Philippine Indigenous Plant Seedlings Classification Using Deep Learning

Plant taxonomists have specialized knowledge on a specific plant species. The shortage of these experts and their unbalanced distribution throughout the globe remains to be a serious problem worldwide. Deep learning technology can be used to create tools that will help them speed up the process of plant classification. This study implements three deep learning models to classify images of Philippine indigenous plant seedlings into five species. AlexNet, GoogLeNet, and ResNet50 were fine-tuned for this purpose. In the three pre-trained models, the weight and bias learning rate factors of the fully connected layers were both increased to 20 to speed up learning. Augmentation techniques such as rotation, random flip, and random horizontally and vertically translations were performed to training images to avoid the effects of overfitting. The resulting deep learning models can be implemented to classify Philippine indigenous plant seedlings as they all achieved above 90% accuracy on the validation set, with ResNet50 exhibited the highest with 98.93%.