Adaptive surface inspection via interactive evolution

An increasingly frequent application of Machine Vision technologies is in automated surface inspection for the detection of defects in manufactured products. Such systems offer significant benefits in terms of cost, detection rates, and user-satisfaction over conventional human inspection systems. However, they usually require significant investment of expert time to set up, are “brittle” in the sense of being highly specialised to the task for which they are tuned, and are consequently sensitive to changes in operating conditions or product specifications. This raises problems within an industrial setting, where operating conditions or requirements may change, and the end-users are experts in their manufacturing field, but not in image processing.In this paper, we describe the development of a rapidly reconfigurable system in which the users’ tacit knowledge and requirements are elicited via a process of Interactive Evolution, finding the image processing parameters to achieve the required goals without any need for specialised knowledge of the machine vision system. We show that the resulting segmentation can be quickly and easily evolved from scratch, and achieves detection rates comparable to those of a hand-tuned system on a hot-rolled steel defect recognition problem.