Eucalyptus Species Identification: Utilizing Optimized VGG Architectures

Although existing studies have explored the identification of Eucalyptus species, the image datasets specific to this domain are typically limited in size, and there is still considerable room for improvement in the accuracy of current recognition algorithms. In response, this work introduces an enhanced VGG neural network model that incorporates residual learning, dropout regularization, and several other incremental enhancements such as small kernel layers, green channel enhancement, and attention mechanisms to improve feature extraction capabilities. Utilizing web scraping techniques, a substantial dataset of Eucalyptus tree images across various scales was collected and preprocessed for training the proposed model. Experimental results demonstrate a significant improvement of 14.78% in validation accuracy compared to baseline model, effectively enhancing the identification of Eucalyptus species. This breakthrough significantly contributes to the conservation of Eucalyptus diversity.