A Machine Learning/Data Science Approach to Sustainability in the Manufacturing and Disposal of Fashion Products
The fashion industry faces significant challenges with textile waste, necessitating the adoption of sustainable approaches. This project aims to contribute to sustainable solutions by utilising data science for automation and improvement in efficiency and economic viability, encouraging fashion businesses to embrace sustainable practices. Specifically, the research is targeted towards the textile recycling process.
The project aims to identify existing gaps in the current textile waste recycling pipeline and explore how integrating data-driven approaches can address these gaps. While earlier research has extensively covered areas such as fabric defect detection or classification of certain fabric properties, much of the existing literature tends to be centred around manufacturing and quality control processes, with limited assessment made in the context of textile recycling. Our aim, thus, extends to exploring the potential of deep learning (DL) based strategies specifically to innovate and potentially automate the textile sorting process.
- How can the challenges posed by the fast fashion industry, especially in waste reduction, be effectively mitigated through recycling?
- What are the primary gaps and challenges in the current textile waste recycling pipeline, and how can they be addressed through the experimental design for the project?
- What is the state-of-the-art in textile waste classification and sorting, and how can digital transformation and automation, using DL-based solutions, refine these processes?
The expected outcome of this project was to create a proof-of-concept in the form of preliminary Deep Learning models that could serve as a starting point for more in-depth research into fully transforming the textile recycling process.
- Examine various aspects of textile recycling, identifying opportunities to enhance sustainability through an extensive review of existing literature.
- Identify several relevant datasets that are representative of the areas in the textile recycling process suitable for automation.
- Apply various deep learning techniques to model these areas and interpret the significance of the results.
Fabric Classification/: This directory contains the source code for the CNN model proposed for fabric texture classification built on the Fabrics Dataset.Garment Instance Segmentation/: This directory contains the source code for the Mask R-CNN model developed for Garment localisation and Instance Segmentation built on the DeepFashion2 Dataset.