PDD-Net: Plant Disease Diagnoses Using Multilevel and Multiscale Convolutional Neural Network Features

Overlooked diseases in agriculture severely impact crop growth, which results significant losses for farmers. Unfortunately, manual field visits plant disease diagnosis (PDD) are costly and time consuming. Although various methods of PDD have been proposed, many challenges yet to be investigated, such as early stage leaf diagnosis, class variations diseases, cluttered backgrounds, computational complexity the system. In this paper, we propose a Convolutional Neural Network (CNN)-based framework (i.e., PDD-Net), employs data augmentation techniques incorporates multilevel multiscale features create scale-invariant architecture. The Flatten-T Swish (FTS) activation function is utilized prevent gradient vanishing exploding problems, while focal loss used mitigate imbalance during PDD-Net training. method outperforms baseline models, achieving an average precision 92.06%, recall 92.71%, F1 score 92.36%, accuracy 93.79% on PlantVillage dataset. It also achieves 86.41%, 85.77%, 86.02%, 86.98% cassava These demonstrate efficiency robustness diagnosis.