ResNet18-SaakNet: A Real-Time Deep Learning Model for Plant Disease Detection

• Karan Soni,
• Rakesh Chandra Gangwar,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.003, Download Paper (Free):

• Deep Learning
• Leaf disease
• Plant disease classification
• ResNet18
• Saak Transform

Abstract

Classification of leaf diseases is a challenging problem because of small inter-class differences, varied environmental conditions, and visual similarities between disease symptoms. In this paper, introduced a light-weight and explainable deep learning model, ResNet18-SaakNet, which integrates Saak (Subspace approximation with augmented kernels) Transform-based pre-processing with ResNet18 as the backbone to learn discriminative patterns in leaf images successfully. The data set utilized pre-processed images of plant leaves from nine different disease classes, selected from the PlantVillage repository. This work introduces a novel hybrid model, ResNet18-SaakNet, which integrates handcrafted spatial-frequency features from the Saak Transform with the deep hierarchical learning capability of ResNet18. Unlike conventional CNNs, our approach ensures real-time applicability with low computational overhead while maintaining high accuracy (96.12% on test data). Comparison with light-weight CNNs also defines its efficiency and interpretability. Saak Transform maintains high-frequency spatial-detail information and enhances feature separability, and ResNet18, light-weight convolutional neural network, achieves significant visual representations. The model reaches 100% training and 96.8% validation accuracy and demonstrates better classification capability with low computational resources. Further, the combining of spectral characteristics through Saak Transform facilitates the assistance of enhancing overall generalization and avoidance of overfitting within deep convolutional layers. These findings validate that ResNet18-SaakNet is a suitable real-time deployable-scalable tool for diagnosing leaf diseases that can be capable of keeping up with precision agriculture and allow growers to diagnose and react to leaf diseases in time.

Statistics

Cite this article