Geological disaster monitoring and early warning method based on AI vision

Abstract

The invention provides a geological disaster monitoring and early warning method based on AI vision, and relates to the technical field of geological disaster monitoring and early warning, and the method comprises the steps: obtaining geological disaster pictures, including daytime pictures and night pictures; marking the pictures to form a geological disaster picture data set; training an AI model through the picture data set to obtain a trained AI model; carrying out lightweight processing on the trained AI model, and putting the AI model into an edge computer; importing the shot pictures into an AI model in an edge computer to identify geological disaster features, identifying the geological disaster features by using a previous and later picture rolling comparison algorithm, and performing mutual verification on two results; and through mutual verification of the two results, the occurrence of the geological disaster or the geological disaster precursor is judged, and an early warning is given out. According to the invention, accurate monitoring and early warning of geological disasters are realized, and timeliness and effectiveness of disaster response are improved.

Description

Technical Field

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

[0002] Geological disasters such as landslides and collapses are one of the major natural disasters in my country, causing hundreds of deaths every year. Monitoring and early warning are the most effective defense measures.

[0003] At present, there are still some shortcomings in my country's current geological monitoring and early warning. The casualties caused by geological disasters every year also prove the need for further improvement to reduce underreporting.

[0004] The first change of a slope is the occurrence of local cracks. For example, if there is a 1-meter-long crack in a certain local area, if the sensor is not at this location, this change cannot be detected. Therefore, it is necessary to install monitoring devices at a high density, otherwise it will be missed. The construction cost and maintenance cost of high-density monitoring are very high. In fact, due to the problem of construction investment, most of the geological disaster monitoring devices are not deployed at a high density, and there are even no basic technical prevention measures, resulting in "regional blind spots".

[0005] Rainfall is the main inducing factor of geological disasters. The monitored rainfall at the potential risk point is in the "past perfect tense". Geological disasters such as mudslides are likely to occur during rainfall. For example, the threshold value is set to 50mm rainfall in 24 hours, which is the cumulative rainfall in the past 24 hours, while mudslides are likely to occur when the concentrated rainfall is 30mm in a certain period of time. There is an underreporting, and casualties occur. There is no advance warning for risk warning. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a geological disaster monitoring and early warning method based on AI vision, which realizes more accurate geological disaster early warning and effectively improves the timeliness and accuracy of disaster prevention.

[0007] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0008] In the first aspect, a geological disaster monitoring and early warning method based on AI vision is provided, the method comprising:

[0009] Obtain pictures of geological disasters, including daytime and nighttime pictures;

[0010] Label the pictures to form a geological disaster picture dataset;

[0011] Train the AI ​​model through the image data set to obtain a trained AI model;

[0012] Lightweight processing of trained AI models is placed in edge computers;

[0013] In the edge computer, the captured images are imported into the AI ​​model to identify the characteristics of geological disasters, and the rolling comparison algorithm of the previous and next images is used to identify the characteristics of geological disasters, and the two results are mutually verified;

[0014] By verifying the two results, it is possible to determine whether a geological disaster has occurred or whether there are precursors to a geological disaster, and to issue an early warning.

[0015] Furthermore, the AI ​​model is trained by the image data set to obtain the trained AI model including:

[0016] Clean and preprocess the image data set to obtain a processed image data set;

[0017] Determine the AI ​​model and define the training parameters based on the characteristics of the processed image dataset;

[0018] Divide the processed image dataset into training set, validation set and test set;

[0019] Extract data from the training set in batches, input it into the AI ​​model framework, and perform forward propagation to calculate the predicted value;

[0020] Based on the predicted value and the true label, the loss function is calculated, and the gradient of the loss function to the AI ​​model parameters is calculated through back propagation;

[0021] Update AI model parameters according to the gradient. After each training iteration, use the validation set to evaluate AI model performance to obtain evaluation results.

[0022] Based on the evaluation results, determine the learning rate and AI model architecture to optimize model performance;

[0023] When the performance of the model on the validation set no longer improves, the training is terminated to obtain the trained AI model.

