FWNet Wood Waste Dataset: A Hierarchical Image Collection for Furniture Manufacturing Recycling

This dataset contains a hierarchical image collection specifically designed for the automated classification and sorting of wood waste in the furniture manufacturing industry. It was developed to support the FWNet research, demonstrating the application of Deep Learning (Convolutional Neural Networks) to maximize material reuse and promote circular economy practices. The dataset is organized to support a two-layer classification architecture: Level 1 (Binary Classification): "Recyclable" vs. "Non-Recyclable" materials. Level 2 (Multiclass Subclassification): The "Recyclable" stream is further categorized into four factory-critical sub-streams: Scraps (Solid wood, MDF, MDP, or Melamine offcuts). Biomass (Sawdust, shavings, and small irregular pieces). Metals (Screws, hinges, nails, handles). Plastics (Dowels and other recyclable components). Data Collection & Characteristics: The images were curated by aggregating selected class-relevant images from public generic datasets (e.g., TrashNet) and web-scraping, additionally supplementing them with original photographs acquired directly within active furniture manufacturing workshops. Original images were captured using mobile devices (iPhone 14 and Samsung Galaxy A23) under diverse angles, lighting conditions, and partial obstructions to simulate realistic production environments. Directory Structure: The provided .zip file contains the raw, unaugmented images structured hierarchically: /Non-Recyclable /Recyclable /Scraps /Biomass /Metals /Plastics Usage: This dataset serves as the foundational data for training and evaluating computer vision models (such as InceptionV3, ResNet50, MobileNetV2) tailored for industrial wood waste management.