AI identification and labeling method and system for geological disaster samples based on expert scoring system

Abstract

The present invention discloses an AI identification and labeling method and system for geological disaster samples based on an expert scoring system, comprising the following steps: S1, obtaining geological disaster sample data, assigning weights to the geological disaster sample data, obtaining initial weight samples, and training the initial weight samples to obtain a recognition model; S2, based on the recognition model, identifying and labeling the labeled geological disaster sample data and the unlabeled geological disaster sample data in the geological disaster sample data; S3, evaluating the identified and labeled geological disaster sample data based on the expert scoring system, assigning weights to the evaluation results to obtain weight samples, selecting qualified weight samples and importing them into a sample library; S4, retraining the recognition model based on the weight samples, and re-identifying and labeling unqualified weight samples; S5, returning to S3 until all the geological disaster sample data are imported into the sample library to complete the identification and labeling.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic identification and annotation of geological disaster remote sensing image samples, and in particular relates to an AI identification and annotation method and system for geological disaster samples based on an expert scoring system. Background Art

[0002] Geological disasters refer to disastrous geological events caused by natural geological processes and human activities during the development and evolution of the Earth. They are characterized by suddenness, multiple occurrences, clustered occurrences, and chain reactions. China has a vast territory and diverse terrain, and geological disasters occur frequently, posing a serious threat to people's lives and property.

[0003] Since the 1960s, remote sensing technology has developed rapidly. The platform has evolved from satellites to drones, and the band covers the ultraviolet-visible light-infrared-microwave range. The spatial resolution has increased from nearly 100 meters to 0.5 meters, and the revisit period has been shortened from dozens of days to 1 day. Satellite images and terrain products are extremely rich, providing important technical support for the identification and detection of geological disasters. There are four main methods for identifying historical geological disasters: (1) Visual interpretation method. This method mainly uses the knowledge and experience of professionals to manually judge historical geological disasters. The visual interpretation results are generally accurate, but they are very dependent on the experience of professionals and are very time-consuming and labor-intensive. (2) Feature threshold method. This method often uses pixel-based methods or object-based multi-scale segmentation methods to statistically analyze the spectrum, texture, landform or terrain characteristics of the geological disaster area, and set one or more thresholds for identification. Since its judgment standard is the statistical value of specific features of a specific area, its scope of application is relatively small. (3) Change detection method. This method requires two or more periods of remote sensing data at the same location to find the changed areas caused by geological disasters. The data source can be two-dimensional optical images or three-dimensional terrain data. Change detection has a better application effect on new geological disasters and requires remote sensing data with time series. (4) Machine learning method. This method is similar to the feature threshold method and change detection in the data preparation part. It needs to extract various relevant features of the data used, but the final identification is not distinguished by a single threshold. Instead, it uses various classifiers of machine learning to identify geological disasters. Its degree of automation is higher than the previous method. Commonly used methods include Bayesian, logistic regression, support vector machine, random forest and deep learning. As a data-driven method, machine learning has become a hot topic in the research of geological disaster identification methods with the explosive growth of geological disaster observation data. Machine learning algorithms are quite different from traditional algorithms. The most prominent point is that machine learning algorithms require a large amount of sample data to train a high-precision model.

[0004] In existing technologies, the collection and labeling of geological disaster samples requires manual or semi-manual work. This is a relatively simple task, but the workload is enormous. This has led to the emergence of data / sample annotators, demonstrating the immense workload of data / sample labeling. To address the significant human and material resources required for manual or semi-manual labeling in AI-powered identification and labeling algorithms for geological disaster samples, this paper proposes a method and system for AI-powered identification and labeling of geological disaster samples based on an expert scoring system. Summary of the Invention

[0005] The purpose of this invention is to propose an AI identification and labeling method and system for geological disaster samples based on an expert scoring system, thereby improving the accuracy and efficiency of labeling geological disaster sample data.

