Identification method for elevator steel belt surface damage
Through multimodal data acquisition and processing technology, combined with decision tree and knowledge graph analysis, real-time and accurate identification of surface damage of elevator steel belts is achieved, solving the problems of low detection accuracy and high cost in the existing technology, and improving the safety and reliability of elevators.
Patent Information
- Application Number
- CN202510218216.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-26
- Publication Date
- 2025-06-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing elevator steel belt damage detection methods have problems such as expensive equipment, complex operation and low detection accuracy, making it difficult to achieve real-time and accurate damage recognition.
Multimodal data acquisition technology is used to collect video, image, audio and spectral data of the surface of elevator steel belt in real time, and the identification of surface damage of elevator steel belt is achieved through preprocessing, feature extraction, decision tree construction and knowledge graph analysis.
Real-time and accurate identification of surface damage of elevator steel belts is achieved, improving the safety and reliability of elevators and reducing inspection costs.
Smart Images
Figure CN120191818A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of elevator steel belt damage detection, and relates to a method for identifying surface damage of an elevator steel belt. Background Art
[0002] As an important tool for vertical transportation, the safety of an elevator is of crucial importance. The elevator steel belt is a key component in the elevator system, and the detection and identification of its surface damage are of great significance for ensuring the safe operation of the elevator. Traditional methods for detecting elevator steel belt damage mainly rely on regular manual inspections, which have problems such as low detection efficiency, large errors, and inability to monitor in real time. With the development of technology, non-destructive testing techniques such as electromagnetic testing, life prediction value method, and resistance testing method have been introduced into elevator steel belt damage detection, but these methods have problems such as high device prices, complex operations, and low detection accuracy. Summary of the Invention
[0003] In order to overcome the deficiencies of the existing elevator steel belt damage detection methods, such as high device prices, complex operations, and low detection accuracy, the present invention provides a method for identifying surface damage of an elevator steel belt, which can identify the surface damage of the elevator steel belt in real time and accurately, and improve the safety and reliability of the elevator.
[0004] The technical solution adopted by the present invention to solve its technical problems is:
[0005] A method for identifying surface damage of an elevator steel belt, comprising the following steps:
[0006] Step 1, obtaining multi-modal data on the surface of the elevator steel belt;
[0007] Step 2, preprocessing the multi-modal data;
[0008] Step 3, extracting damage feature information on the surface of the elevator steel belt;
[0009] Step 4, analyzing the causes of surface damage of the elevator steel belt;
[0010] Step 5, constructing a knowledge graph for identifying surface damage of the elevator steel belt;
[0011] Step 6, comprehensively making a decision based on the output result of the decision tree and the comparison result of the knowledge graph to generate a final damage identification result.
[0012] Further, in the above step 1, the multi-modal data includes video, image, audio, and spectral data.
[0013] In the above step 2, the preprocessing steps include image contrast enhancement, image smoothing, image segmentation and extraction, image cleaning, and data enhancement processing.
[0014] In step 3, edge detection, texture analysis, and image processing techniques are used to extract edge features, texture features, spectral features, and audio features.
[0015] In step 4, the analysis of the damage causes includes establishing a damage category library and using machine learning algorithms to classify and identify damage events.
[0016] In step 5, a knowledge graph for identifying surface damage of elevator steel belts is constructed using graph database technology. The process is as follows: Based on the damage feature information and historical data, analyze the possible causes of surface damage of elevator steel belts and establish a damage category library; Use graph database technology to construct a knowledge graph for identifying surface damage of elevator steel belts. This graph contains entities such as damage features, damage causes, damage types, etc. and their mutual relationships, providing an intuitive view for the understanding and analysis of damage events.
[0017] In step 6, the process of constructing a decision tree is as follows:
[0018] 6.1. Data preparation: Before constructing a decision tree, training data needs to be prepared. The training data includes multimodal data (such as images, audio, spectral data, etc.) collected from the surface of elevator steel belts and corresponding damage labels. The process is as follows:
[0019] 6.1.1. Data collection: Use a high-resolution camera to collect images and videos of the steel belt surface, use a microphone to collect audio signals during the operation of the steel belt, and use a spectral analyzer to collect spectral data of the steel belt surface; Ensure that the data covers various damage types (such as scratches, corrosion, wear, fractures, etc.) and normal states;
[0020] 6.1.2. Data annotation: Professional technicians annotate the collected data to clarify the damage type or normal state corresponding to each piece of data. The annotation results serve as the supervision signal for decision tree training;
[0021] 6.1.3. Data preprocessing: Perform contrast enhancement, noise removal, normalization, etc. on the image data, perform filtering, feature extraction (such as MFCC, zero-crossing rate, etc.) on the audio data, perform smoothing processing, feature extraction (such as specific band intensity) on the spectral data, and convert all data into a unified feature vector form for easy input into the decision tree model.
