Dataset of Packaging Defects and Associated Cost of Poor Quality in a Wine Bottling Process

This dataset contains 9,651 images acquired from a real wine bottling packaging process. The dataset is intended for applications in quality control, computer vision, and process improvement in industrial environments.
The images are organized into three defect classes: missing cap, missing bottle, and wrong bottle. Each image reflects actual production conditions, where bottles are arranged in boxes with a capacity of 12 units, consistent with the real packaging configuration.
To ensure variability and robustness in visual features, multiple cap colors are included, namely black, red, orange, blue, purple, green, and silver. This diversity supports the development of more generalizable defect detection and classification models.
The dataset was divided into training (70%), validation (20%), and testing (10%) subsets, facilitating its direct use in supervised learning tasks.
Additionally, the dataset includes a configuration file (data.yaml) that specifies the directory paths for each subset (training, validation, and testing) as well as the corresponding class labels. This file enables straightforward integration with deep learning frameworks, particularly those based on YOLO architectures.
This dataset can support research in defect detection, automated inspection systems, and quality improvement methodologies such as Six Sigma and Industry 4.0 applications.