KNUST Research Week and Scientific Conference: Empowering Researchers to Achieve Impactful Outcomes

This article presents MeatScan, a curated image dataset developed to support deep learning-based binary classification of cow meat as fresh or spoiled. The dataset comprises 11,000 high-resolution RGB images (5627 fresh and 5373 spoiled) captured in real-world Ghanaian environments, including open-air markets, butcher shops, and cold storage facilities. Images were labeled based on observable visual cues such as texture, colour, and surface condition, with annotations verified under natural lighting by trained data collectors. MeatScan provides structured and contextually rich visual data for supervised learning in food quality monitoring. It addresses a key gap between advances in computer vision and practical food safety inspection, especially in low-resource settings. The dataset supports experimentation with convolutional neural networks, transfer learning, and data augmentation, and serves as a real-world benchmark for evaluating model robustness to lighting variability, diverse meat textures, and complex backgrounds.