Joint Asset Defect Detection Model for Utility Grids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently identifying and classifying utility asset types and their corresponding defects from images, particularly due to the long-tail distribution of defect types and asset classes.
Innovation Solution
A joint model for asset-defect detection is proposed, which performs long-tail object detection by jointly classifying asset type and defect status using feature data from asset classifications, thereby leveraging the long-tailed nature of training examples and data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If separately training two models to perform asset classification and defect detection, then model training simplicity is improved, but classification accuracy deteriorates due to long-tail distribution
Solution Approach 1:
The patent combines asset classification and defect detection into a single joint model that processes images simultaneously for both tasks. This merging allows the model to learn shared features and relationships between asset types and their defects, improving classification accuracy while handling the long-tail distribution of defect data more effectively than separate models.
Solution Approach 2:
The joint model performs multiple functions - both asset classification and defect detection - within a single unified framework. This multi-functional approach enables the model to leverage common patterns across different asset types and defect categories, improving overall performance on the imbalanced long-tail data distribution.
2Measurement precision
If using a joint model for asset-defect detection, then classification accuracy is improved, but model complexity increases
Solution Approach 1:
The joint model is segmented into distinct functional components: an asset classification branch and a defect detection branch, both sharing a common feature extraction backbone. This segmentation allows the model to handle complex tasks through specialized sub-components while reusing shared features, managing complexity through modular design.
Solution Approach 2:
The patent transforms the problem from two separate single-task models into a unified multi-task model that operates in an expanded parameter space. By jointly optimizing both asset classification and defect detection objectives, the model leverages additional dimensional information and relationships that improve accuracy while the shared architecture prevents exponential complexity growth.
3Reliability
If leveraging long-tailed nature of training examples in joint model, then defect detection performance is improved, but training data requirements increase
Solution Approach 1:
The patent employs parameter changes through loss function weighting and sampling strategies that adapt to the long-tail distribution of defect data. By dynamically adjusting the importance of different defect classes during training and using techniques like focal loss or oversampling of rare defects, the model achieves reliable defect detection performance without requiring proportional increases in training data for all classes.
Data Source
AI summary
This disclosure describes a system, method, and computer storage medium for joint asset and defect detection. The approach includes receiving input data including an input image of a utility asset, the input image including one or more objects. Deep neural networks are configured to generate embeddings for classification labels of the one or more objects, each embedding corresponding to a classification label and including a mapping between the classification label and a subset of feature vectors. Defect classifiers are configured to determine a likelihood of an object from the one or more objects in the input image containing a type of defect. Each defect classifier is trained to determine a type of defect based on the embeddings for the one or more classification labels. The approach includes generating an output image that includes bounding boxes for the objects and an annotation corresponding a respective object from the objects.