Plant phenotyping with limited annotation: Doing more with less

Abstract

Deep learning (DL) methods have transformed the way we extract plant traits—both under laboratory as well as field conditions. Evidence suggests that “well‐trained” DL models can significantly simplify and accelerate trait extraction as well as expand the suite of extractable traits. Training a DL model typically requires the availability of copious amounts of annotated data; however, creating large‐scale annotated dataset requires nontrivial efforts, time, and resources. This limitation has become a major bottleneck in deploying DL tools in practice. Self‐supervised learning (SSL) methods give exciting solution to this problem, as these methods use unlabeled data to produce pretrained models for subsequent fine‐tuning on labeled data and have demonstrated superior transfer learning performance on down‐stream classification tasks. We investigated the application of SSL methods for plant stress classification using few labels. We select a plant stress classification problem to test the effectiveness of SSL, as it is a fundamentally challenging problem due to (a) disease classification which depends on the abnormalities in a small number of pixels, (b) high data imbalance across different classes, and (c) fewer annotated and available plant stress images than in other domains. We compared seven SSL approaches spanning four broad classes of SSL methods on soybean [Glycine max L. (Merr.)] plant stress dataset and report that pretraining on unlabeled plant stress images significantly outperforms transfer learning methods using random initialization for plant stress classification. In summary, SSL‐based model initialization and data curation improves annotation efficiency for plant stress classification tasks and will circumvent data annotation challenges associated with DL methods.

Self-supervised maize kernel classification and segmentation for embryo identification

“…Finally, we expect that such methods to be easily applied to, and very useful to a broad range of plant phenotyping applications. Recent examples of successful applications of such SSL training strategies include disease classification (Nagasubramanian et al, 2022) and insect detection (Kar et al, 2021).…”

“…Data annotation by an expert with domainspecific knowledge is a tedious and expensive task. The DL community is exploring various strategies to break this dependency on a large quantity of annotated data to train DL models in a label-efficient manner, including approaches like active learning (Nagasubramanian et al, 2021), transfer learning (Jiang and Li, 2020), weakly supervised learning (Ghosal et al, 2019;Körschens et al, 2021) and the more recent advances in selfsupervised learning (Jing and Tian, 2020;Marin Zapata et al, 2021;Nagasubramanian et al, 2022). Transfer learning has been widely utilized in plant phenomics applications for classification and segmentation tasks (Wang et al, 2019;Kattenborn et al, 2021).…”

“…Transfer learning has been widely utilized in plant phenomics applications for classification and segmentation tasks (Wang et al, 2019;Kattenborn et al, 2021). Recently, self-supervised learning has been applied to improve classification and segmentation models (Güldenring and Nalpantidis, 2021;Nagasubramanian et al, 2022;Lin et al, 2023). In this work, we focus on deploying self-supervised learning approaches to the problem of characterizing maize kernels that are imaged in a commercial high-throughput seed imaging system [Qsorter technologies (QualySense)].…”

Using Machine Learning Enabled Phenotyping To Characterize Nodulation In Three Early Vegetative Stages In Soybean

“…Combining advances in phenotyping as presented here with improved methods of genomic prediction will further enable breeding advancements related to nodulation (de Azevedo Peixoto et al, 2017; J. M. Shook, Lourenco, et al, 2021). Work could also be done using a similar method to evaluate other pulse crops' ability to nodulate differentially in taproot and non-taproot zones , potentially using less hands-on annotation (Nagasubramanian et al, 2022). We also foresee and recommend that farmers utilize imaging-based tools such as SNAP, if they are packaged in a smartphone app, allowing them to study nodulation and further work on root health for system-wide analysis and farm-based applications.…”

“…Manual measurements and counts are inefficient due to the complexity of the number, size, and placement of nodules on roots. Therefore, automated methods, including computer vision and machine learning-based models, are valuable avenues to solve these challenges (Kar et al, 2022;Mochida et al, 2019;Nagasubramanian et al, 2022;.…”