Wei-Lun Lin

Title: Wei-Lun Lin and Adaptive Recognition of Unknown Objects.

Introduction

Wei-Lun Lin is an inventor based in Taichung, Taiwan. His patent work focuses on neural-network methods for identifying and learning about objects that an image-recognition model cannot initially recognize. He is associated with Feng Chia University and has 2 patents.

Latest Patents

Lin’s patent, “Image Analyzation Method with Unrecognized Object Memory Mechanism,” describes a system that detects when an object does not fit a known category and stores information about it. The method extracts image features, assigns an attention representative value, and groups target objects into designated directories according to their values. Once a directory reaches a specified quantity threshold, the system assigns a new category label to the images and uses them as retraining data. This approach is designed to help recognition models learn new categories without requiring people to label every previously unrecognized object manually.

His patent, “Comprehension Method of Neural Network for Unrecognizable Objects,” compares the number of objects recognized by a model with the total number of objects identified through image segmentation. If the counts differ, the system determines that the image contains one or more unknown objects and trains the model using that information. The method aims to give neural networks a way to detect and learn about unfamiliar objects.

Career Highlights

Lin’s inventions address a practical challenge in computer vision: recognition systems may encounter objects outside the categories on which they were trained. His patented methods combine object detection, feature analysis, memory mechanisms, and model retraining to support ongoing learning.

Collaborations

Lin is associated with Feng Chia University. His coworker, Shao-Mao Wang, is also named in connection with his work.

Conclusion

Wei-Lun Lin’s patents explore ways for neural networks to identify unfamiliar objects and incorporate them into future recognition, reducing reliance on manual labeling.