Asymmetric Autoencoder for Image Identification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image identification technologies often misjudge images as abnormal due to minor displacements or rotations, leading to reduced accuracy and increased operation time, and require frequent updates of abnormal images in databases, making the process cumbersome and inefficient.
Innovation Solution
An image identification method using an asymmetric automatic codec, trained on normal state images, performs feature extraction and reconstruction to generate reconstructed images, allowing for accurate comparison and determination of normal or abnormal states without the need for frequent database updates, and includes a robotic arm for automated handling of defective products.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image comparison methods are used to identify defects by comparing acquired images against standard images, then defect detection capability is provided, but misjudgment occurs due to minor displacements or rotations leading to reduced identification accuracy
Solution Approach 1:
Instead of comparing the acquired image against a standard image to detect defects, the patent inverts the approach by comparing the standard image against the acquired image. This inversion allows the system to adapt to variations in the acquired image (such as displacements or rotations) while still effectively detecting actual defects, thereby reducing misjudgments and improving identification accuracy
Solution Approach 2:
The patent changes the comparison parameters by introducing a similarity threshold and allowing for geometric transformations (displacement, rotation) during the comparison process. By adjusting these parameters, the system can tolerate minor variations without triggering false defect detections, thus improving both accuracy and reliability
2Reliability
If all produced abnormal images are stored in a database for continuous updates, then comprehensive defect coverage is achieved, but operation becomes tedious and inefficient due to frequent database updates
Solution Approach 1:
The patent extracts only the essential defect characteristics from abnormal images rather than storing the entire image database. By extracting and storing only the key defect features, the system maintains comprehensive defect detection coverage while significantly reducing the burden of database updates and improving operational efficiency
Solution Approach 2:
Instead of storing and managing numerous abnormal images in a database, the patent creates simplified representations or copies of the defect characteristics. These copied defect features are then used for comparison, achieving the same defect detection coverage with much easier and faster operations
3Reliability
If conventional image comparison methods are used, then defect detection is provided, but operation time increases due to tedious database updates and misjudgment corrections
Solution Approach 1:
The patent performs preliminary actions by pre-processing the acquired image to correct for displacements and rotations before the comparison step. This preliminary alignment reduces misjudgments and eliminates the need for time-consuming database updates, thereby maintaining reliable defect detection while significantly reducing operation time
Data Source
AI summary
An image identification method is provided, including: storing at least one normal state image of at least one test object; an automatic codec receiving the at least one normal state image to become a trained automatic codec; at least one camera device capturing at least one state image of the at least one test object; a computer device receiving the at least one state image, and the trained automatic codec performing feature extraction and reconstruction on the at least one state image to generate at least one reconstructed state image; and the computer device comparing the at least one state image and the at least one reconstructed state image, and determining whether the at least one state image is a normal state image. The present invention also provides an image identification system.