Detection of Small Screws Using Machine Learning

Small screws are commonly employed in assenrbling similar or dissimilar components of various industrial or consumer products. In automated assembly lines, detecting and sorting small targets like screws require accurate sensing and machine learning. Identifying the size of the screw is crucial for assembling and disassembling. This study focuses on the detection of small screws captured in images using a recent machine learning platform named You Only Look Once (YOLO) Version 8. This algorithmic platform offers powerful supervised machine-learning engines with highly competitive accuracy and speed. The data set from this study consists of images of small black carbon steel flat head screws. The dataset is divided into training, development, and testing subsets. There are 24 classes of screws with extremely small difference in attributes in their sizes. The dataset is annotated using the Make Sense open-source annotating tool. The results indicate that the machine learning method demonstrates 99.3% mean average precision (mAP). This detection performance is promising in comparison to the results documented by previous studies.