A Two-Stage Framework for Small Character Detection in the Manufacturing Industry

In recent years, research on machine vision systems with artificial intelligence has rapidly increased in the manufacturing industry for identifying defects. Specifically, Optical Character Recognition (OCR) technology has witnessed significant advancements, enabling the effective recognition of characters inscribed on parts. However, a key limitation of prior approaches lies in their restriction to character-centric images. This constraint renders them unsuitable for real-world manufacturing environments where small characters are encountered within a large field of view (FOV). In order to address this limitation, we present a novel two-stage framework. This framework commences with the detection of resistor regions within the image. Subsequently, the identified regions containing the small characters are fed into an OCR model for accurate recognition. In addition, we propose a novel labeling method that leverages a clustering algorithm. Experimental results demonstrate the effectiveness and efficiency of our framework in the real manufacturing industry setting.