Defect Detection through Customised Reduction and Hybrid Convolution Classification over Super-pixel Clusters

Abstract

Defect detection is the process of locating defects or anomalies within an object that include changes in textures, features, patterns, missing part, along with other object modifications. The paper discusses some of the main challenges of defect detection including details on sample selection, object orientation, semantic segmentation and image defect classification. This paper focuses on applying modified machine and deep learning models to analyse defects with wide object invariance. We demonstrate algorithms that perform multi-class classification with improvements in the image segmentation process that directly connect to the deep model architecture. Before applying learning algorithms, the paper also demonstrates the value of sample selection together with a more simplified normalised dimension reduction based on image downscaling even before using the convolution operation of the convolutional neural networks (CNN).