Plant Growth Prediction Based on Hierarchical Auto-encoder

As plants grow, the area of the leaves changes arbitrarily and the growth rate varies from leaf to leaf. In controlled environments such as plant factories, accurate plant growth prediction models are required for efficient cultivation. In this paper, we propose a new deep learning network that can predict plant growth. First, for predicting the shape of a plant, hierarchical auto-encoders are adopted for shape prediction. After the plant shape is predicted first, its RGB information is replenished by fusing the shape with a current RGB image to generate a future RGB plant image. A variety of experiments have been performed with a dataset produced from a plant factory. Experimental results show that the proposed method is resistant to predicting global and local growth of plant leaves. It also predicts dynamic plant movements well, leading to the accurate prediction of a future plant image.