RCSnet——Flower Classification Network Design Based on Transfer Learning and Channel Attention Mechanism

The current global landscape contains a vast array of flowering plant species. Given the limitations of small-scale datasets and the complexity of real-world photography, developing an efficient and accurate flower classification system is a significant challenge. This study introduces a novel flower classification network architecture, RCS-net, which uses transfer learning and channel attention mechanisms to improve the accuracy and efficiency of flower image classification. RCS-net is based on ResNet50, which was pre-trained on the ImageNet dataset. By incorporating channel attention mechanisms, it enhances the convolutional neural network’s ability to capture key features within flower images, thereby improving classification performance. Experimental validation on a publicly available flower dataset on the Kaggle platform shows that RCS-net achieves an accuracy rate of 96.1% on a three-class flower dataset, outperforming currently popular flower classification models. In addition, ablation studies verify the effectiveness of combining channel attention mechanisms with convolutional neural networks, proving that RCS-net not only improves classification accuracy, but also optimises computational efficiency and model robustness. Future work will focus on extending the scope of the model, exploring optimisation strategies and improving generalisation capabilities to support a wider range of plant classification tasks, considering its potential application in resource-constrained environments.