[0022] 6.2. Feature selection: Feature selection is one of the key steps in constructing a decision tree. Good features can significantly improve the performance and accuracy of the decision tree. The process is as follows:
[0023] 6.2.1. Feature Extraction: Image features include edge detection (such as the Canny operator), texture features (such as GLCM), color histograms, etc.; audio features include MFCC (Mel Frequency Cepstral Coefficients), zero-crossing rate, spectral centroid, etc.; spectral features include the intensity of specific bands, the shape of spectral curves, etc.; other features include environmental parameters such as the running speed of the steel strip and the load.
[0024] 6.2.2. Feature Evaluation: Use metrics such as Information Gain (IG), Gain Ratio, and Gini Impurity to evaluate the classification ability of each feature, and select the feature that is most helpful for damage classification as the decision node.
[0025] 6.3. Construction of Decision Tree: The construction of a decision tree is a recursive process. By continuously dividing the dataset until the stopping condition is met, the process is as follows:
[0026] 6.3.1. Select the Optimal Feature: Calculate the information gain or Gini impurity of each feature, and select the optimal feature as the splitting criterion for the current node;
[0027] 6.3.2. Divide the Dataset: According to the selected feature, divide the dataset into multiple subsets, and each subset corresponds to a possible value of the feature;
[0028] 6.3.3. Recursive Construction: Recursively repeat the above process for each subset until the stopping condition is met:
[0029] The samples in the subset belong to the same class, that is, the damage types of all samples are the same;
[0030] The number of samples in the subset is less than the preset threshold;
[0031] The decision tree reaches the preset maximum depth;
[0032] Leaf Node Labeling:
[0033] When the recursion reaches the leaf node, label the output result of the leaf node according to the majority class of the samples in the subset;
[0034] 6.4. Pruning: Decision trees are prone to overfitting, especially when the depth of the tree is relatively deep. Therefore, it is necessary to prune the decision tree to improve the generalization ability of the model. The process is as follows:
[0035] 6.4.1. Pre-Pruning: During the growth of the decision tree, stop the growth of the tree in advance. For example, set the maximum tree depth or the minimum number of samples;
[0036] 6.4.2. Post-Pruning: After the decision tree has grown completely, reduce the complexity of the tree through pruning.
[0037] Common pruning methods include: Cost-Complexity Pruning: Select the optimal pruning position by introducing a penalty term (such as the complexity of the tree). Or Minimum Error Pruning: Select the tree structure with the minimum error after pruning through the validation set.
[0038] 6.5. Model Validation: To ensure the accuracy and reliability of the decision tree model, it needs to be validated. The process is as follows:
[0039] 6.5.1. Cross-Validation: Use cross-validation (such as K-fold cross-validation) to evaluate the performance of the model;
[0040] Divide the dataset into K subsets. Each time, use K - 1 subsets to train the model, and the remaining 1 subset is used for validation; repeat K times, and calculate metrics such as the average accuracy, recall rate, and F1 score of the model.
[0041] 6.5.2. Test Set Validation: Validate the performance of the model on an independent test set to ensure that the model can accurately identify damage on unseen data.
[0042] 6.6. Optimization of the Decision Tree: According to the validation results, optimize the decision tree to further improve its performance. The process is as follows:
[0043] 6.6.1. Feature Engineering: Add new features or adjust existing features to improve the classification ability of the model;
[0044] 6.6.2. Parameter Tuning: Adjust the parameters of the decision tree, such as the maximum depth, minimum number of samples, splitting criterion, etc.; Use Grid Search or Random Search to find the optimal parameter combination.
[0045] 6.6.3. Ensemble Learning: Use ensemble methods (such as random forest, Boosting) to improve the performance of the decision tree.
