Analyzing Design Typicality by Image Classification with Deep Learning

Abstract

While deep learning has been successfully applied to many domains and industries, it is still in the first step to investigate the potential application to the field of industrial design. This paper discusses how image classification with deep learning can be used to analyze design typicality, which is the primary factor in designing product appearance and brand image. For promoting it to no-code designers such as industrial designers to adopt the process based on the machine learning tool, Waikato Environment for Knowledge Analysis (Weka) is introduced. A pilot study shows a promising approach for the designers to build datasets, pre-process, train models, test models, and measure prediction performance. This study suggests bridging image classification techniques to product design processes to advance design research and practice.