Divided Training Image Labeling for Accurate Micrograph Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image identification models for micro-compositions of materials require extensive manual expertise for training data preparation, and dividing micrographic images can lead to misclassification due to altered structures, necessitating improved methods for accurate and efficient model building.
Innovation Solution
An image identification system that includes a model building unit, a training data creating unit, and units for image dividing, texture index computation, tentative label prediction, label comparison, and modification, enabling the creation of high-accuracy models by refining labels of divided training images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If micrographic images are divided to expand training data quantity, then the quantity of training data increases, but the structure of divided images may differ from original images leading to misclassification
Solution Approach 1:
The patent applies segmentation by dividing training images into multiple smaller images to expand the quantity of training data. The image dividing unit automatically divides training images into multiple divided training images, increasing the available training data while maintaining structural integrity through automated division processes.
Solution Approach 2:
The patent implements feedback mechanisms through label comparison and verification processes. The label comparison unit compares labels of divided images with original images to detect discrepancies, and the verification process uses multiple feature quantities (chromaticity, brightness, contrast, structure pattern, frequency, entropy) to verify label accuracy, providing feedback loops that ensure classification reliability.
2Reliability
If feature quantity thresholds are predefined by experts to filter divided images, then classification accuracy is maintained, but the complexity of the process increases and requires preliminary knowledge
Solution Approach 1:
The patent applies self-service by enabling automated label verification and modification without requiring expert intervention. The system automatically calculates multiple feature quantities, compares labels, identifies discrepancies, and modifies labels using the label modification unit, allowing the system to verify and correct its own training data without external expert input.
Solution Approach 2:
The patent utilizes parameter changes by evaluating multiple image feature quantities (chromaticity, brightness, contrast, structure pattern, frequency, entropy) instead of relying on predefined thresholds. This multi-parameter approach automatically adapts to different image characteristics and eliminates the need for expert-defined threshold parameters.
3Measurement precision
If manual label verification is performed on divided images, then label accuracy is improved, but the time and labor required for training data preparation increases
Solution Approach 1:
The patent replaces manual mechanical verification processes with automated computational systems. The label comparison unit automatically compares labels using computational algorithms, and the label modification unit automatically corrects discrepancies, substituting human manual verification with automated image processing and comparison mechanisms that operate rapidly without manual intervention.
Solution Approach 2:
The patent implements continuous automated label verification and modification throughout the training data preparation process. The system continuously calculates feature quantities, compares labels, and modifies discrepancies in an automated workflow, eliminating interruptions and maintaining continuous productive action without manual intervention cycles.
Data Source
AI summary
An image identification system includes: an image dividing unit which divides training images of a first training data set and assigns a label assigned to a training image from which dividing occurs to the divided training images as tentative labels; a texture index computing unit which computes texture indexes for each of the divided training images; a tentative label prediction model building unit which builds a tentative label prediction model to predict tentative labels assigned to the divided training images based on the texture indexes; and a label comparison unit which compares first tentative labels assigned to the divided training images with second tentative labels predicted with respect to the divided training images by the tentative label prediction model and extracts divided training images for which there is discrepancy between the first and second tentative labels as those images for which it is necessary to modify tentative labels.