CNN with Multi Stage Image Data Augmentation Methods for Indonesia Rare and Protected Orchids Classification

Image classification of Indonesian rare and protected orchids is one of the solutions to prevent illegal trade, especially in online commerce that uses images as one of the display features. Image classification using deep Convolutional Neural Network (CNN) is a major breakthrough at this time where the extraction feature is done automatically through a series of convolution layers. Deep CNN requires a lot of training data to produce a good classification result. Image data of Indonesian rare and protected orchids are relativHeruely difficult to obtain when searched through relevant sources to avoid inaccurate information. In this paper, we propose a new approach of multi-stage image data augmentation to overcome limited data problem in deep CNN for Indonesian rare and protected orchids image classification. Image data augmentation method using basic image augmentation divided into geometric transformation and distortion injection. ResNet with transfer learning as CNN model is used for the classification. The proposed system is experimentally evaluated in the form of data augmentation methods combination such as stage 1 and stage 2 augmented dataset and results show its convincing performance compared to existing methods.