[0046] Preferably, in 6.3.3, set a threshold to verify and adjust the recognition result.
[0047] More preferably, in 6.6.3, the random forest can effectively reduce overfitting by constructing multiple decision trees and voting to select the final result.
[0048] The present invention adopts multi-modal data acquisition technology to collect image, video, audio and spectral data on the surface of elevator steel belts in real time. These data go through a series of preprocessing steps, including image contrast enhancement, image smoothing, image segmentation and extraction, image cleaning and data augmentation, to improve data quality and prepare for subsequent feature extraction. The preprocessed data will be used to extract damage feature information on the surface of elevator steel belts, and these feature information include but are not limited to edge features, texture features, spectral features and audio features. By constructing a comprehensive diagnostic decision tree and combining machine learning algorithms such as support vector machine (SVM), neural network, etc., the extracted features are analyzed to identify and classify different damage types. In addition, this method also includes a damage cause analysis module, which can analyze the possible causes of the damage on the surface of elevator steel belts according to the damage feature information and historical data, and establish a damage category library. Using graph database technology, an identification knowledge graph of the damage on the surface of elevator steel belts is constructed, which contains entities such as damage features, damage causes, damage types and their mutual relationships, providing an intuitive view for the understanding and analysis of damage events; finally, by comparing the output results of the comprehensive decision tree and the results of the knowledge graph, the identification results of the damage on the surface of elevator steel belts are generated.
[0049] The beneficial effects of the present invention are mainly manifested in that: this method verifies and adjusts the identification results by setting thresholds to ensure the accuracy and reliability of the identification results; it not only improves the identification efficiency of the damage on the surface of elevator steel belts, but also reduces costs and enhances the safety and reliability of elevators. Brief Description of the Drawings
[0050] Figure 1 It is a schematic flowchart of a method for identifying damage on the surface of elevator steel belts.
[0051] Figure 2 It is a principle block diagram of a system for implementing a method for identifying damage on the surface of elevator steel belts. Detailed Embodiments
[0052] The present invention will be further described below with reference to the drawings.
[0053] Refer to Figure 1 and Figure 2 , a method for identifying damage on the surface of elevator steel belts, includes the following steps:
[0054] Step 1, obtain multi-modal data on the surface of elevator steel belts, and the multi-modal data includes video, image, audio and spectral data;
[0055] Step 2, preprocess the multi-modal data; the preprocessing steps include image contrast enhancement, image smoothing, image segmentation and extraction, image cleaning and data augmentation;
[0056] Step 3: Extract the damage feature information on the surface of the elevator steel belt: Use edge detection and texture analysis image processing techniques to extract edge features, texture features, spectral features, and audio features;
[0057] Step 4: Analyze the reasons for the damage on the surface of the elevator steel belt. The damage cause analysis includes establishing a damage category library and using machine learning algorithms to classify and identify damage events.
[0058] Step 5: Construct an identification knowledge graph for the damage on the surface of the elevator steel belt;
[0059] Use graph database technology to construct an identification knowledge graph for the damage on the surface of the elevator steel belt. The process is as follows: According to the damage feature information and historical data, analyze the possible reasons for the damage on the surface of the elevator steel belt and establish a damage category library; Use graph database technology to construct an identification knowledge graph for the damage on the surface of the elevator steel belt. This graph contains entities such as damage features, damage reasons, damage types, etc. and their mutual relationships, providing an intuitive view for the understanding and analysis of damage events.
[0060] Step 6: Integrate the output results of the decision tree and the comparison results of the knowledge graph to generate the final damage identification result.
[0061] In Step 6, the process of constructing the decision tree is as follows:
[0062] 6.1 Data preparation: Before constructing the decision tree, it is necessary to prepare training data. The training data includes multimodal data (such as images, audio, spectral data, etc.) collected from the surface of the elevator steel belt and the corresponding damage labels. The process is as follows:
[0063] 6.1.1 Data collection: Use a high-resolution camera to collect images and videos of the steel belt surface, use a microphone to collect audio signals during the operation of the steel belt, and use a spectral analyzer to collect spectral data of the steel belt surface; Ensure that the data covers various damage types (such as scratches, corrosion, wear, fracture, etc.) and normal states;
[0064] 6.1.2 Data annotation: Professional technicians annotate the collected data to clarify the damage type or normal state corresponding to each piece of data. The annotation results serve as the supervision signal for decision tree training;
[0065] 6.1.3 Data preprocessing: Perform contrast enhancement, noise removal, normalization, etc. on the image data, perform filtering and feature extraction (such as MFCC, zero-crossing rate, etc.) on the audio data, perform smoothing processing and feature extraction (such as specific band intensity) on the spectral data, and convert all data into a unified feature vector form for easy input into the decision tree model.
