Review: Deep learning in image-based plant phenotyping

As a writer and an editor, I am horrified by the idea that thinking can be replaced by artificial intelligence. But I do recognize that deep learning / machine learning / artificial intelligence can provide major opportunities for data analysis, as eloquently described in this review article by Murphy et al. that focuses on the applications of deep learning for plant image data and phenotyping. They start off with some clear definitions (including helpful diagrams) of the terminology employed, whether about computer vision and image segmentation or different types of deep learning and neural networks. They next discuss best practices, emphasizing the importance of carefully selecting the training and validation datasets, and define the different metrics that can be used to assess the success of the deep learning model. They wrap up with some examples of how deep learning can be used in plant classification (including weed versus non-weed) and phenomics, including early detection of stress or disease, and a call for more training in these methods. Like it or not, machine learning is part of our lives now, and this review is an excellent way to start to appreciate how it can make our science more efficient and effective. (Summary by Mary Williams @PlantTeaching) Annu. Rev. Plant Biol. 10.1146/annurev-arplant-070523-042828