Object Classification Using Verifiable Representative Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object classification methods using deep learning struggle with precise classification due to the lack of proof that features learned are effective, making it difficult to accurately classify objects in microscopic images.
Innovation Solution
An object classifying apparatus that includes an object area computing section, feature selecting section, feature extracting section, feature classifying section, and output section, which computes object areas, selects and extracts features, classifies them, and outputs classification results with representative features, enabling precise object classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning is used to automatically learn features, then classification precision can be enhanced, but it becomes difficult to verify whether features are learned correctly from specific sites
Solution Approach 1:
The patent introduces an intermediary verification mechanism that bridges the black-box deep learning feature extraction process and the user's need for verification. The system extracts features using deep learning, then verifies whether these features are correctly learned from specific sites by comparing them with manually designed features or reference features. This intermediary verification step allows the system to maintain the automation benefits of deep learning while providing transparency and verification capability.
2Loss of information
If manual feature design is used, then feature verification is possible, but precise assessment is not possible when surfaces are unstructured
Solution Approach 1:
The patent merges the advantages of both manual feature design and deep learning feature extraction. It combines manually designed features (which provide verification capability) with automatically learned features from deep learning (which provide precision for unstructured surfaces). The system integrates both feature types and uses them together for classification, thereby achieving both verification capability and high precision classification.
3Ease of operation
If only classification results are output, then the system is simple to operate, but users cannot verify the classification basis
Solution Approach 1:
The patent implements a feedback mechanism that provides users with classification basis information. After performing classification, the system outputs not only the classification results but also the extracted features and their correspondence to classification decisions. This feedback loop allows users to verify the classification basis while maintaining ease of operation through automated presentation of the information.
Data Source
AI summary
Representative features which are representative features for deciding the types or states of objects are extracted, the types or states of the objects are identified on the basis of the representative features to classify the objects, and the object classification results and the representative features are output in association with an image.