Implementation of YOLO26 for Mold Detection on White Bread Based on Digital Imagery

Abstract

White bread is highly susceptible to visible mold contamination, which causes physical deterioration and potential health risks. Conventional manual visual inspection is slow, subjective, and inconsistent, necessitating an automated detection system. This study implemented the YOLO26n algorithm for mold contamination detection on white bread based on digital imagery. A primary dataset of 300 images (150 fresh bread and 150 moldy breads) was collected independently, annotated via Roboflow, and split into 70% training, 20% validation, and 10% testing. The model was trained on Google Colab using the MuSGD optimizer with 200 epochs. The YOLO26n model achieved an overall precision of 0.827, recall of 0.734, and mAP50 of 0.711, with an inference speed of 8.1 ms per image, demonstrating its potential as a fast and lightweight solution for automated mold inspection, though further improvement in moldy bread detection performance is required before reliable deployment in bakery production lines.