Outlier Classification in Object Recognition Using Inlier and Outlier Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition processes face challenges in accurately classifying unlabeled data, leading to misclassification errors when new objects are encountered, as they cannot be defined using existing labels, and all objects cannot be labeled, resulting in errors during classification.
Innovation Solution
A method and device for classifying outliers using unlabeled data and labeled data to set an inlier region and an outlier region, where a deep convolutional neural network generates attributes, and the control unit selects attributes based on internal density and external separation to map unlabeled data to the appropriate region, preventing misclassification by forming a separate outlier region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object recognition uses labeled data to define classification classes, then classification accuracy for known objects is improved, but inability to handle new unlabeled objects causes misclassification errors
Solution Approach 1:
The patent segments the classification space into two distinct regions: inlier region for known objects and outlier region for new objects. This segmentation allows the system to handle both labeled and unlabeled data appropriately, preventing misclassification of new objects while maintaining accuracy for known objects.
Solution Approach 2:
The patent introduces a new dimension to the classification problem by adding an 'outlier region' to the traditional classification space. This dimensional extension enables the system to accommodate new unlabeled objects without forcing them into existing classes, thereby improving both accuracy and adaptability.
2Ease of manufacture
If all objects are forced into existing classification labels, then labeling process is simplified, but classification errors occur for objects that do not fit existing labels
Solution Approach 1:
The patent extracts objects that do not fit existing labels into a separate outlier region. This extraction process maintains the simplicity of labeling for known objects while ensuring reliability by isolating objects that cannot be confidently classified into existing categories.
3Measurement precision
If deep convolutional neural network is used to generate attributes from unlabeled data, then feature extraction capability is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary attribute extraction using deep convolutional neural networks on unlabeled data during the training phase. By pre-computing these attributes beforehand, the system reduces computational complexity during inference, as the complex feature extraction has already been performed in advance.
Data Source
AI summary
The present invention relates to a method, device, and robot for classifying an outlier during object recognition learning using artificial intelligence. The method or device for classifying an outlier during object recognition learning according to an embodiment of the present invention sets an inlier region and an outlier region through learning using unlabeled data and labeled data.