: Using the first loss function to perform network training control.

[0073] Preferably, as an example, using the first loss function to perform network training control includes:

[0074] The image group is input into the geological hazard detection network, the corresponding loss value is calculated using the first loss function and the second loss function, the network parameters in the geological hazard detection network are updated using the gradient descent method according to the corresponding loss value, and the image group is input into the geological hazard detection network in sequence until the loss value converges, thereby completing the training of the geological hazard detection network.

[0075] S41: To achieve geological disaster monitoring.

[0076] Preferably, as an example, to realize geological disaster monitoring, the method includes:

[0077] A preset number of newly collected mountain image frames are input into the trained geological disaster detection network to obtain the geological disaster detection result. If the geological disaster detection result is that there is a landslide, an alarm is issued; if the geological disaster detection result is that there is no landslide, no alarm is issued.

[0078] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

[0079] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A geological disaster monitoring and early warning method based on artificial intelligence, characterized in that: Includes steps: The mountain videos collected in history are obtained and recorded as historical mountain videos; Obtaining a pre-constructed geological disaster detection network, wherein each convolution layer in the geological disaster detection network comprises a plurality of convolution kernel groups, and each convolution kernel group comprises a horizontal convolution kernel and a vertical convolution kernel; Input a preset number of frames of historical mountain images in the historical mountain video into the geological disaster detection network, use the horizontal convolution kernel to perform convolution processing in the historical mountain image to obtain the historical mountain image after horizontal convolution, and use the vertical convolution kernel to perform inter-image convolution processing on all the historical mountain images after horizontal convolution obtained in the convolution kernel group to obtain the historical mountain image after vertical convolution; construct a first loss function , the sequence of all pixels at a position in the historical mountain image after all horizontal convolutions obtained by each convolution kernel group of each convolution layer is recorded as the analysis sequence of the position, Indicates Convolutional layer The sequence of autocorrelation values ​​of the analysis sequence at all positions of the convolution kernel group is composed of, Indicates Convolutional layer The sequence of all pixels in the historical mountain image after the longitudinal convolution of the convolution kernel group, represents the calculation of correlation, , Represent the number of convolution kernel groups and the number of convolution layers in each convolution layer respectively; The first loss function is used to perform network training control to achieve geological disaster monitoring.

2. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 1 is characterized in that: The step of obtaining a pre-built geological disaster detection network includes: Constructing a geological disaster detection network comprising a plurality of convolutional layers, a plurality of downsampling layers and a plurality of fully connected layers, wherein the convolutional layers comprise a plurality of convolutional kernel groups and a preset number of longitudinal convolutional kernels; Each convolution kernel group contains a preset number of horizontal convolution kernels and one vertical convolution kernel.

3. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 2 is characterized in that: Another method for obtaining the loss function of the geological disaster detection network includes: The image flows into the first convolution layer, and the horizontal convolution kernel of the convolution kernel group is used to obtain the historical mountain image after horizontal convolution. The vertical convolution kernel outside the convolution kernel group is used to perform inter-image convolution processing on all the historical mountain images after horizontal convolution corresponding to a frame of historical mountain image to obtain a comprehensive convolution image, which is downsampled and input into the next convolution layer; after completing the processing of each convolution layer, the comprehensive convolution image of the last convolution layer is obtained, and the sequence of downsampled images of the comprehensive convolution image of the last convolution layer and the sequence of downsampled images of the historical mountain images after the vertical convolution of each convolution layer are spliced ​​together and input into the fully connected layer, and the mean square error loss function is calculated using the output data and labels of the fully connected layer, which is recorded as the second loss function.

4. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 3 is characterized in that: The using the first loss function to perform network training control includes: The historical mountain video is divided into several image groups containing a preset number of frame images, the image groups are input into the geological disaster detection network, the corresponding loss values ​​are calculated using the first loss function and the second loss function, the network parameters in the geological disaster detection network are updated using the gradient descent method according to the corresponding loss values, the image groups are input into the geological disaster detection network in turn until the loss values ​​converge, and the training of the geological disaster detection network is completed.

5. The method for monitoring and early warning of geological disasters based on artificial intelligence according to claim 1 is characterized in that: The performing convolution processing in the historical mountain image includes: The transverse convolution kernel slides on the historical mountain image, and after each slide, the transverse convolution kernel is used to perform convolution processing on the pixels in the historical mountain image masked by the transverse convolution kernel.

6. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 1 or 3 is characterized in that: The performing of inter-image convolution processing comprises: An arbitrary pixel position is taken as the target position, and the pixel sequence at the target position in a preset number of frames of historical mountain images is recorded as the pixel sequence of the target position, and the pixel sequence of each position is obtained; the vertical convolution kernel in the convolution kernel group is used to perform convolution processing on the pixel sequence of each position.

7. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 4 is characterized in that: The above-mentioned geological disaster monitoring includes: A preset number of newly collected mountain image frames are input into the trained geological disaster detection network to obtain the geological disaster detection result. If the geological disaster detection result is that there is a landslide, an alarm is issued; if the geological disaster detection result is that there is no landslide, no alarm is issued.

8. The artificial intelligence-based geological disaster monitoring and early warning method according to claim 1 is characterized in that: The method for obtaining the autocorrelation value comprises: Perform autocorrelation detection on the sequence to obtain the autocorrelation value.

Images