Bolt classification system for motorcycle retail shop

Abstract

The COVID-19 situation in Thailand has led to a rise in online purchase orders, resulting in a higher demand for using motorcycles for shipment, which has also increased the demand for essential components, notably durable and aesthetically pleasing bolts. To enrich their business opportunities, bolts of 19 classes are gathered from a motorcycle shop to establish a systematic bolt classification procedure containing feature extraction stage and classification stage. A feature extraction is formulated from utilization of steps, which are background removal, contour extraction, image rotation, cropping, structural analysis, dominant color analysis, hole detection, and calculating head-to-whole length ratio. Subsequently, five classification models, comprising multi-layer perceptron, random forest, decision tree, support vector machine, and logistic regression, are employed to identify the appropriate class for each bolt. The results indicate that the multi-layer perceptron stands out as the most effective classification model with the proposed features.