MBFFN: A Multi-Backbone Feature Fusion Network for Fine-Grained Plant Species Classification

Automatic plant species identification plays a vital role in agriculture, biodiversity conservation, and ecological monitoring. However, leaf-based recognition remains challenging due to high interclass similarity, intraclass variation, and complex visual patterns. To address these issues, this paper proposes a multi-backbone feature fusion network (MBFFN) for reliable plant species identification using leaf images. The proposed approach integrates complementary deep representations extracted from multiple convolutional neural network architectures, enabling effective modeling of both local texture information and global structural characteristics. The MBFFN framework fuses features from individual backbone models and employs a fully connected classifier for final prediction, improving robustness compared with single-model systems. This design is well suited for practical deployment where high accuracy and consistency are required. The effectiveness of the proposed method is evaluated on two widely used benchmark datasets, Flavia and Swedish Leaf, which present diverse and challenging classification scenarios. Experimental results conducted using a 70:15:15 training-validation-testing split demonstrate that the proposed MBFFN achieves a classification accuracy of 98.11% on the Flavia Leaf Dataset and 98.51% on the Swedish Leaf Dataset, outperforming individual deep learning models and several existing approaches. These results indicate that multi-backbone feature fusion provides a reliable and scalable solution for real-world plant species identification. The proposed MBFFN framework has strong potential for applications in automated plant analysis, precision agriculture, and ecological decision-support systems.