[0024] Furthermore, the calculation formula of the predicted value is:

[0025]

[0026] in, Represents the predicted value of the model; L represents the total number of layers of the neural network; W l represents the weight matrix of the lth layer; h l (X) represents the result of the activation function of the first layer; b l Represents the bias vector of the lth layer; l represents the layer index of the neural network.

[0027] Furthermore, the loss function is calculated as:

[0028]

[0029] Among them, L(θ) represents the loss function; m represents the number of samples; y i represents the true value of the i-th sample; λ represents the regularization parameter; i represents the index; ∥θ∥ 2 Represents the norm of the model parameters.

[0030] Furthermore, the calculation formula for updating the AI ​​model parameters is:

[0031]

[0032] Among them, θ t+1 represents the parameter vector of the AI ​​model at the t+1th iteration; θ t represents the parameter vector of the AI ​​model at the tth iteration; α represents the learning rate; represents the gradient of the model parameter θ at the tth iteration of the loss function L; t represents the number of iterations.

[0033] Furthermore, the captured images are imported into the AI ​​model on the edge computer to identify the characteristics of geological disasters, and the rolling comparison algorithm of the previous and next images is used to identify the characteristics of geological disasters, and the two results are mutually verified, including:

[0034] When new pictures are taken on site, they are transmitted to the edge computer in real time;

[0035] The edge computer determines whether the current time is day or night to determine the current time of the picture;

[0036] Based on the current time of the image, the edge computer determines the corresponding AI model and performs image analysis;

[0037] The AI ​​model performs forward propagation calculations on the image, identifies the geological hazard features in the image, and sets targets at the geological hazard feature risk points. When local cracks occur at the risk points, the targets move or deform.

[0038] The camera takes a picture of the target object and compares the pixels with the previous picture. If the pixels change, it means that the target object has moved or deformed.

[0039] When the target moves or deforms, the two results are verified against each other.

[0040] Furthermore, by verifying the two results, it is possible to determine whether a geological disaster or a precursor of a geological disaster has occurred and issue an early warning, including:

[0041] The two results verify each other. If the two methods reach the same conclusion, it is determined that a local geological disaster has occurred and a geological disaster warning is issued;

[0042] If the two results are inconsistent, take another local photo of the characteristic area and make another judgment;

[0043] If a local photo shows that the target position has moved or deformed, or if AI identifies local geological disaster characteristics, it will be judged as a precursor to a geological disaster and an early warning will be issued.

[0044] The second aspect is a geological disaster monitoring and early warning system based on AI vision, including:

[0045] An acquisition module is used to acquire pictures of geological disasters, including daytime pictures and nighttime pictures;

[0046] Label the images to form a geological disaster image dataset; train the AI ​​model through the image dataset to obtain a trained AI model; lightweight process the trained AI model and place it in the edge computer;

[0047] The processing module is used to import the captured images into the AI ​​model in the edge computer to identify the characteristics of geological disasters, and use the rolling comparison algorithm of the previous and next images to identify the characteristics of geological disasters, and verify the two results against each other; by verifying the two results against each other, it is determined that a geological disaster has occurred or a precursor of a geological disaster has occurred, and an early warning is issued.

[0048] According to a third aspect, a computing device includes:

[0049] one or more processors;

[0050] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0051] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0052] The above solution of the present invention includes at least the following beneficial effects:

[0053] By acquiring geological disaster images including daytime and nighttime, it ensures that the system can effectively monitor at any time period and improves the all-weather working ability of the early warning system. By annotating the images, a geological disaster image data set is formed, which not only enriches the learning materials of the AI ​​model, but also improves the model's recognition accuracy of different geological disaster characteristics. Through the training and application of the AI ​​model, the intelligent recognition of geological disaster characteristics is realized, the error of human judgment is reduced, and the accuracy and efficiency of early warning are improved. The trained AI model is lightweight and placed in the edge computer, so that the monitoring and early warning of geological disasters can be carried out in real time on site, reducing the delay of data transmission and improving the response speed. The mutual verification of the two methods of AI model recognition and the rolling comparison algorithm of the previous and next images increases the reliability of the early warning and reduces the possibility of false alarms and missed reports. Through the mutual verification of the two results, it is possible to more accurately judge the geological disasters that have occurred or the precursors of geological disasters, and issue early warnings in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a geological disaster monitoring and early warning method based on AI vision provided by an embodiment of the present invention.

