MobileNetV3Small-CM FusionNet: A Lightweight Deep Learning Framework for Multi-Class Arecanut Disease Classification Using Feature Fusion

Abstract

This study presents a novel deep learning-based framework, MobileNetV3Small-CM FusionNet, for the multi-class classification of arecanut plant diseases. The proposed model combines lightweight convolutional features extracted from the MobileNetV3Small architecture with handcrafted color moment descriptors (mean, standard deviation, and skewness) to enhance classification accuracy, particularly for visually similar and minority classes. A comprehensive dataset containing images of healthy and diseased arecanut samples was used for training, validation, and testing. The model was benchmarked against a baseline CNN and the unmodified MobileNetV3Small. Experimental results demonstrate that the proposed fusion model significantly outperforms the baselines, achieving a test accuracy of 99.54%, along with near-perfect precision, recall, and F1-scores across all nine disease classes. In contrast, the CNN and MobileNetV3Small achieved 93.74% and 83.87% accuracy, respectively, with notable misclassifications in rare disease categories. The superior performance of the proposed model validates the effectiveness of combining deep and handcrafted features for robust plant disease recognition.