Hybrid MobileNetV2 and Graph Convolutional Networks for Clove Leaf Nutrient Deficiency Classification

Abstract

Nutrient deficiency is one of the major factors affecting the productivity of clove (Syzygium aromaticum) plants. Conventional diagnostic methods, including laboratory analysis and visual inspection, are often time-consuming, destructive, or subjective. This study proposes a hybrid deep learning framework that integrates MobileNetV2 and Graph Convolutional Networks (GCNs) for automatic classification of nitrogen, phosphorus, and potassium (NPK) deficiencies in clove leaves. MobileNetV2 was employed as a feature extractor to generate visual embeddings, which were transformed into a graph representation using the K-Nearest Neighbors (KNN) algorithm. The resulting graph was then classified using a Graph Convolutional Network to exploit structural relationships among visually similar leaf samples. Experimental evaluation on four classes (Healthy, Nitrogen deficiency, Phosphorus deficiency, and Potassium deficiency) achieved an overall classification accuracy of 94.57%, with macro-average precision, recall, and F1-score of approximately 95%. These results indicate that integrating graph-based relational learning with convolutional feature extraction effectively improves nutrient deficiency classification in clove leaves and demonstrates the potential of the proposed framework for automated plant health monitoring in precision agriculture.   Keywords - Graph Convolutional Networks, MobileNetV2, NPK deficiency, Graph-based deep learning, Clove leaves.