[0055] Figure 2 It is a schematic diagram of a geological disaster monitoring and early warning system based on AI vision provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0057] like Figure 1 As shown, an embodiment of the present invention proposes a geological disaster monitoring and early warning method based on AI vision, and the method includes the following steps:

[0058] Step 1, obtaining pictures of geological disasters, including daytime pictures and nighttime pictures;

[0059] Step 2: annotate the images to form a geological disaster image dataset;

[0060] Step 3: Train the AI ​​model using the image data set to obtain a trained AI model;

[0061] Step 4: Lightweight processing of the trained AI model is placed in the edge computer;

[0062] Step 5: In the edge computer, the captured images are imported into the AI ​​model to identify the characteristics of geological disasters, and the rolling comparison algorithm of the previous and next images is used to identify the characteristics of geological disasters, and the two results are mutually verified;

[0063] Step 6, by verifying the two results, it is determined that a geological disaster or a precursor of a geological disaster has occurred, and an early warning is issued.

[0064] In an embodiment of the present invention, pictures of geological disasters are obtained, including daytime pictures and nighttime pictures, to ensure the comprehensiveness and diversity of the data. Pictures of relevant geological disasters can be collected regardless of whether it is a well-lit day or a dimly lit night, providing a rich data basis for the training of the AI ​​model and improving the generalization ability of the model. Through meticulous annotation work, a high-quality geological disaster picture dataset is constructed. It helps to improve the AI ​​model's ability to recognize geological disaster characteristics. Using a carefully constructed dataset to train the AI ​​model enables the model to learn the key features of geological disasters. The trained AI model will have higher recognition accuracy and stability. Through lightweight processing, the AI ​​model can run efficiently on edge computers with limited resources. This not only reduces hardware costs, but also improves the real-time and response speed of the geological disaster monitoring and early warning system.

[0065] The use of a double verification mechanism, that is, combining the recognition results of the AI ​​model and the rolling comparison algorithm of the previous and next pictures, can significantly improve the accuracy and reliability of geological disaster identification. This comprehensive verification method helps to reduce false alarms and missed alarms and improve the performance of the early warning system. By combining the two verification results, it is possible to more accurately judge the occurrence or precursor of geological disasters and issue early warnings in a timely manner. This helps relevant departments and personnel to quickly take countermeasures, reduce casualties and property losses caused by geological disasters, and improve society's disaster prevention and mitigation capabilities.

[0066] In a preferred embodiment of the present invention, the above step 1, obtaining pictures of geological disasters, including daytime pictures and nighttime pictures, may include:

[0067] By using high-definition cameras to shoot on-site, we can capture the actual scenes before and after the geological disaster. These pictures not only contain the intuitive expression of geological disasters, but also reflect the environmental background of the disaster.

[0068] Since a large number of pictures may be generated during the shooting process, including many pictures of low quality or irrelevant to geological disasters, preliminary screening is required. The purpose is to remove those pictures that are useless for subsequent analysis, reduce data redundancy, and improve work efficiency. Classification and storage of screened pictures is an important part of data management. Through reasonable classification, they can be retrieved and used according to different dimensions (such as disaster type, time, location, etc.).

