Sewer Line Inspection Using Hybrid Supervised and Unsupervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sewer inspection methods are time-consuming, error-prone, and subjective due to the lack of advanced image analysis techniques, with existing automated systems relying on limited supervised learning methods and failing to achieve human expert quality accuracy, and lacking 3D reconstruction capabilities from 2D videos.
Innovation Solution
A computer-implemented method combining supervised and unsupervised learning to analyze sewer line videos, where supervised learning identifies defects and unsupervised learning provides geometrical information for 3D reconstruction, synergistically improving accuracy and report generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If supervised learning methods are used for defect identification, then automation is improved, but accuracy does not reach human expert quality
Solution Approach 1:
The patent combines supervised learning (for defect classification) and unsupervised learning (for anomaly detection and geometric analysis) into a hybrid system. This merging allows the system to achieve both automation and high accuracy by leveraging the strengths of both approaches: supervised learning provides automated defect categorization while unsupervised learning enhances detection accuracy by identifying patterns without human labels and performing 3D reconstruction.
2Measurement precision
If manual inspection is conducted, then accuracy can be maintained, but the inspection process becomes time-consuming
Solution Approach 1:
The system performs preliminary automated analysis using both supervised and unsupervised learning methods to pre-identify defects and generate initial assessments. This preliminary action filters and prioritizes critical defects before human expert review, reducing the time required for manual inspection while maintaining high accuracy through the combination of automated detection and expert verification.
3Ease of operation
If only supervised learning is used, then defect classification is achieved, but 3D reconstruction capability is lost
Solution Approach 1:
The patent implements a universal learning framework where unsupervised learning components serve multiple functions: they enable 3D reconstruction from 2D video frames, perform geometric analysis of defects, and supplement supervised learning for anomaly detection. This multi-functionality allows the system to maintain defect classification capabilities while simultaneously recovering 3D reconstruction information that would be lost with supervised learning alone.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed herein is a computer-implemented method of automated or assisted inspection of a sewer line. The method includes performing a supervised data analysis stream on an input survey video of the sewer line, which includes performing prediction with a supervised learning method on frames of the input survey video to analyse the frames regarding defects. The method further includes performing an unsupervised data analysis stream on the input survey video which includes performing prediction with an unsupervised learning method on the input survey video to obtain geometrical information; and performing a 3D reconstruction of the sewer line based on the geometrical information. Then results of the supervised data analysis stream and the unsupervised data analysis stream are combined.