Share
Export Citation
SE-ResNet: An Attention Enhanced CNN for Classification of Indonesian Medicinal Plants
Rahmat M.A.
Proceedings of the International Conference on Information Technology and Electrical Engineering Icitee
Abstract
Correctly identifying medicinal plants is very important for protecting biodiversity, supporting pharmaceutical research, and keeping traditional knowledge alive, especially in areas with a lot of biodiversity like Indonesia. This study suggests a new version of the SE-ResNet-50 architecture for classifying Indonesian medicinal plants by their leaf images. The model uses a pretrained ResNet-50 on ImageNet to extract hierarchical features. A Squeeze and Excitation (SE) block is added at the last convolutional layer to make high level discriminative features stand out more. We used a set of $\mathbf{1, 5 0 0}$ pictures of plants in ten different categories to train and test the model. The experimental results show that the proposed model has a classification accuracy of 97.65%, which is much better than a baseline ResNet- 50 model without pretrained weights, which only got 89.94%. These results show that using residual learning, attention mechanisms, and transfer learning together works well for very detailed plant classification tasks.