PunicaNet: A shortened-ResNet18 and transformer hybrid model for pomegranate disease detection
Pomegranate ( Punica granatum ) is a widely consumed fruit with high commercial value, making its cultivation important. Image-based pomegranate disease detection is essential for autonomous systems that support efficient value chain management. Literature survey reveals that exploration of Transformer based deep learning models in computer vision domain is an active area of research due to inherent low prediction accuracy of the Transformer model in its stand alone form for image processing applications. To address this research gap, the objectives of the proposal are defined as follows (i) image based identification of pomegranate fruit diseases using deep learning (ii) designing Transformer based deep learning model suitable for pomegranate fruit disease detection (iii) comparing the proposed model with existing mainstream models for performance appraisal. The proposed PunicaNet is a hybrid model involving shortened-ResNet18 and Transformer, trained to identify and classify pomegranate disease labels. The design uniqueness of PunicaNet is that it provides a design template for integrating Convolutional Neural Network (CNN) with Transformer as hybrid models. CNN captures local features and Transformer captures global image features. The proposed model is evaluated for Precision, Recall, F1-Score, overall accuracy, labelwise accuracy and has achieved performance improvements of approximately 0.26–3.50% compared to baseline models. PunicaNet presents a template for designing Transformer-CNN hybrid models for pomegranate fruit disease identification.
Authors
- Anupama V
- Shrisha H S
Publication Details
- Journal
- Journal of Integrative Horticulture
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1016/j.jih.2026.100010
- Primary Topic
- Smart Agriculture and AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00