Multilabel Classification of Orchid Features based on Deep Learning

There are many different types of orchids, with many inter-class similarities. Classifying them using the conven-tional method of comparing the features to registered orchid types is very time consuming. Thus it would be useful to have an image based orchid feature classifier. In this paper a single output multilabel classifier using transfer learning is designed to classify six different orchid features. The design is based on experiments on different loss functions, pre-trained models, number of neurons in the dense layer, dropout rates and fine-tuning. The final model uses an Xception model with one untrainable layer as feature extractor. The classifier consists of two dense layers with a dropout of 0.5 in between. The final model gets a macro average f1-score of 0.85. The model is reliable for the non-color features, however it doesn’t perform well on some rare classes of the color features.