Graph Neural Network Classification for Linear Infrastructure Elements
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for classifying elements in infrastructure models representing linear infrastructure, such as roads, face challenges due to inconsistent data nomenclature, human error, and the limitations of machine learning methods that require large datasets and fail to consider contextual information.
Innovation Solution
The proposed solution involves extracting cross sections from infrastructure models, generating graph representations to capture contextual relationships, and applying these graphs to a trained Graph Neural Network (GNN) model for accurate classification, with the option for user review and model retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual classification is used, then classification accuracy can be maintained, but time consumption increases significantly
Solution Approach 1:
The system enables automated self-classification of infrastructure elements using a graph neural network that processes cross-section data and contextual relationships, eliminating the need for manual intervention while maintaining high classification accuracy
Solution Approach 2:
The patent replaces manual mechanical classification processes with an automated computational system that uses graph neural networks to classify infrastructure elements based on geometric and contextual features
2Extent of automation
If semantic rules are used for classification, then automation is achieved, but expertise requirements increase and edge cases are not covered
Solution Approach 1:
The system replaces complex semantic rule systems with a graph neural network that automatically learns classification patterns from data, eliminating the need for expert rule crafting while handling edge cases through data-driven generalization
Solution Approach 2:
The patent transforms the classification approach by changing from rule-based parameters to data-driven model parameters, where the graph neural network automatically adjusts to different infrastructure types and edge cases through training on diverse cross-section data
3Extent of automation
If traditional machine learning techniques are used, then automation is achieved, but classification accuracy decreases due to lack of contextual information
Solution Approach 1:
The patent adds the dimension of contextual relationships between infrastructure elements to traditional machine learning approaches. By constructing graphs that capture spatial and semantic relationships across cross-sections, the system enables automated classification with high accuracy that traditional methods lacking this contextual dimension cannot achieve
Solution Approach 2:
The system merges multiple types of information including geometric features, contextual relationships, and cross-sectional data into a unified graph representation that is processed by the graph neural network, combining the strengths of various data sources to achieve both automation and high accuracy
4Measurement precision
If large datasets are used for training, then machine learning model accuracy improves, but data availability becomes a constraint for linear infrastructure
Solution Approach 1:
The patent segments the infrastructure model into discrete cross-sections along the linear element, creating manageable data units that can be processed individually. This segmentation allows the graph neural network to learn from a smaller number of segmented cross-sections while maintaining the ability to accurately classify elements throughout the entire infrastructure, overcoming the limitation of small overall datasets
Data Source
AI summary
In example embodiments, improved techniques are provided for classifying elements of an infrastructure model that represents linear infrastructure (e.g., roads). The techniques may extract a set of cross sections perpendicular to a centerline of the linear infrastructure from the infrastructure model, generate a graph representation of each cross section to produce a set of graphs having nodes that represent elements and edges that represent contextual relationships, provide the set of graphs to a trained graph neural network (GNN) model, and produce therefrom class predictions for the elements. The class predictions may include one or more predicted classes for each element with a respective confidence. A best predicted class for each element may be selected and assigned to the element, thereby creating a new version of the infrastructure model. For elements that extend through multiple cross sections, the selection may involve aggregating predicted classes originating from the different graphs.