Automated Fig (Ficus carica L.) variety classification based on deep learning using fruit and leaf images

Accurate identification of fig varieties is essential for orchard management, conservation of phytogenetic resources, quality control, and commercial traceability along the entire agri-food supply chain. Conventional identification methods based on morphological, biochemical, and molecular analyses are often limited by their subjectivity, destructive procedures, high costs, and poor scalability. As a consequence, their use is extremely restricted when applied to real-time systems or large-scale datasets. In recent years, computer vision and deep learning approaches have shown remarkable success in agricultural image analysis. However, their deployment for fig variety classification remains relatively underexplored, particularly under real-world conditions involving visually similar cultivars. This study proposes a multi-view deep learning framework for fig variety classification that exploits complementary visual information captured at both fruit and plant levels. Multiple visual representations, including external fig appearance, internal (cut) view, ostiole region, and leaf morphology, are considered to comprehensively characterize each cultivar. After preprocessing, deep features are independently extracted from each view using pre-trained convolutional neural networks through transfer learning, and subsequently integrated via feature-level fusion to construct a unified and discriminative representation. The resulting fused feature vector is classified using a Support Vector Machine (SVM) to enhance robustness against inter-class similarity and intra-class variability. Extensive experiments are conducted on a locally collected multi-view fig dataset from northern Algeria to evaluate the effectiveness of the proposed framework. Results demonstrate that multi-view fusion outperforms single-view configurations, enhancing accuracy and generalization. This approach shows high potential for smart agriculture, automated sorting, and decision-support tools in the agri-food industry.