[0069] In a preferred embodiment of the present invention, the above step 2, annotating the pictures to form a geological disaster picture dataset, may include:

[0070] During the labeling process, it is necessary to accurately identify the characteristics of geological disasters in the image and to annotate key information such as its type, location, scope and severity in detail. This information is an important basis for subsequent AI model training and directly affects the recognition accuracy and performance of the model. After the labeling is completed, data quality inspection is an essential step. Through inspection, labeling errors or omissions can be discovered and corrected in a timely manner to ensure the accuracy and completeness of the data set. The labeled data set is sorted and stored to form a standardized geological disaster image data set.

[0071] In a preferred embodiment of the present invention, the above step 3, training the AI ​​model through the image data set to obtain the trained AI model includes:

[0072] Step 31, cleaning and preprocessing the image data set to obtain a processed image data set;

[0073] Step 32, determine the AI ​​model and define the training parameters according to the characteristics of the processed image data set;

[0074] Step 33, dividing the processed image data set into a training set, a validation set and a test set;

[0075] Step 34, extracting data from the training set in batches, inputting it into the AI ​​model framework, and performing forward propagation to calculate the predicted value;

[0076] Step 35, calculate the loss function based on the predicted value and the true label, and calculate the gradient of the loss function to the AI ​​model parameters through back propagation;

[0077] Step 36, update the AI ​​model parameters according to the gradient, and after each training iteration, use the validation set to evaluate the AI ​​model performance to obtain the evaluation results;

[0078] Step 37, based on the evaluation results, determine the learning rate and AI model architecture to optimize model performance;

[0079] Step 38: When the performance of the model on the validation set no longer improves, the training is terminated to obtain a trained AI model.

[0080] In an embodiment of the present invention, before training the AI ​​model, the original image data set is cleaned and preprocessed. This includes removing duplicate, blurred or invalid images, normalizing and enhancing the data, etc., to improve the quality and diversity of the data. According to the characteristics of the preprocessed data set, such as the size of the image, the number of color channels, etc., a suitable AI model architecture, such as a convolutional neural network (CNN), is selected. At the same time, key parameters in the training process, such as learning rate, batch size, number of iterations, etc., are defined. The data set is divided into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to evaluate the performance of the model and adjust the model parameters during the training process, and the test set is used to finally evaluate the generalization ability of the model. During the training process, data is extracted from the training set in batches, input into the AI ​​model, and forward propagation calculation is performed to obtain the predicted value of the model. The loss function is calculated by comparing the predicted value of the model with the true label of the image. The loss function measures the gap between the model prediction and the true value. Next, the gradient of the loss function to the model parameters is calculated to determine how to adjust the model parameters to reduce the loss. The parameters of the model are updated using an optimization algorithm (such as the gradient descent method). After each training iteration, use the validation set to evaluate the performance of the model, such as accuracy, recall, etc. Based on the evaluation results on the validation set, dynamically adjust the learning rate and model architecture to optimize the performance of the model. This may include reducing the learning rate to obtain more refined parameter adjustments, or increasing the number of network layers to increase the complexity of the model. When the performance of the model on the validation set reaches a stable state or no longer improves significantly, end the training process and save the trained AI model.

[0081] Improve data quality and reduce the impact of noise and invalid data on model training, thereby enhancing the stability and generalization ability of the model. Ensure that the model structure matches the data characteristics to improve training efficiency and model performance. Through independent validation sets and test sets, the performance of the model can be evaluated more accurately to avoid overfitting.

[0082] Batch training and gradient updates: speed up the training process while ensuring that the model can gradually learn the inherent laws of the data. Dynamic adjustment of learning rate and model architecture: enable the model to continuously optimize itself during the training process and adapt to the changes and complexity of the data. Prevent the model from continuing to train when the performance of the validation set no longer improves, resulting in wasted resources and performance degradation.

[0083] In a preferred embodiment of the present invention, in the above step 34, the calculation formula of the predicted value is:

[0084]

[0085] in, Represents the predicted value of the model; L represents the total number of layers of the neural network; W l represents the weight matrix of the lth layer; hl (X) represents the result of the activation function of the first layer; b l Represents the bias vector of the lth layer; l represents the layer index of the neural network.

