Skip to main content

Accelerating image-based plant phenotyping and pattern recognition: deep learning or few-shot learning?

This thematic series is open for submissions. For more information, click here.
To submit your manuscript, click here.

In recent years, deep learning methods have played a great role in the plant sciences and achieved a series of remarkable achievements in many fields, such as yield prediction and estimation, crop pest identification, plant disease detection, physiological trait indication, seedling development monitoring, plant irrigation strategy, cultivar recognition, leaf counting, etc. However, the applications based on typical deep learning rely heavily on big-scale datasets requiring substantial manual annotation of training data, which is a serious shortcoming. After all, large-scale real-world agricultural datasets are time-consuming and expensive to collect and label by experts for every potential application. In order to alleviate this problem, few-shot learning is emerging, also called learning from few data. Few-shot learning is a new branch of deep learning, which aims to develop an intelligent model with good generalization from only few data, towards the combination of machine intelligence with flexibility and extensibility. Both deep learning and few-shot learning are technological explorations in the field of plant sciences that have the potential to greatly accelerate their applications. Therefore, we encourage you to share your research and opinions in this area and submit your paper to this thematic series. 

This page was last updated in May 2021.