A Large-scale Evaluation of Pretraining Paradigms for the Detection of Defects in Electroluminescence Solar Cell Images
Authors:
David Torpey,
Lawrence Pratt,
Richard Klein
Abstract:
Pretraining has been shown to improve performance in many domains, including semantic segmentation, especially in domains with limited labelled data. In this work, we perform a large-scale evaluation and benchmarking of various pretraining methods for Solar Cell Defect Detection (SCDD) in electroluminescence images, a field with limited labelled datasets. We cover supervised training with semantic…
▽ More
Pretraining has been shown to improve performance in many domains, including semantic segmentation, especially in domains with limited labelled data. In this work, we perform a large-scale evaluation and benchmarking of various pretraining methods for Solar Cell Defect Detection (SCDD) in electroluminescence images, a field with limited labelled datasets. We cover supervised training with semantic segmentation, semi-supervised learning, and two self-supervised techniques. We also experiment with both in-distribution and out-of-distribution (OOD) pretraining and observe how this affects downstream performance. The results suggest that supervised training on a large OOD dataset (COCO), self-supervised pretraining on a large OOD dataset (ImageNet), and semi-supervised pretraining (CCT) all yield statistically equivalent performance for mean Intersection over Union (mIoU). We achieve a new state-of-the-art for SCDD and demonstrate that certain pretraining schemes result in superior performance on underrepresented classes. Additionally, we provide a large-scale unlabelled EL image dataset of $22000$ images, and a $642$-image labelled semantic segmentation EL dataset, for further research in develo** self- and semi-supervised training techniques in this domain.
△ Less
Submitted 27 February, 2024;
originally announced February 2024.
H2-Golden-Retriever: Methodology and Tool for an Evidence-Based Hydrogen Research Grantsmanship
Authors:
Paul Seurin,
Olusola Olabanjo,
Joseph Wiggins,
Lorien Pratt,
Loveneesh Rana,
Rozhin Yasaei,
Gregory Renard
Abstract:
Hydrogen is poised to play a major role in decarbonizing the economy. The need to discover, develop, and understand low-cost, high-performance, durable materials that can help maximize the cost of electrolysis as well as the need for an intelligent tool to make evidence-based Hydrogen research funding decisions relatively easier warranted this study.In this work, we developed H2 Golden Retriever (…
▽ More
Hydrogen is poised to play a major role in decarbonizing the economy. The need to discover, develop, and understand low-cost, high-performance, durable materials that can help maximize the cost of electrolysis as well as the need for an intelligent tool to make evidence-based Hydrogen research funding decisions relatively easier warranted this study.In this work, we developed H2 Golden Retriever (H2GR) system for Hydrogen knowledge discovery and representation using Natural Language Processing (NLP), Knowledge Graph and Decision Intelligence. This system represents a novel methodology encapsulating state-of-the-art technique for evidence-based research grantmanship. Relevant Hydrogen papers were scraped and indexed from the web and preprocessing was done using noise and stop-words removal, language and spell check, stemming and lemmatization. The NLP tasks included Named Entity Recognition using Stanford and Spacy NER, topic modeling using Latent Dirichlet Allocation and TF-IDF. The Knowledge Graph module was used for the generation of meaningful entities and their relationships, trends and patterns in relevant H2 papers, thanks to an ontology of the hydrogen production domain. The Decision Intelligence component provides stakeholders with a simulation environment for cost and quantity dependencies. PageRank algorithm was used to rank papers of interest. Random searches were made on the proposed H2GR and the results included a list of papers ranked by relevancy score, entities, graphs of relationships between the entities, ontology of H2 production and Causal Decision Diagrams showing component interactivity. Qualitative assessment was done by the experts and H2GR is deemed to function to a satisfactory level.
△ Less
Submitted 15 November, 2022;
originally announced November 2022.