[0066] 6.2. Feature Selection: Feature selection is one of the key steps in constructing a decision tree. Good features can significantly improve the performance and accuracy of the decision tree. The process is as follows:
[0067] 6.2.1. Feature Extraction: Image features include edge detection (such as the Canny operator), texture features (such as GLCM), color histograms, etc.; audio features include MFCC (Mel Frequency Cepstral Coefficients), zero-crossing rate, spectral centroid, etc.; spectral features include the intensity of specific bands, the shape of spectral curves, etc.; other features include environmental parameters such as the running speed of the steel strip and the load.
[0068] 6.2.2. Feature Evaluation: Use metrics such as Information Gain (IG), Gain Ratio, and Gini Impurity to evaluate the classification ability of each feature, and select the feature that is most helpful for damage classification as the decision node.
[0069] 6.3. Construction of the Decision Tree: The construction of the decision tree is a recursive process. By continuously dividing the dataset until the stopping condition is met, the process is as follows:
[0070] 6.3.1. Select the Optimal Feature: Calculate the information gain or Gini impurity of each feature, and select the optimal feature as the splitting criterion for the current node;
[0071] For example, if a certain feature can divide the dataset into purer subsets (i.e., the samples in the subset have a more homogeneous class), then the information gain of this feature is higher.
[0072] 6.3.2. Divide the Dataset: According to the selected feature, divide the dataset into multiple subsets, and each subset corresponds to a possible value of this feature;
[0073] For example, for a binary feature (such as "scratched" and "not scratched"), the dataset will be divided into two subsets.
[0074] 6.3.3. Recursive Construction: Recursively repeat the above process for each subset until the stopping condition is met:
[0075] The samples in the subset belong to the same class, that is, the damage types of all samples are the same;
[0076] The number of samples in the subset is less than the preset threshold;
[0077] The decision tree reaches the preset maximum depth;
[0078] Leaf Node Labeling:
[0079] When the recursion reaches the leaf node, label the output result of the leaf node according to the majority class of the samples in the subset;
[0080] 6.4. Pruning: Decision trees are prone to overfitting, especially when the tree depth is relatively deep. Therefore, it is necessary to prune the decision tree to improve the generalization ability of the model. The process is as follows:
[0081] 6.4.1. Pre-pruning: During the growth of the decision tree, stop the growth of the tree in advance. For example, set the maximum tree depth or the minimum number of samples.
[0082] For example, when the number of samples in a node is less than 10, stop further splitting.
[0083] 6.4.2. Post-pruning: After the decision tree has grown completely, reduce the complexity of the tree through pruning.
[0084] Common pruning methods include: Cost-Complexity Pruning: Select the optimal pruning position by introducing a penalty term (such as the complexity of the tree). Or Minimum Error Pruning: Select the tree structure with the minimum error after pruning through the validation set.
[0085] 6.5. Model Validation: To ensure the accuracy and reliability of the decision tree model, it is necessary to validate it. The process is as follows:
[0086] 6.5.1. Cross-validation: Use cross-validation (such as K-fold cross-validation) to evaluate the performance of the model;
[0087] Divide the dataset into K subsets. Each time, use K - 1 subsets to train the model, and the remaining 1 subset is used for validation; repeat K times, and calculate metrics such as the average accuracy, recall rate, and F1 score of the model.
[0088] 6.5.2. Test Set Validation: Validate the performance of the model on an independent test set to ensure that the model can accurately identify damage on unseen data.
[0089] 6.6. Optimization of Decision Trees: According to the validation results, optimize the decision tree to further improve its performance. The process is as follows:
[0090] 6.6.1. Feature Engineering: Add new features or adjust existing features to improve the classification ability of the model;
[0091] For example, combining image texture features and audio features may be more effective than single features.
