Hassan, H., Amir, A., Abd El-Ghany, M., Salih, S., Ouf, S. (2025). Aspergillus detection based on deep learning model using YOLOv8 with a small custom dataset.. Egyptian Journal of Botany, 65(2), 211-226. doi: 10.21608/ejbo.2025.342052.3109
Hosam M. Hassan; Asmaa Amir; Mohamed Naguib Mohamed Abd El-Ghany; Said A. Salih; Salama A. Ouf. "Aspergillus detection based on deep learning model using YOLOv8 with a small custom dataset.". Egyptian Journal of Botany, 65, 2, 2025, 211-226. doi: 10.21608/ejbo.2025.342052.3109
Hassan, H., Amir, A., Abd El-Ghany, M., Salih, S., Ouf, S. (2025). 'Aspergillus detection based on deep learning model using YOLOv8 with a small custom dataset.', Egyptian Journal of Botany, 65(2), pp. 211-226. doi: 10.21608/ejbo.2025.342052.3109
Hassan, H., Amir, A., Abd El-Ghany, M., Salih, S., Ouf, S. Aspergillus detection based on deep learning model using YOLOv8 with a small custom dataset.. Egyptian Journal of Botany, 2025; 65(2): 211-226. doi: 10.21608/ejbo.2025.342052.3109
Aspergillus detection based on deep learning model using YOLOv8 with a small custom dataset.
1Department of Mathematics, Faculty of Science, Cairo University, Giza 12613, Egypt
2Department of Biotechnology, Faculty of Science, Cairo University, Giza 12613, Egypt.
3Botany and Microbiology Department, Faculty of Science, Cairo University, Giza, Egypt
4Chemistry department, Faculty of Science, Cairo university
Abstract
Over the past years, there has been a growing interest in studying the effects of fungal respiratory diseases by the predominant species identified in respiratory cultures from this genus Aspergillus. Machine learning autonomously identifies the five distinct species of fungus. We selected a diverse array to show a wide array of color combinations, dimensions, and configurations, which enhance the incorporation of diversity and intricacy in our research. The split was conducted in a random manner, allocating 70% of the data to the training set, 20% to the validation set, and 10% to the test set. The photos assessed the heterogeneity among various forms of Aspergillus. The photographs were taken against two distinct backgrounds: one in copper and the other in grey. Multiple elevations and shooting angles were taken into consideration. The crowdedness of the Aspergillus also varied randomly per image. We utilized a smartphone camera boasting a resolution of 32 megapixels. A grand total of 337 photographs were captured, including 5 objects that were appropriately identified. CSPDarknet53 acts as the fundamental structure for YOLOv8, which is constructed on top of DenseNet. The YOLOv8 model attained a mean average precision (mAP) of 90%. YOLOv8 has a significant advantage in terms of its speed in detecting objects, making it suitable for real-time identification situations that demand both high accuracy and few false positives. The results exhibited that YOLOv8 exhibited outstanding precision and detecting skills. This technique is highly effective and efficient in detecting many species of Aspergillus.