Semantic Segmentation of Source Geometry Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for creating detailed and informative electronic maps are inefficient, time-consuming, and require significant human involvement, as they rely on manual identification and labeling of different usage areas on maps.
Innovation Solution
A method and system that perform semantic inference on images of locations using artificial intelligence, which involves converting vector or raster images into labeled representations by assigning pixel and partition class labels based on usage types, and utilizing neural networks for classification and labeling, allowing for automatic generation of detailed and informative electronic maps with minimal human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual identification and labeling of usage areas is performed, then map accuracy and detail can be achieved, but the process becomes inefficient and time-consuming
Solution Approach 1:
The patent replaces the manual mechanical process of identification and labeling with an automated computer-based system that uses image processing and neural networks to perform semantic segmentation, thereby substituting human labor with automated computational methods
Solution Approach 2:
The system enables automated self-labeling of map features through neural networks that automatically perform semantic segmentation, allowing the system to service itself without requiring manual human intervention for each labeling task
2Loss of information
If manual labeling is performed to create detailed electronic maps, then comprehensive usage area identification can be achieved, but significant human involvement and costs are required
Solution Approach 1:
The patent replaces manual human labeling operations with automated neural network-based semantic segmentation, eliminating the need for significant human involvement while maintaining comprehensive identification of usage areas through automated image analysis
Solution Approach 2:
The system creates automated copies of labeling processes through trained neural networks that can replicate the identification and labeling of usage areas without requiring original human expert intervention, thereby reducing human involvement while preserving information quality
3Productivity
If automated neural network methods are used for semantic segmentation, then map creation speed and efficiency improve, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by training neural networks in advance on labeled datasets before deployment. This pre-training phase creates ready-to-use models that can perform semantic segmentation automatically, thereby increasing map creation speed while managing system complexity through preparatory work
Solution Approach 2:
The trained neural network serves as an intermediary between the input map image and the final labeled output. This intermediary component encapsulates the complexity of automated analysis, providing a simplified interface that improves productivity while containing system complexity within the trained model
Data Source
AI summary
Systems and methods for semantic segmentation of a source geometry are provided. The method includes providing a vector image of a location; generating a raster image from the vector image, the raster image including at least one partition, the at least one partition including a plurality of pixels; assigning a pixel class label to each of the plurality of pixels, the pixel class label including a usage type; and assigning a partition class label to the at least one partition based on the pixel class labels assigned to the plurality of pixels, wherein the partition class label includes a usage type for the partition.