Boundary-Aware Demons Algorithm with Applications in Electronic Waste Recycling

Abstract

Electronic waste (e-waste) refers to electronic devices that are nearing the end of their useful life, and are discarded, donated, or given away. Valuable metallic and plastic components in e-waste (gold, silver, platinum) are estimated to value upwards of $60 billion and although e-waste represents only 2% of solid waste streams, it can represent up to 70% of the hazardous waste that ends up in landfills. As such, e-waste recycling has garnered the attention of private companies and governments alike and represents an excellent opportunity for economic growth and environmental protection. It also poses a unique set of challenges that promote technological advancements in artificial intelligence (AI) and computer vision pipelines that can be integrated in waste recycling systems for increased automation, improved efficiency, and reduced processing time.

We focus on defect or change detection of e-waste through x-ray image analysis. Defect analysis of components in e-waste can identify 1) regions of significant change in a device, 2) damage of certain components, and 3) fraudulent components. All of which can inform downstream disassembly tasks including component segmentation and identification, device handling policies, and component removal policies. Demons deformable image registration serves as the basis for our change detection algorithm. By aligning an x-ray of a pristine device with an x-ray of a potentially defective device, a dense vector field is produced that can be analyzed to measure and quantify local deformations/transformations between the images. Using a priori annotations in the form of segmentation masks of high value components (i.e. batteries, cameras, speakers, and screws), the motion or change status of individual components can be measured.

The main contribution of this thesis is to identify current limitations of the Demons deformable image registration algorithm and address them by introducing a novel Boundary-Aware Image Registration (BAIR) algorithm that modifies demons to allow discontinuity during a normally continuous deformation process. We evaluate our method on several phone x-rays with known deformations to demonstrate that BAIR reduces the occurrence of false positives for identifying areas of extreme damage/change. Additionally, we present additional use cases for BAIR for improved image registration where discontinuity in the image transformation process is desired.