[0086] In the embodiment of the present invention, the original input data is sent to the first layer of the neural network. In each layer l, the input data (or the output of the previous layer) is first compared with the weight matrix W l Perform a dot multiplication operation. This process actually performs a weighted combination of the input features to extract higher-level features. Then, the weighted result is passed through the activation function h l (X) performs nonlinear transformation and introduces nonlinear factors, so that the neural network can learn and simulate more complex functional relationships. Each layer contains a bias vector b l , which provides an adjustable offset for the output of the neuron, increasing the flexibility of the model and enabling the model to better fit the data. As the data is passed layer by layer in the neural network, each layer extracts and transforms features based on the output of the previous layer. Finally, in the last layer of the neural network, the final prediction value is obtained through the activation function.

[0087] The weight matrix can effectively perform weighted combination of input features to extract information useful for the task. During the training process, the weight matrix will be adjusted according to the feedback of the loss function so that the model can better adapt to the data distribution and task requirements. The activation function introduces nonlinear factors to the neural network, allowing the model to learn and simulate more complex nonlinear relationships. Through nonlinear transformations, the activation function enhances the expressive power of the neural network, enabling it to handle more complex tasks. The bias vector provides an adjustable offset for the output of the neuron, increasing the flexibility of the model and helping the model to better fit the data. By adjusting the bias vector, the model can converge to the final solution faster during the training process.

[0088] In a preferred embodiment of the present invention, in the above step 35, the loss function is calculated as follows:

[0089]

[0090] Among them, L(θ) represents the loss function; m represents the number of samples; y i represents the true value of the i-th sample; λ represents the regularization parameter; i represents the index; ∥θ∥ 2 Represents the norm of the model parameters.

[0091] In an embodiment of the present invention, The mean squared error term calculates the model prediction value and the true value y iThe average of the sum of squares of the differences between λ and θ. It reflects the accuracy of the model prediction. The smaller the difference, the closer the model prediction is to the actual situation. 2 The regularization term is a penalty for the complexity of the model, where λ is the regularization parameter that controls the strength of the regularization, ∥θ∥ 2 The norm of the model parameters (such as weights and biases) is the square root of the sum of the squares of the elements of the parameter vector. The purpose of the regularization term is to prevent the model from overfitting and to limit the complexity of the model by introducing a penalty on the model parameters.

[0092] When the model prediction value is far from the true value, the mean square error term will increase, thereby increasing the value of the loss function. The regularization parameter λ determines the degree of influence of the regularization term. A larger λ value means stronger regularization, that is, a stronger penalty for model complexity, which helps prevent overfitting; while a smaller λ value weakens the impact of regularization, which may make the model more complex, but may also lead to overfitting. By shrinking the mean square error, the model's prediction value can be made closer to the true value, thereby improving the accuracy of the prediction.

[0093] The mean square error provides a clear optimization target, so that gradient descent or other optimization algorithms in the training process can adjust model parameters more effectively. By introducing regularization terms, the overfitting of the model on the training data can be effectively prevented. Regularization can prompt the model to choose a simpler parameter configuration, thereby enhancing the model's ability to generalize to new data. By adjusting the regularization parameter λ, the complexity of the model can be flexibly controlled to adapt to different problems and data set characteristics.

[0094] In a preferred embodiment of the present invention, the calculation formula for updating the AI ​​model parameters in step 36 is:

[0095]

[0096] Among them, θ t+1 represents the parameter vector of the AI ​​model at the t+1th iteration; θ t represents the parameter vector of the AI ​​model at the tth iteration; α represents the learning rate; represents the gradient of the model parameter θ at the tth iteration of the loss function L; t represents the number of iterations.

[0097] In an embodiment of the present invention, It means that after each iteration, the model's parameter vector θ is adjusted according to the current gradient Update. α is the learning rate, which controls the step size of parameter update. A larger learning rate may lead to unstable training, while a smaller learning rate may lead to slow training. represents the gradient of the loss function with respect to the model parameters. The gradient indicates how the model parameters should be adjusted in order to minimize the loss function. t represents the number of iterations. As the number of iterations increases, the model parameters will gradually adjust to better values, thereby reducing the value of the loss function.

