Deep learning-based detection of tropical candida in Saccharomyces cerevisiae production

• 1、Biotechnology School, Jiangnan University,wuxi 214122
• 2、Internet of Things Engineering School, Jiangnan University,wuxi 214122

Abstract：ISaccharomyces cerevisiae holds significant value in the fermentation industry, yet its fermentation process and yield are susceptible to contamination by Candida tropicalis. This study integrated microscopic imaging with object detection technology to analyze physiological parameter variations of S. cerevisiae under different C. tropicalis contamination levels, including growth dynamics, budding rate, flocculation capacity, and morphological characteristics. An enhanced YOLOv10-based object detection model was constructed to detect morphological changes in both yeast species. Experimental results demonstrated that the model accurately identified contaminant growth phases and contamination severity, achieving high mean accuracy with outstanding performance in precision, recall, and mean Average Precision (mAP). Featuring compact parameters and low computational complexity, the model is deployable on edge devices. This approach provides an effective solution for real-time contamination monitoring and quality control in S. cerevisiae fermentation processes, while also establishing theoretical and experimental foundations for investigating yeast-yeast interactions.

Keywords： Image recognition Saccharomyces cerevisiae；Candida tropicalis；Deep Learning；YOLOv10；Cellular Morphology