[0092] 6.6.2. Parameter Tuning: Adjust the parameters of the decision tree, such as the maximum depth, minimum number of samples, splitting criterion, etc.;
[0093] Use Grid Search or Random Search to find the optimal parameter combination.
[0094] 6.6.3. Ensemble learning: Use ensemble methods (such as random forest, Boosting) to improve the performance of decision trees.
[0095] Random forest can effectively reduce overfitting by constructing multiple decision trees and voting for the final result.
[0096] In step 6, set a threshold to verify and adjust the recognition result.
[0097] This embodiment also provides a recognition system for elevator steel belt surface damage, including a collection module, a preprocessing module, a model construction module, an analysis module, a knowledge graph construction module, and a recognition module.
[0098] To achieve accurate and rapid recognition of elevator steel belt surface damage, the implementation process is as follows:
[0099] First, this embodiment prepares devices such as highly sensitive and high-resolution sensors and cameras for real-time collection of multi-modal data on the elevator steel belt surface, including video, images, audio, and spectral information. These data are the basis for subsequent processing and recognition work.
[0100] In the data preprocessing stage, the collected image data is enhanced to improve the clarity and contrast of the images, making the detailed information on the elevator steel belt surface more obvious; at the same time, algorithms such as filtering are used to remove noise and interference in the images to ensure the accuracy of subsequent processing; in addition, image segmentation algorithms are used to separate the elevator steel belt surface from the background for more refined processing and analysis.
[0101] In the feature extraction link, edge features, texture features, spectral features, and audio features of the elevator steel belt surface are extracted, and these features are crucial for identifying damage on the elevator steel belt surface.
[0102] Next, a recognition knowledge graph of elevator steel belt surface damage is constructed using machine learning or deep learning algorithms; these graphs learn and identify damage features on the elevator steel belt surface by training a large number of sample data; in the graph training stage, the preprocessed data and the extracted feature information are input into the recognition graph for training and optimization, continuously adjusting the parameters and structure of the graph to improve the recognition accuracy and robustness of the graph.
[0103] Finally, the trained atlas is applied to the actual elevator steel belt surface damage recognition task, and the recognition results are output, including information such as the location, type, and severity of the damage. After analyzing and evaluating these recognition results, corresponding maintenance and repair suggestions are put forward to ensure the safe operation of the elevator. At the same time, the recognition results and relevant information are recorded and fed back to relevant departments and personnel for subsequent tracking and processing.
[0104] This embodiment covers the whole process from data acquisition, preprocessing, feature extraction to damage recognition, aiming to achieve accurate and rapid recognition of elevator steel belt surface damage, thereby improving the safety and efficiency of the elevator industry.
[0105] The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept and is only for illustrative purposes. The protection scope of the present invention should not be regarded as limited to the specific forms stated in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be conceived by those of ordinary skill in the art according to the inventive concept of the present invention.
Claims
1. A method for identifying surface damage of elevator steel belt, characterized in that: The method comprises the following steps: Step 1, obtaining multimodal data of the elevator steel belt surface; Step 2: Preprocess the multimodal data; Step 3, extracting damage characteristic information of the elevator steel belt surface; Step 4: Analyze the cause of damage to the elevator steel belt surface; Step 5: Construct a knowledge graph for identifying surface damage of elevator steel belts; Step 6: Combine the decision tree output results and the knowledge graph comparison results to generate the final damage identification results.
2. A method for identifying surface damage of an elevator steel belt according to claim 1, characterized in that: In step 1, the multimodal data includes video, image, audio and spectral data.
3. A method for identifying surface damage of an elevator steel belt according to claim 1 or 2, characterized in that: In step 2, the preprocessing steps include image contrast enhancement, image smoothing, image segmentation and extraction, image cleaning and data enhancement processing.
4. A method for identifying surface damage of an elevator steel belt according to claim 1 or 2, characterized in that: In step 3, edge detection and texture analysis image processing technology are used to extract edge features, texture features, spectral features and audio features.
5. A method for identifying surface damage of an elevator steel belt according to claim 1 or 2, characterized in that: In step 4, the damage cause analysis includes establishing a damage category library and using a machine learning algorithm to classify and identify damage events.