[0006] On one hand, to achieve the above-mentioned purpose, the present invention provides an AI identification and labeling method for geological disaster samples based on an expert scoring system, comprising the following steps:

[0007] S1. Obtain geological disaster sample data, assign weights to the geological disaster sample data, obtain initial weight samples, and train the initial weight samples to obtain a recognition model; the geological disaster samples refer to geological disaster remote sensing image samples;

[0008] S2. Based on the recognition model, identify and label the labeled geological disaster sample data and the unlabeled geological disaster sample data in the geological disaster sample data;

[0009] S3. Evaluate the identified and labeled geological disaster sample data based on an expert scoring system, assign weights to the evaluation results to obtain weighted samples, and select qualified weighted samples to import into a sample library;

[0010] S4. Retraining the recognition model based on the weight samples, and re-identifying and labeling unqualified weight samples;

[0011] S5. Return to S3 until all the geological disaster sample data are imported into the sample library and identification and labeling are completed.

[0012] Optionally, the qualified weighted sample includes: sample data with a score of not less than 90 points in the weighted sample.

[0013] Optionally, obtaining the initial weighted sample specifically includes: extracting geological disaster sample data according to a sample database, assigning initial weights to the geological disaster sample data, and assigning initial weighted samples according to the scoring scores.

[0014] Optionally, obtaining the recognition model specifically includes:

[0015] A geological disaster identification model is preset, and the initial weight samples are used as a training set to train the parameters in the geological disaster identification model to obtain a recognition model.

[0016] Optionally, the identified and annotated sample data is scored based on an expert scoring system, and the calculation is as follows:

[0017]

[0018] Among them, P i is the scoring value, A represents the preset geological disaster area marked manually or the preset geological disaster area marked by the recognition model in the previous cycle; B represents the preset geological disaster area marked by the recognition model in this cycle; S A∩B represents the area of ​​the intersection of area A and area B; S A∪B It represents the area of ​​the union of region A and region B.

[0019] On the other hand, to achieve the above-mentioned purpose, the present invention provides an AI identification and labeling system for geological disaster samples based on an expert scoring system, comprising:

[0020] A database module, a recognition model module and an expert system module, wherein the database module, the recognition model module and the expert system module are communicatively connected;

[0021] The database module is used to store geological disaster sample data;

[0022] The recognition model module is used to train the recognition model using machine learning based on the weight samples;

[0023] The expert system module is used to score the sample results identified and labeled by the recognition model.

[0024] Optionally, the database module includes a data disaster type judgment unit, a data storage unit, a data acquisition unit and a data preprocessing unit.

[0025] The data geological disaster type judgment unit is used to judge the type of geological disaster of the sample by using machine learning;

[0026] The data storage unit is used to store new geological disaster sample data and samples with qualified scores scored by the expert system;

[0027] The data acquisition unit is used to acquire sample data of a preset geological disaster type from a sample library;

[0028] The data preprocessing unit is used to preprocess the sample data obtained from the sample library.

[0029] Optionally, the recognition model module includes a recognition model training unit and a sample recognition unit.

[0030] The recognition model training unit is used to train the recognition model;

[0031] The sample identification unit is used to identify and mark the new geological disaster sample data and samples with unqualified scores assigned by the expert system using the identification model.

[0032] Technical effect of the present invention: The present invention discloses an AI identification and labeling method and system for geological disaster samples based on an expert scoring system, which solves the problems of manual labeling of geological disaster sample data in related technologies, low labeling efficiency, inaccurate labeling results, and consumption of a large amount of manpower and material resources; AI automatic identification (machine learning) is used to improve work efficiency and save human labor; but the accuracy of machine learning is largely determined by the number and quality of samples; the more samples there are, the stronger the representativeness, and the more accurate the labeling, the more reliable the obtained recognition model; but with traditional manual labeling methods, the more samples there are and the higher the quality, the more labor is required; the present invention improves the accuracy and efficiency of labeling of geological disaster sample data, reduces labor intensity, and has good economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0034] Figure 1 This is a flowchart of a method for AI identification and labeling of geological disaster samples based on an expert scoring system according to an embodiment of the present invention;

[0035] Figure 2 A flow chart for training a geological disaster identification model is preset for an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of the structure of the AI ​​identification and labeling system for geological disaster samples in the expert scoring system provided by an embodiment of the present invention;

[0037] Figure 4 A flowchart of the AI ​​identification and labeling method and system for landslide disaster samples in the expert scoring system provided in an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the scoring results of the expert system for landslide sample identification and annotation in an embodiment of the present application. DETAILED DESCRIPTION

[0039] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0041] The AI ​​identification and labeling technology and system for geological disaster samples in the expert scoring system provided in the embodiments of the present application can be specifically applied to smart terminals such as desktop computers, notebooks, mainframe computers, and tablet computers.

