StomataSeg: Semi-Supervised Instance Segmentation for Sorghum Stomatal Components

TL;DR

This work tackles automated, multi-class segmentation of sorghum stomatal components, a challenging task due to tiny, nested pore and guard-cell structures. It introduces StomataSeg, a patch-based semi-supervised framework that combines high-resolution patch preprocessing with pseudo-labeling to expand the training set, coupled with a dedicated stomatal dataset comprising 11,060 manual and 56,428 pseudo-labelled patches across multiple genotypes. Benchmarking across semantic and instance segmentation models shows notable gains: semantic $\text{mIoU}$ improves from $65.93\%$ to $70.35\%$, and instance $\text{AP}$ from $28.30\%$ to $46.10\%$, with pore area $\text{AP}$ rising to $35.40\%$, demonstrating robust detection of small stomatal features. The approach enables scalable, AI-driven stomatal phenotyping in crops, providing a practical path toward higher-throughput analysis of traits linked to water-use efficiency and drought resilience, while acknowledging limitations such as species scope and potential pseudo-label noise.

Abstract

Sorghum is a globally important cereal grown widely in water-limited and stress-prone regions. Its strong drought tolerance makes it a priority crop for climate-resilient agriculture. Improving water-use efficiency in sorghum requires precise characterisation of stomatal traits, as stomata control of gas exchange, transpiration and photosynthesis have a major influence on crop performance. Automated analysis of sorghum stomata is difficult because the stomata are small (often less than 40 $μ$m in length in grasses such as sorghum) and vary in shape across genotypes and leaf surfaces. Automated segmentation contributes to high-throughput stomatal phenotyping, yet current methods still face challenges related to nested small structures and annotation bottlenecks. In this paper, we propose a semi-supervised instance segmentation framework tailored for analysis of sorghum stomatal components. We collect and annotate a sorghum leaf imagery dataset containing 11,060 human-annotated patches, covering the three stomatal components (pore, guard cell and complex area) across multiple genotypes and leaf surfaces. To improve the detection of tiny structures, we split high-resolution microscopy images into overlapping small patches. We then apply a pseudo-labelling strategy to unannotated images, producing an additional 56,428 pseudo-labelled patches. Benchmarking across semantic and instance segmentation models shows substantial performance gains: for semantic models the top mIoU increases from 65.93% to 70.35%, whereas for instance models the top AP rises from 28.30% to 46.10%. These results demonstrate that combining patch-based preprocessing with semi-supervised learning significantly improves the segmentation of fine stomatal structures. The proposed framework supports scalable extraction of stomatal traits and facilitates broader adoption of AI-driven phenotyping in crop science.

Figures (11)

• Figure 1: Closed stomata show no visible pore; open stomata exhibit a distinct pore.
• Figure 2: Digital microscopy was performed inside a solarweave greenhouse.
• Figure 3: Distribution of the 318 human-annotated original images across genotype, leaf level, surface and region dimensions. Each point denotes a specific genotype–leaf-level combination; colour (green = abaxial, blue = adaxial) indicates leaf surface, shape (circle = base, square = mid, triangle = tip) indicates sampling region, and the number shows image count. Background bands separate the five genotypes. The y-axis tracks developmental progression from L9 (lower leaves) to FL (flag leaf, uppermost fully-emerged leaf).
• Figure 4: Annotation interface on the V7 platform showing pixel-wise labelling of a sorghum image. Each stomatal complex is delineated, with inner guard cell and pore areas annotated as distinct objects. Annotation classes are colour-coded and listed in the sidebar to support real-time collaboration, instance management, and quality monitoring. The interface also displays metadata and annotation progress to ensure consistent and efficient data curation.
• Figure 5: Dataset curation and expansion workflow. The pipeline comprises: (1) image acquisition was performed inside a solarweave greenhouse with microscopy; (2) quality filtering of the 318 original images ensuring balanced genotype, surface and region coverage; (3) manual pixel-wise annotation by trained annotators with expert review; and (4) pseudo-labelling expansion using only high-confidence predictions from the seed segmentation model.