A Faster RCNN Architecture for Simultaneous Detection of Fruit type and Disease : A Multi-task Learning Approach

Abstract

Recent advances in computer vision have significantly impacted agriculture by enabling more precise monitoring of fruits. Deep learning techniques applied to fruit analysis enable the identification of fruit types and the detection of diseases, leading to improved yields and reduced environmental impact through timely intervention. However, most existing methods employ separate single-task models for each attribute (e.g., fruit classification or disease detection), resulting in increased complexity, higher computational cost, and the need for large labeled datasets for each task. In contrast, we propose a unified multi-task learning framework based on the Faster R-CNN architecture that simultaneously performs fruit type classification and disease detection within a single model. By sharing convolutional feature representations across both tasks, our approach leverages synergies between classification and detection, enhancing efficiency and accuracy. Training on a diverse dataset of fruit images annotated with both fruit identities and disease labels, our model jointly optimizes both tasks to learn more generalizable and discriminative features. Experimental evaluations demonstrate that this multi-task Faster R-CNN achieves competitive accuracy on both tasks while requiring fewer training samples and less computational resources than separate single-task models. The unified model not only simplifies system design but also accelerates inference and training, improving real-world scalability and robustness. This integrated approach provides a robust, efficient solution for automated fruit attribute analysis in agricultural applications.