Toward automatic plant phenotyping: starting from leaf counting

The development of automatic plant phenotyping systems has drawn great attention in the recent years. It can help improve the throughput of phenotyping measurements and reduce the associated human labor cost. Working towards the goal of automatic plant phenotyping, here we begin with developing an automatic method for leaf counting. Most of the previous approaches for leaf counting are based on regression modeling or instance segmentation. In contrast to these approaches, we consider the task of leaf counting as a object detection problem. In particular, we perform object detection and localization for leaves in the input images. The location and size of a leaf is indicated by a bounding box. Thus, we can obtain the number of leaves by counting the number of bounding boxes. We develop our leaf counting network architecture based on YOLOv3. In order to evaluate our proposed method, we utilize the cauliflower images from the ABRC (Agricultural Biotechnology Research Center, Academia Sinica) and the Arabidopsis images from the CVPPP (Computer Vision Problems in Plant Phenotyping) dataset. Our proposed method achieves state of the art results on these datasets.