Embargoed and Restricted Access
Reason: Under embargo until April 2024. After this date a copy can be supplied under Section 51(2) of the Australian Copyright Act 1968 by submitting a document delivery request through your library
Spectrally Enhanced Image Processing and Nonlinear Alternating Decision Trees
thesis
posted on 2021-04-18, 08:29 authored by SANUSH KHYLE ABEYSEKERAVisual inspection is a sub-class of non-destructive testing extensively used in high volume manufacturing. Existing inspection technologies are heavily reliant on complex machine learning algorithms and expensive cameras for visual recognition. This research investigates an alternative framework whereby light incident on the target object is optimized to enhance object contrast before classification. Optimized illumination spectra are calculated by a machine learning algorithm called Alternating Decision Trees (ADTree). The ADTree algorithm is a powerful classification algorithm on its own right. Further extensions are proposed to ADTree to improve its effectiveness as a tool for data exploration.