Aspergillus Species Fungi Identification Using Microscopic Scale Images
Robert Kerwin C. Billones, Edwin J. Calilung, Elmer P. Dadios, Nelson Santiago Vispo
Abstract
This paper presents a technique for microscopic classification of Aspergillus fungi species. In another study, the authors described a process of fungi identification using macroscopic images. However, some fungi macroscopic images look similar even if it belongs to different Aspergillus fungi species. The process described in this study includes the microscopic identification and classification of a 9-type Aspergillus fungi species. A machine learning model is trained and validated using 4545 microscopic images. The CNN model v1 achieved 87.50% accuracy in training, and 95.65% accuracy in validation. The model is re-calibrated to improve the training performance. CNN model v1.1 used softmax instead of sigmoid activation, and dropout rate is reduced from 0.5 to 0.2. The re-calibrated model achieved 94.20% accuracy in training, and 94.31% accuracy in validation.