Lightweight Attention-Enhanced Deep Learning for Flower Species Recognition

Abstract

Accurate image-based flower species recognition has gained increasing attention due to its applications in biodiversity monitoring, agricultural management, and educational tools. Traditional computer vision approaches rely on handcrafted features and shallow classifiers but fail to generalize across diverse species and complex real-world conditions. With the advent of deep learning, convolutional neural networks (CNNs) and transformer-based models have significantly improved image recognition accuracy. This paper proposes a hybrid deep learning architecture for flower classification that combines a CNN backbone with attention-based feature refinement and transfer learning from large-scale natural image datasets. To address limited labeled data, we integrate data augmentation and knowledge distillation to reduce overfitting and computational cost. Extensive experiments on the Oxford 102 Flowers and FGVC datasets demonstrate that our approach outperforms conventional CNN baselines and achieves competitive results compared to recent vision transformers. Furthermore, we analyze model interpretability using Grad-CAM visualizations to highlight discriminative regions in flower images. This study provides a practical and efficient solution for real-world flower recognition, with potential applications in smartphone-based plant identification and smart agriculture.