[0042] Example 1:

[0043] like Figure 1 As shown, this embodiment provides an AI identification and labeling method for geological disaster samples based on an expert scoring system, including the following steps:

[0044] Step 1: Obtain geological disaster sample data, assign weights to the geological disaster sample data, obtain initial weight samples, and train the initial weight samples to obtain a geological disaster identification model;

[0045] Step 2: Based on the recognition model, identify and label the labeled geological disaster sample data and the unlabeled geological disaster sample data in the geological disaster sample data;

[0046] Step 3: Evaluate the identified and labeled geological disaster sample data based on the expert scoring system, assign weights to the evaluation results to obtain weighted samples, and select qualified weighted samples to import into the sample library;

[0047] Step 4: Retrain the recognition model based on the weight samples, and re-identify and label unqualified weight samples;

[0048] Step 5: Return to step 3 until all the geological disaster sample data are imported into the sample library and the identification and labeling are completed.

[0049] As an implementation manner of the present invention, the qualified weighted samples include: sample data with a score of not less than 90 points in the weighted samples.

[0050] As an embodiment of the present invention, the recognition model is obtained by machine learning using weight samples, and further Figure 2 As shown:

[0051] Obtain weighted samples: The weighted samples include the preset geological disaster type samples extracted from the sample database and assigned initial weights and weighted samples assigned weights according to the scoring scores. The sample data set F1 = {S1, S2, S3, ..., S p}, initialize the weight or assign the weight to W1={C1,C2,C3,...,C p}, after assigning weights, the sample data set FF1 = {C1*S1,, C2*S2, C3*S3,..., C p *S p};

[0052] Training the recognition model: Select an appropriate preset geological disaster recognition model, use the obtained weighted sample FF1 as the training set, and fine-tune the recognition model parameters so that samples with larger weights have a greater impact on the recognition model;

[0053] Prevent overfitting in training: Use L1 regularization, L2 regularization, Dropout regularization and other methods to prevent overfitting in training.

[0054] As an embodiment of the present invention, the expert system scoring method, for manually labeled samples, uses the sample result area labeled by the recognition model and the manually labeled preset disaster area to perform correlation analysis; for unlabeled new samples, uses the sample result area labeled by the recognition model and the sample labeling results in the previous cycle to perform correlation analysis. The higher the correlation, the higher the score given by the expert system to the sample result labeled by the recognition model; the scoring value P is obtained according to the following formula i :

[0055]

[0056] Among them, A represents the preset geological disaster area marked manually or the preset geological disaster area marked by the recognition model in the previous cycle; B represents the preset geological disaster area marked by the recognition model in this cycle; S A∩B represents the area of ​​the intersection of area A and area B; S A∪B represents the area of ​​the union of region A and region B; S A∩B When P is 0, i =0, indicating that area A and area B are separated and there is no intersection area. The recognition and labeling effect of this sample is poor and needs to be re-recognized; The closer it is to 1, the larger the overlapping area of ​​the two regions is, and the better the recognition effect of the sample is.

[0057] Example 2:

[0058] like Figure 3 As shown, an embodiment of the present invention further provides an AI identification and labeling system for geological disaster samples based on an expert scoring system, comprising:

[0059] Database module, used to store geological disaster sample data;

[0060] The recognition model module is used to train the recognition model using machine learning methods based on weight samples;

[0061] An expert system module, used to score the sample results identified and labeled by the recognition model;

[0062] As an embodiment of the present invention, the database module is a spatial database; the data table structure of the spatial database is designed to include: image ID, image geographic location, number of image bands, image resolution, imaging time, geological disaster type, annotation results, expert system score, etc.

