Root phenotyping is crucial for advancing our understanding of plant development and adaptation. However, existing platforms often face challenges in balancing high-throughput capacity with long-term, high-frequency monitoring. To overcome this limitation, we present HTPRootSlides, an integrated root phenotyping platform designed for dynamic and scalable trait analysis. Its design features a circulating zone that accommodates 141 specialized root boxes for high-throughput operation synchronously. Root boxes follow a continuous S-shaped trajectory step by step, facilitating repetitive imaging for high-throughput, time-series data acquisition. To address challenges such as water vapor condensation and fine root entanglement, we developed a dedicated segmentation algorithm, achieving 89.56% accuracy in root isolation. Combining morphological and skeleton-based feature extraction techniques, the platform ensures comprehensive and efficient phenotypic trait quantification. We validated HTPRootSlides by dynamically monitoring root development in four staple crops (soybean, maize, wheat, and rice) during early-stage germination (< 14 d). The results demonstrate the capability of HTPRootSlides for high-frequency, high-precision and large-scale root phenotyping (< 1 h with 141 root boxes per run), offering researchers a powerful tool to investigate root dynamics and optimize crop performance through trait selection.

For fast in-situ assessment of tiller phenotypes in rice breeding, we introduce the TillerPET model, an improved transformer-based deep learning solution that permits phenotyping the number and compactness of rice tillers in images of post-harvest rice stubble. A rice tiller phenotype dataset covering three years of field data and four experimental sites across China was constructed to train and validate the model. TillerPET reports an R² of 0.941 for counting tiller number, demonstrating state-of-the-art performance on the proposed RTP dataset. Beyond its minimal errors in estimating tiller number, TillerPET also achieves an R² of 0.978 for characterizing tiller compactness. The two phenotypic parameters exhibit a high degree of consistency with expert breeders, offering reliable phenotypic indicators to guide further breeding.

We present Depth-Informed Crop Segmentation (DepthCropSeg), an almost unsupervised crop segmentation approach without manual pixel-level annotations. Crop segmentation is a fundamental vision task in agriculture, which benefits a number of downstream applications such as crop growth monitoring and yield estimation. Over the past decade, image-based crop segmentation approaches have shifted from classic color-based paradigms to recent deep learning-based ones. The latter, however, rely heavily on large amounts of data with high-quality manual annotation such that considerable human labor and time are spent. In this work, we leverage Depth Anything V2, a vision foundation model, to produce high-quality pseudo crop masks for training segmentation models. We compile a dataset of 17,199 images from six public plant segmentation sources, generating pseudo masks from depth maps after normalization and thresholding. After a coarse-to-fine manual screening, 1378 images with reliable masks are selected. We compare four semantic segmentation models and enhance the top-performing one with depth-informed two-stage self-training and depth-informed post-processing. To evaluate the feasibility and robustness of DepthCropSeg, we benchmark the segmentation performance on 10 public crop segmentation testing sets and a self-collect dataset covering in-field, laboratory, and unmanned aerial vehicle (UAV) scenarios. Experimental results show that our DepthCropSeg approach can achieve crop segmentation performance comparable to the fully supervised model trained with manually annotated data (86.91 vs. 87.10). For the first time, we demonstrate almost unsupervised, close-to-full-supervision crop segmentation successfully.

As one of the most widely grown crops in the world, rice is not only a staple food but also a source of calorie intake for more than half of the world’s population, occupying an important position in China’s agricultural production. Thus, determining the inner potential connections between the genetic mechanisms and phenotypes of rice using dynamic analyses with high-throughput, nondestructive, and accurate methods based on high-throughput crop phenotyping facilities associated with rice genetics and breeding research is of vital importance. In this work, we developed a strategy for acquiring and analyzing 58 image-based traits (i-traits) during the whole growth period of rice. Up to 84.8% of the phenotypic variance of the rice yield could be explained by these i-traits. A total of 285 putative quantitative trait loci (QTLs) were detected for the i-traits, and principal components analysis was applied on the basis of the i-traits in the temporal and organ dimensions, in combination with a genome-wide association study that also isolated QTLs. Moreover, the differences among the different population structures and breeding regions of rice with regard to its phenotypic traits demonstrated good environmental adaptability, and the crop growth and development model also showed high inosculation in terms of the breeding-region latitude. In summary, the strategy developed here for the acquisition and analysis of image-based rice phenomes can provide a new approach and a different thinking direction for the extraction and analysis of crop phenotypes across the whole growth period and can thus be useful for future genetic improvements in rice.