Fruit Ripeness Classification with Few-Shot Learning

Deep learning based image classification systems require large amount of training data and long training time. However, the availability of large annotated image dataset is usually limited and expensive to generate, which limits a vision system to adapt to new task efficiently. In this paper, a few-shot classification framework is proposed which can adapt one fruit ripeness classification system to classify new types of fruits using only a few training samples. The proposed framework adopts the meta-learning paradigm where a base network learns to extract meta-features and few-shot classification tasks from the base classes with abundant training samples and then apply the network to similar task on the novel classes using only a few support samples. Experimental results indicate that the proposed framework is able to achieve over 75% ripeness classification accuracy on various fruits using a little as five samples.