[0063] As an embodiment of the present invention, the database module includes a data disaster type judgment unit, a data storage unit, a data acquisition unit and a data preprocessing unit.

[0064] The data geological disaster type judgment unit is used to judge the sample geological disaster type using a machine learning model;

[0065] The data storage unit is used to store new geological disaster sample data and samples with qualified scores scored by the expert system;

[0066] The data acquisition unit is used to acquire sample data of a preset geological disaster type from a sample library;

[0067] The data preprocessing unit is used to perform preprocessing such as resampling on the sample data obtained from the sample library so that the sample data meets the requirements of the recognition model module;

[0068] As an embodiment of the present invention, the recognition model module includes a recognition model training unit and a sample recognition unit.

[0069] The recognition model training unit is used to train the recognition model;

[0070] The sample identification unit is used to identify and mark the new geological disaster sample data and samples with unqualified scores assigned by the expert system using the identification model.

[0071] Example 3

[0072] like Figure 4 As shown, the embodiment of the present invention also provides an AI identification and labeling method and system for landslide disaster samples in an expert scoring system, including:

[0073] Step 1: Obtain sample data of landslide type, assign weights to the sample data, obtain initial weight samples, and train the initial weight samples to obtain a landslide recognition model;

[0074] Step 2: Based on the landslide identification model, identifying and labeling the labeled landslide sample data and the unlabeled landslide sample data in the landslide sample data;

[0075] Step 3: Evaluate the identified and marked landslide sample data based on the expert scoring system, assign weights to the evaluation results to obtain weighted samples, and select qualified weighted samples to import into the sample library;

[0076] Step 4: retraining the landslide identification model based on the weighted samples, and re-identifying and marking unqualified weighted samples;

[0077] Step 5: Return to step 3 until all the landslide sample data are imported into the sample library and the identification and labeling are completed.

[0078] As an implementation manner of the present invention, the qualified weighted samples include: sample data with a score of not less than 90 points in the weighted samples.

[0079] As an embodiment of the present invention, when identifying and labeling the unlabeled landslide samples in step 2, machine learning must first be used to determine whether the sample geological disaster type is a landslide. The machine learning model can use a convolutional neural network.

[0080] As an embodiment of the present invention, the landslide identification model is obtained by machine learning using weighted samples, and the further steps are as follows:

[0081] Obtain weighted samples: The weighted samples include landslide disaster type samples extracted from the sample database and assigned initial weights and weighted samples assigned weights according to the scoring scores. The sample data set F1 = {S1, S2, S3, ..., S p}, initialize the weight or assign the weight to W1={C1,C2,C3,...,C p}, after assigning weights, the sample data set FF1 = {C1*S1,, C2*S2, C3*S3,..., C p *S p};

[0082] Training the landslide recognition model: Using the obtained weighted sample FF1 as the training set, fine-tune the parameters of the landslide recognition model so that samples with larger weights have a greater impact on the recognition model. The landslide recognition model can use a mask region-based convolutional neural network (Mask R-CNN). First, feature pyramid extraction is performed on the image. The network is extracted in the training area and a region of interest is generated. Then, category prediction and boundary regression are performed on the region of interest. The activation function uses the ReLu function: f(x) = max(0, x).

[0083] Prevent overfitting in training: Use the Dropout regularization method to prevent overfitting in training.

[0084] As an embodiment of the present invention, the expert system scoring method, for manually labeled samples, uses the sample landslide area labeled by the recognition model and the manually labeled preset landslide area to perform correlation analysis; for unlabeled new samples, uses the sample landslide area labeled by the recognition model and the sample labeling results in the previous cycle to perform correlation analysis. The higher the correlation, the higher the score given by the expert system to the sample result labeled by the recognition model; the scoring value P is obtained according to the following formula i :

[0085]

[0086] Among them, A represents the landslide area marked manually or the landslide area marked by the recognition model in the previous cycle; B represents the preset geological disaster area marked by the recognition model in this cycle; S A∩B represents the area of ​​the intersection of area A and area B; S A∪B represents the area of ​​the union of region A and region B; Figure 5 An example of scoring the expert system's identification and annotation results for landslide samples.

