Related Experiment Video
Updated: Feb 17, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Panicle-SEG: a robust image segmentation method for rice panicles in the field based on deep learning and superpixel
Xiong Xiong1, Lingfeng Duan2, Lingbo Liu1
1Britton Chance Center for Biomedical Photonics, Wuhan National Laboratory for Optoelectronics, and Key Laboratory of Ministry of Education for Biomedical Photonics, Department of Biomedical Engineering, Huazhong University of Science and Technology, Wuhan, 430074 People's Republic of China.
This study introduces Panicle-SEG, a novel algorithm for accurate rice panicle segmentation. This method enhances rice breeding by improving yield estimation through robust image analysis.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Breeding
Background:
- Accurate rice panicle segmentation is crucial for image-based phenotyping in rice breeding.
- Challenges include variable illumination, complex backgrounds, and diverse rice characteristics.
Purpose of the Study:
- To develop a robust and efficient algorithm for rice panicle segmentation.
- To improve the accuracy and speed of image-based rice phenotyping.
Main Methods:
- Proposed Panicle-SEG algorithm utilizes simple linear iterative clustering (SLIC) superpixels, convolutional neural network (CNN) classification, and entropy rate superpixel optimization.
- A dataset of 684 training and 48 testing images was used to build and evaluate the Panicle-SEG-CNN model.
Main Results:
- Panicle-SEG demonstrated superior segmentation accuracy compared to existing methods (HSeg, i2 hysteresis thresholding, jointSeg).
- Achieved average segmentation results of 0.626 (Qseg), 0.730 (Sr), 0.891 (SSIM), 0.821 (Precision), 0.730 (Recall), and 76.73% (F-measure).
- Improved execution speed through multithreading and CUDA parallel acceleration.
Conclusions:
- Panicle-SEG is a robust method for rice panicle segmentation, applicable across various conditions and rice types.
- This algorithm offers new possibilities for non-destructive yield estimation in rice.
- The developed software and dataset are publicly available.
