Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 May 26:16:1589161.
doi: 10.3389/fpls.2025.1589161. eCollection 2025.

Rice-SVBDete: a detection algorithm for small vascular bundles in rice stem's cross-sections

Xiaoying Zhu  1 , Weiyu Zhou  1 , Jianguo Li  2   3 , Mingchong Yang  2   3 , Haiyu Zhou  2   3 , Jiada Huang  2   3 , Jiahua Shi  4 , Jun Shen  5 , Guangyao Pang  1 , Lingqiang Wang  2   3   6
Affiliations

Rice-SVBDete: a detection algorithm for small vascular bundles in rice stem's cross-sections

Xiaoying Zhu et al. Front Plant Sci. .

Abstract

Introduction: Vascular bundles play a vital role in the growth, development, and yield formation of rice. Accurate measurement of their structure and distribution is essential for improving rice breeding and cultivation strategies. However, the detection of small vascular bundles from cross-sectional images is challenging due to their tiny size and the noisy background typically present in microscopy images.

Methods: To address these challenges, we propose Rice-SVBDete, a specialized deep learning-based detection algorithm for small vascular bundles in rice stem cross-sections. Our approach enhances the YOLOv8 architecture by incorporating three key innovations: Dynamic Snake-shaped Convolution (DSConv) in the Backbone network to adaptively capture intricate structural details of small targets. A Multi-scale Feature Fusion (MFF) mechanism, combining features from the Backbone, Feature Pyramid Network (FPN), and Path Aggregation Network (PAN), to better handle objects at multiple scales. A new Powerful Intersection over Union (PIoU) loss function that emphasizes spatial consistency and positional accuracy, replacing the standard CIoU loss.

Results: Experimental evaluations show that Rice-SVBDete achieves a precision of 0.789, recall of 0.771, and mean Average Precision (mAP@.5) of 0.728 at an IoU threshold of 0.50. Compared to the baseline YOLOv8, Rice-SVBDete improves precision by 0.179, recall by 0.201, and mAP@.5 by 0.227, demonstrating its effectiveness in small object detection.

Discussion: These results highlight Rice-SVBDete's potential for accurately identifying small vascular bundles in complex backgrounds, providing a valuable tool for rice anatomical analysis and supporting advancements in precision agriculture and plant science research.

Keywords: YOLO; deep learning; deformable convolution; rice vascular bundles; small object detection.

PubMed Disclaimer

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Figures

Figure 1
Figure 1
Challenges in the detection of small vascular bundles.
Figure 2
Figure 2
Network structure of Rice-SVBDete. C2f-DSConv module in the Backbone network improves the model’s accuracy in recognizing the fine structures at the boundary of small vascular bundles. Multi-feature fusion mechanism of Neck network accurately captures fine-grained target features.
Figure 3
Figure 3
Rice stem cross-sections microscopic image.
Figure 4
Figure 4
Dataset annotation. (A) shows the overall annotation result, while (B, C) are zoomed-in views of specific regions in (A).
Figure 5
Figure 5
Detection results of State-of-the-Art Methods.
Figure 6
Figure 6
Comparison of manual labeling and algorithmic detection results. (a, b) show the fitting of area and count of small vascular bundles between the predictions and the labels, (c, d) present the least squares fitting results for these two parameters.
Figure 7
Figure 7
Heatmap examples without MFF and with MFF.
Figure 8
Figure 8
Visualization of detection results before and after applying DSConv and MFF modules.
See this image and copyright information in PMC

References

    1. Bapat V. A., Kavi Kishor P., Jalaja N., Jain S. M., Penna S. (2023). Plant cell cultures: biofactories for the production of bioactive compounds. Agronomy 13, 858. doi: 10.3390/agronomy13030858 - DOI
    1. Chen Z., Liu Z., Liu J. (2024. c). “Local contrast prior-guided cross aggregation model for effective infrared small target detection,” in ICASSP 2024–2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (Seoul, Korea: IEEE; ), 4315–4319.
    1. Chen J., Wen R., Ma L. (2024. a). Small object detection model for uav aerial image based on yolov7. Signal. Image Video Process. 18, 2695–2707. doi: 10.1007/s11760-023-02941-0 - DOI
    1. Chen T., Ye Z., Tan Z., Gong T., Wu Y., Chu Q., et al. . (2024. b). Mim-istd: Mamba-in-mamba for efficient infrared small target detection. IEEE Trans. Geosci. Remote Sens. 62, 1–13. doi: 10.1109/TGRS.2024.3485721 - DOI
    1. Chen D., Zhang L. (2024). SL-YOLO: A stronger and lighter drone target detection model. arXiv preprint arXiv:2411.11477. (Ithaca, NY, USA: Cornell University Library; ) doi: 10.48550/arXiv.2411.11477 - DOI