PhytoNet: Mish-Optimized Deep Learning Architecture for Enhanced Tomato Leaf Disease Detection

Tomato leaf diseases significantly impact agricultural productivity, necessitating accurate and efficient diagnostic methods. Deep learning has emerged as a robust approach for plant disease detection, but challenges such as inefficient feature extraction, classification, model complexity and limited computational resources hinder its widespread adoption. This study introduces a PhytoNet (Mish-Optimized SqueezeNet) Framework to enhance tomato leaf disease prediction. The SqueezeNet architecture, known for its lightweight design, is optimized with the Mish activation function to improve feature extraction and classification capabilities while maintaining computational efficiency. The methodology involves training the SqueezeNet model on 10 classes of mendaley dataset of tomato leaf images, encompassing multiple disease classes and healthy samples. Data preprocessing techniques, including image augmentation and normalization, are employed to ensure model robustness. The integration of the Mish activation function in critical layers enhances non-linearity, aiding in better gradient flow and improved performance during training. Model evaluation is conducted using metrics such as accuracy, precision, recall and F1-score. Experimental results demonstrate that the PhytoNet outperforms traditional SqueezeNet and other lightweight architectures in terms of classification accuracy, achieving over 0.9957 accuracy on the dataset. Additionally, the model maintains low computational overhead, making it suitable for deployment on resource-constrained devices. Hence, the proposed framework effectively balances accuracy and efficiency, addressing critical limitations in existing plant disease detection models. This work underscores the potential of lightweight and activation-optimized deep learning frameworks for real-time agricultural applications, paving the way for scalable and sustainable solutions in precision farming.