6. A method for identifying surface damage of an elevator steel belt according to claim 1 or 2, characterized in that: In step 5, a knowledge graph for identifying surface damage of elevator steel strips is constructed using graph database technology. The process is as follows: based on damage feature information and historical data, possible causes of surface damage of elevator steel strips are analyzed, and a damage category library is established; and a knowledge graph for identifying surface damage of elevator steel strips is constructed using graph database technology. The graph contains entities of damage features, damage causes, damage types, and their interrelationships, providing an intuitive view for understanding and analyzing damage events.
7. A method for identifying surface damage of an elevator steel belt according to claim 1 or 2, characterized in that: In step 6, the process of constructing a decision tree is as follows: 6.
1. Data preparation: Before building a decision tree, it is necessary to prepare training data, which includes multimodal data collected from the surface of the elevator steel belt and the corresponding damage labels. The process is as follows: 6.1.
1. Data collection: Use a high-resolution camera to collect images and videos of the steel strip surface, use a microphone to collect audio signals when the steel strip is running, and use a spectrum analyzer to collect spectrum data on the steel strip surface; ensure that the data covers various damage types and normal states; 6.1.2 Data labeling: Label the collected data to clarify the damage type or normal state corresponding to each piece of data. The labeling results are used as supervisory signals for decision tree training; 6.1.
3. Data preprocessing: perform contrast enhancement, noise removal, and normalization on image data, filter and feature extraction on audio data, smooth and feature extraction on spectral data, and convert all data into a unified feature vector form to facilitate the input of the decision tree model; 6.
2. Feature selection, the process is: 6.2.
1. Feature extraction: Image features include edge detection, texture features and color histogram; audio features include MFCC, zero crossing rate and spectrum centroid; spectral features include the intensity of a specific band and the shape of the spectrum curve; other features include the running speed and load of the steel belt; 6.2.
2. Feature evaluation: Use information gain, gain ratio, and Gini impurity indicators to evaluate the classification ability of each feature, and select the feature that is most helpful for damage classification as the decision node; 6.
3. Construction of decision tree: The construction of decision tree is a recursive process, which continuously divides the data set until the stopping condition is met. The process is: 6.3.
1. Select the optimal feature: Calculate the information gain or Gini impurity of each feature and select the optimal feature as the partitioning criterion for the current node; 6.3.
2. Divide the data set: Divide the data set into multiple subsets based on the selected features, each subset corresponds to a possible value of the feature; 6.3.
3. Recursive construction: Repeat the above process recursively for each subset until the stopping condition is met: The samples in the subset belong to the same category, that is, all samples have the same injury type; The number of samples in the subset is less than the preset threshold; The decision tree reaches the preset maximum depth; Leaf node tags: When the recursion reaches a leaf node, the output result of the leaf node is labeled according to the majority category of the samples in the subset; 6.
4. Pruning: The process is as follows: 6.4.
1. Pre-pruning: Stop the growth of the decision tree in advance during its growth, for example, by setting the maximum tree depth or the minimum number of samples. 6.4.
2. Post-pruning: After the decision tree is fully grown, pruning is used to reduce the complexity of the tree. Pruning methods include: cost complexity pruning, which introduces penalty terms to select the optimal pruning position; or minimum error pruning: select the tree structure with the minimum error after pruning through the validation set; 6.
5. Model verification, the process is: 6.5.
1. Cross-validation: Use cross-validation to evaluate the performance of the model. Divide the data set into K subsets, use K-1 subsets to train the model each time, and use the remaining 1 subset for validation. Repeat K times and calculate the average accuracy, recall, and F1 score of the model. 6.5.
2. Test set verification: Verify the performance of the model on an independent test set to ensure that the model can accurately identify damage on unseen data; 6.
6. The optimization process of decision tree is as follows: 6.6.
1. Feature Engineering: Add new features or adjust existing features to improve the classification ability of the model; 6.6.
2. Parameter adjustment: Adjust the parameters of the decision tree, including the maximum depth, minimum number of samples, and splitting criteria; use grid search or random search to find the optimal parameter combination; 6.6.
3. Ensemble learning: Use ensemble methods to improve the performance of decision trees.
8. A method for identifying surface damage of an elevator steel belt as claimed in claim 7, characterized in that: In 6.3.3, a threshold is set to verify and adjust the recognition result.
9. A method for identifying surface damage of an elevator steel belt as claimed in claim 7, characterized in that: In 6.6.3, random forest constructs multiple decision trees and votes to select the final result.