[0098] By calculating the gradient of the loss function with respect to the model parameters and updating the parameters in the opposite direction of the gradient, the value of the loss function can be effectively reduced, thereby optimizing the performance of the model. The setting of the learning rate α allows for flexible control of the amplitude of parameter updates during training. By properly adjusting the learning rate, the training speed can be accelerated while ensuring training stability. By updating the model parameters through multiple iterations, the point of the final value of the loss function can be gradually approached, thereby improving the prediction accuracy and performance of the model.

[0099] In a preferred embodiment of the present invention, the above step 4, placing the trained AI model lightweight processing into the edge computer, may include:

[0100] Model compression is to reduce the size of the model by reducing redundant parameters of the model, merging similar computing layers, or using quantization technology. In this process, methods such as weight reuse, pruning, and coding optimization are used to remove unnecessary parts of the model, thereby reducing storage space and computing requirements. Model optimization mainly makes further adjustments to the compressed model to improve its operating efficiency on edge devices. This includes optimizing the model's calculation graph, reducing unnecessary computing operations, or performing specific hardware optimization on the model to make it more suitable for execution on edge computers. Since specific deep learning frameworks (such as TensorFlow) are used when training the model, these models need to be converted into a format that can be recognized and executed by edge devices before being deployed to edge computers. The conversion process involves serialization and deserialization of the model, as well as adaptation for specific hardware platforms. The converted model file is transferred to the edge computer and configured to ensure that the model can run smoothly on the device. This includes installing the necessary operating environment, configuring related parameters, and conducting preliminary model testing. Performance testing of the deployed model is performed on the edge computer to ensure that it meets the requirements of real-time and accuracy. This process includes testing key indicators such as the model's inference speed, memory usage, and recognition accuracy.

[0101] In a preferred embodiment of the present invention, the above step 5, in the edge computer, uses the captured pictures to import into the AI ​​model to identify the geological disaster characteristics, and uses the rolling comparison algorithm of the previous and next pictures to identify the geological disaster characteristics, and mutually verifies the two results, which may include:

[0102] Step 51, when a new picture is taken on site, the picture is transmitted to the edge computer in real time;

[0103] Step 52, the edge computer determines whether the current time is day or night to determine the current time of the picture;

[0104] Step 53, according to the current time of the picture, the edge computer determines the corresponding AI model and performs picture analysis;

[0105] Step 54, the AI ​​model performs forward propagation calculation on the image, identifies the geological disaster features in the image, and sets targets at the geological disaster feature risk points. When local cracks occur at the risk points, the targets move or deform.

[0106] Step 55, the camera takes a picture of the target object and compares the picture taken last time with the pixel comparison. If a pixel change is found, it means that the target object has moved or deformed.

[0107] Step 56, when the target object moves or deforms, the two results are mutually verified.

[0108] In an embodiment of the present invention, after the on-site camera captures a new picture, the picture data is transmitted to the edge computer in real time by wired or wireless means to ensure the timeliness and accuracy of the data. The edge computer determines the current time through an internal clock or an external time synchronization service, and judges whether it is day or night, so as to select a suitable AI model for analysis. The edge computer selects the pre-trained AI model of the corresponding time period for picture analysis according to whether the current time is day or night. The AI ​​model analyzes the picture through forward propagation calculation and identifies the characteristics of geological disasters. Once potential geological disaster risk points are identified, the system will set virtual or actual targets at these points as references. The camera regularly takes pictures of the target object and compares it with the previous picture at the pixel level to detect whether the target object has moved or deformed. When the target object is detected to have moved or deformed through pixel comparison, this result will be mutually verified with the geological disaster characteristics identified by the AI ​​model.

