Plant disease detection on PlantVillage dataset using EfficientNetV2-B0.
In this code, I focused on only 7 labels of this dataset. First PlantVillage dataset has been downloaded and loaded. Then these 7 labels has been seperated for preprocessing step. Processing steps are resizing and augmentation. After preprocessing, I used EfficientNetV2B0 with pre-traind weights of Imagenet dataset. After fine-tuning, for a better analysis, Macro, Micro, and Weighted has been calculated for Precision, Recall, and F1-score. Moreover, the confusion matrix has been shown. At last, EfficientNetV2B0 could achieve around 98% accuracy on these 7 classes.
Note that this dataset is imbalanced, and for better performance in the real world, we need to solve this problem. You can utilize the method that I used in this link for balancing ckplus dataset: https://github.com/OmidGhadami95/EfficientNetV2_Quantization_CK/blob/main/EfficientNetV2_B2_Quantization_CK%2B.ipynb