A survey on deep-learning-based plant phenotype research in agriculture

Plant phenotype refers to measurable traits of plants, which acts as an observable proxy between gene expression and environmental impact, and is also an important determinant for the yield, quality and stress resistance characteristics of crops. Most of the plant phenotypes can be acquired by digital imaging techniques and processed by image processing algorithms. In recent years, the rapid development of genomics advances the study of plant phenotyping in many aspects, especially in terms of high-precision and high-throughput. Traditional plant phenotype research cannot meet these requirements and revolutions are in urgent need. As a breakthrough in computer science, the emergence of deep learning approaches significantly expands the capability of traditional image processing. For instance, state-of-the-art results in identification and segmentation tasks have been achieved by deep-learning-based methods and the records are continually improved by their variants. It is an interesting topic to study how to incorporate deep-learning techniques into plant phenotype research, and various impactful methods have been proposed in the past few years. The objective of this survey is to provide an overview of the current progress of deep-learning-based plant phenotype research in agriculture. In this survey, we elaborate the work from four different aspects, (i) plant morphology and physiological information extraction, (ii) plant identification and weed detection, (iii) pest detection, and (iv) yield prediction. We also analyze the pros and cons of these methods compared to traditional approaches. The potential future trends of plant phenotyping research are discussed at the end of this survey.