[0109] Since geological disasters behave differently under different lighting conditions, selecting an appropriate AI model based on time can improve the accuracy of recognition. Using a time-dependent AI model can improve the recognition accuracy of geological disaster features because the model is trained under similar lighting conditions. The setting of targets helps to continuously monitor changes in potential hazards and promptly detect precursors of geological disasters. Pixel comparison technology can accurately capture subtle changes in targets, providing sensitive and reliable data for early warning of geological disasters. The double verification mechanism improves the reliability and accuracy of the early warning system and reduces the risk of false alarms and missed alarms. Real-time transmission ensures that edge computers can process the latest image data in a timely manner, providing a data basis for real-time monitoring and early warning of geological disasters.

[0110] In a preferred embodiment of the present invention, the above step 6, through mutual verification of the two results, determines that a geological disaster or a precursor of a geological disaster has occurred, and issues an early warning, may include:

[0111] Step 61, the two results are mutually verified. If the two methods reach the same conclusion, it is determined that a local geological disaster has occurred, and a geological disaster warning is issued;

[0112] Step 62, if the two results are inconsistent, take a local photo of the characteristic area again and make a judgment again;

[0113] Step 63: If the local photo finds that the target position has moved or deformed, or AI identifies local geological disaster characteristics, it is judged as a precursor to a geological disaster and an early warning is issued.

[0114] In an embodiment of the present invention, the system compares the geological disaster characteristics identified by the AI ​​model with the target object movement or deformation results detected by pixel comparison. If both methods indicate that a geological disaster has occurred, the system will confirm this conclusion. When the AI ​​model recognition result is inconsistent with the pixel comparison result, the system will instruct the camera to take local high-definition photos of the characteristic area again for more detailed analysis and judgment. In the re-taken local photos, if the system detects that the position of the target object has moved or deformed, or the AI ​​model identifies obvious geological disaster characteristics, the system will determine that this is a precursor to a geological disaster and immediately issue an early warning.

[0115] Double verification improves the accuracy of early warnings and reduces the possibility of false alarms, thus ensuring that early warnings can be issued in a timely and accurate manner when a geological disaster actually occurs. By reshooting and analyzing local photos, the system can obtain more detailed information, which helps to resolve uncertainties in initial judgments and improve the reliability and accuracy of the early warning system. It ensures that even if there is uncertainty in the initial judgment, the system can still capture the precursors of geological disasters in a timely manner through further analysis, so that preventive measures can be taken in advance to reduce potential losses.

[0116] like Figure 2 As shown, an embodiment of the present invention further provides a geological disaster monitoring and early warning system 20 based on AI vision, comprising:

[0117] The acquisition module 21 is used to acquire pictures of geological disasters, including daytime pictures and nighttime pictures; annotate the pictures to form a geological disaster picture data set; train the AI ​​model through the picture data set to obtain a trained AI model; and lightweight process the trained AI model and place it in the edge computer;

[0118] The processing module 22 is used to import the captured pictures into the AI ​​model in the edge computer to identify the characteristics of geological disasters, and use the rolling comparison algorithm of the previous and next pictures to identify the characteristics of geological disasters, and verify the two results with each other; through the mutual verification of the two results, it is determined that a geological disaster has occurred or a precursor of a geological disaster has occurred, and an early warning is issued.

[0119] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0120] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0121] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0122] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0124] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0125] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0126] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0127] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, 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. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or 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: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0128] In addition, it should be noted that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understood that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using basic programming skills after reading the description of the present invention.

[0129] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0130] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A geological disaster monitoring and early warning method based on AI vision, characterized in that: The method comprises: Obtain pictures of geological disasters, including daytime and nighttime pictures; Label the pictures to form a geological disaster picture dataset; Train the AI ​​model through the image data set to obtain a trained AI model; Lightweight processing of trained AI models is placed in edge computers; In the edge computer, the captured images are imported into the AI ​​model to identify the characteristics of geological disasters, and the rolling comparison algorithm of the previous and next images is used to identify the characteristics of geological disasters, and the two results are mutually verified; By verifying the two results, it is possible to determine whether a geological disaster has occurred or whether there are precursors to a geological disaster, and to issue an early warning.

2. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 1 is characterized in that: The AI ​​model is trained through the image dataset to obtain the following AI models: Clean and preprocess the image data set to obtain a processed image data set; Determine the AI ​​model and define the training parameters based on the characteristics of the processed image dataset; Divide the processed image dataset into training set, validation set and test set; Extract data from the training set in batches, input it into the AI ​​model framework, and perform forward propagation to calculate the predicted value; Based on the predicted value and the true label, the loss function is calculated, and the gradient of the loss function to the AI ​​model parameters is calculated through back propagation; Update AI model parameters according to the gradient. After each training iteration, use the validation set to evaluate AI model performance to obtain evaluation results. Based on the evaluation results, determine the learning rate and AI model architecture to optimize model performance; When the performance of the model on the validation set no longer improves, the training is terminated to obtain the trained AI model.

3. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 2 is characterized in that: The formula for calculating the predicted value is: in, Represents the predicted value of the model; L represents the total number of layers of the neural network; W l represents the weight matrix of the lth layer; h l (X) represents the result of the activation function of the first layer; b l Represents the bias vector of the lth layer; l represents the layer index of the neural network.

4. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 3 is characterized in that: The loss function is calculated as: Among them, L(θ) represents the loss function; m represents the number of samples; y i represents the true value of the i-th sample; λ represents the regularization parameter; i represents the index; ||θ|| 2 Represents the norm of the model parameters.

5. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 4 is characterized in that: The calculation formula for updating AI model parameters is: Among them, θ t+1 represents the parameter vector of the AI ​​model at the t+1th iteration; θ t represents the parameter vector of the AI ​​model at the tth iteration; α represents the learning rate; represents the gradient of the model parameter θ at the tth iteration of the loss function L; t represents the number of iterations.

6. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 5 is characterized in that: In the edge computer, the captured images are imported into the AI ​​model to identify the characteristics of geological disasters, and the rolling comparison algorithm of the previous and next images is used to identify the characteristics of geological disasters. The two results are mutually verified, including: When new pictures are taken on site, they are transmitted to the edge computer in real time; The edge computer determines whether the current time is day or night to determine the current time of the picture; Based on the current time of the image, the edge computer determines the corresponding AI model and performs image analysis; The AI ​​model performs forward propagation calculations on the image, identifies the geological hazard features in the image, and sets targets at the geological hazard feature risk points. When local cracks occur at the risk points, the targets move or deform. The camera takes a picture of the target object and compares the pixels with the previous picture. If the pixels change, it means that the target object has moved or deformed. When the target moves or deforms, the two results are verified against each other.

7. The AI ​​vision-based geological disaster monitoring and early warning method according to claim 6 is characterized in that: Through mutual verification of the two results, it is judged that a geological disaster has occurred or a precursor of a geological disaster has occurred, and an early warning is issued, including: The two results verify each other. If the two methods reach the same conclusion, it is determined that a local geological disaster has occurred and a geological disaster warning is issued; If the two results are inconsistent, take another local photo of the characteristic area and make another judgment; If a local photo shows that the target position has moved or deformed, or if AI identifies local geological disaster characteristics, it will be judged as a precursor to a geological disaster and an early warning will be issued.

8. A geological disaster monitoring and early warning system based on AI vision, characterized in that: include: An acquisition module is used to acquire pictures of geological disasters, including daytime pictures and nighttime pictures; Label the pictures to form a geological disaster picture dataset; Train the AI ​​model through the image data set to obtain the trained AI model; lightweight process the trained AI model and place it in the edge computer; The processing module is used to import the captured images into the AI ​​model in the edge computer to identify the characteristics of geological disasters, and use the rolling comparison algorithm of the previous and next images to identify the characteristics of geological disasters, and verify the two results against each other; by verifying the two results against each other, it is determined that a geological disaster has occurred or a precursor of a geological disaster has occurred, and an early warning is issued.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images