Microscopy-based Fungi Species Classification using a Deep Learning Ensemble Approach

Fungi play a vital role in many industries such as medicine and agriculture, but they also cause infections and diseases, making accurate identification crucial. This study presents an ensemble deep learning model using CNNs, integrated into a system called MycoDiagnosis, to categorize microscopic images of fungal species. Three models—MobileNetV2, InceptionV3, and EfficientNetB0—are combined in an ensemble, where their outputs are combined to improve accuracy and efficiency in identifying fungal species. Additionally, image preprocessing techniques were applied to improve model training performance. The ensemble model effectively reduces human error in fungal classification while providing more accurate and consistent diagnoses. This improved identification is valuable in the study of epidemiology and disease prevention. The model achieved an accuracy of 88%, with a precision of 88.13%, a recall of 88.00%, and an F1 score of 87.92%. Additionally, the model’s performance was compared against the individual CNN models to determine the effectiveness of the ensemble strategy. These results demonstrate that CNN-based ensemble methods, integrated within the MycoDiagnosis system, can significantly enhance fungal diagnostics through advanced microscopy and computational techniques.