RAW Industrial Yarn-Breakage Image Dataset (Textile Manufacturing, 2024)

This dataset provides the RAW version of an industrial yarn-breakage image collection acquired in a real textile manufacturing environment in 2024. All images were captured using a high-resolution industrial monochrome (Mono8) area-scan camera (2592 × 2048 pixels) mounted on a mobile inspection platform operating along the spinning frame under controlled LED ring illumination.The full RAW dataset contains 3833 images, organized into seven yarn-breakage classes, each corresponding to a unique breakage pattern across three spindle positions:000g – All spindles broken (71 images)001g – Spindle 1 & 2 broken (134 images)011g – Spindle 1 broken (300 images)100g – Spindle 2 & 3 broken (120 images)101g – Spindle 2 broken (416 images)110g – Spindle 3 broken (404 images)111g – No breakage (2388 images)Each image belongs to exactly one class, consistent with the physical yarn-breakage state recorded during field acquisition.Two RAW formats are provided:BMP: Original unprocessed images directly generated by the industrial camera.PNG: Lossless PNG versions converted from BMP for easier use and reduced storage size.No preprocessing, enhancement, filtering, normalization, resizing, or data augmentation has been applied. All images preserve their original acquisition resolution (2592 × 2048, Mono8 grayscale) and maintain the real visual characteristics of industrial yarn-breakage scenarios.This dataset supports research in machine vision, machine learning, industrial automation, and textile defect inspection, particularly for yarn-breakage detection and classification under real-world manufacturing conditions.