[0087] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An AI identification and labeling method for geological disaster samples based on an expert scoring system, characterized by: The following steps are involved: S1. Obtain geological disaster sample data, assign weights to the geological disaster sample data, obtain initial weight samples, and train the initial weight samples to obtain a recognition model; the geological disaster samples refer to geological disaster remote sensing image samples; Training the recognition model includes: using the obtained weight sample FF1 as a training set and fine-tuning the recognition model parameters; the recognition model uses a mask region convolutional neural network to extract feature pyramids from the image, extracts the network in the training area and generates a region of interest, and performs category prediction and boundary regression on the region of interest; the activation function uses the ReLu function: ; S2. Based on the recognition model, identify and label the labeled geological disaster sample data and the unlabeled geological disaster sample data in the geological disaster sample data; S3. Evaluate the identified and labeled geological disaster sample data based on an expert scoring system, assign weights to the evaluation results to obtain weighted samples, and select qualified weighted samples to import into a sample library; The expert scoring system includes: for manually labeled samples, performing a correlation analysis on the sample landslide area labeled by the recognition model and the manually labeled preset landslide area; for unlabeled new samples, performing a correlation analysis on the sample landslide area labeled by the recognition model and the sample labeling results in the previous cycle; the higher the correlation, the higher the score given by the expert scoring system to the sample result labeled by the recognition model; Get rating points : *100%; Among them, A represents the manually marked landslide area or the landslide area marked by the recognition model in the previous cycle; B represents the preset geological disaster area marked by the recognition model in this cycle; represents the area of ​​the intersection of region A and region B; represents the area of ​​the union of region A and region B; S4. Retraining the recognition model based on the weight samples, and re-identifying and labeling unqualified weight samples; S5. Return to S3 until all the geological disaster sample data are imported into the sample library and identification and labeling are completed.

2. The method for identifying and labeling geological disaster samples using AI based on an expert scoring system according to claim 1, wherein: The qualified weighted samples include: sample data with a score of not less than 90 points in the weighted samples.

3. The AI ​​identification and labeling method for geological disaster samples based on an expert scoring system according to claim 1, characterized in that: Acquiring the initial weighted samples specifically includes: extracting geological disaster sample data according to the sample database, assigning initial weights to the geological disaster sample data, and assigning initial weighted samples according to the scoring scores.

4. The AI ​​identification and labeling method for geological disaster samples based on an expert scoring system according to claim 1, characterized in that: Obtaining the recognition model specifically includes: A geological disaster identification model is preset, and the initial weight samples are used as a training set to train the parameters in the geological disaster identification model to obtain a recognition model.

5. The AI ​​identification and labeling system for geological disaster samples based on the expert scoring system is characterized by: A method for AI identification and labeling of geological disaster samples based on an expert scoring system according to any one of claims 1 to 4, the system comprising: A database module, a recognition model module and an expert system module, wherein the database module, the recognition model module and the expert system module are communicatively connected; The database module is used to store geological disaster sample data; The recognition model module is used to train the recognition model using machine learning based on the weight samples; The expert system module is used to score the sample results identified and labeled by the recognition model.

6. The AI ​​identification and labeling system for geological disaster samples based on an expert scoring system according to claim 5, characterized in that: The database module includes a data disaster type judgment unit, a data storage unit, a data acquisition unit and a data pre-processing unit. The data geological disaster type judgment unit is used to judge the type of geological disaster of the sample by using machine learning; The data storage unit is used to store new geological disaster sample data and samples with qualified scores scored by the expert system; The data acquisition unit is used to acquire sample data of a preset geological disaster type from a sample library; The data preprocessing unit is used to preprocess the sample data obtained from the sample library.

7. The AI ​​identification and labeling system for geological disaster samples based on an expert scoring system according to claim 6, characterized in that: The recognition model module includes a recognition model training unit and a sample recognition unit. The recognition model training unit is used to train the recognition model; The sample identification unit is used to identify and mark the new geological disaster sample data and samples with unqualified scores assigned by the expert system using the identification model.

Images