Optimization strategies and visual validation for in-situ root image segmentation and repair
Year: 2025
Type: article
Abstract: Root systems play a crucial role in fixing and supporting plants, and are important for absorbing water and nutrients. The analysis of plant phenotypic parameters helps achieve precise management of crop production and enhance resource utilization efficiency. The traditional U-Net is susceptible to interference from soil backgrounds and lacks sufficient focus on root edges, while introducing attention mechanisms can guide the model to concentrate more on root details, meeting the requirements for root repair. Therefore, this experiment proposes an improved U-Net segmentation network that integrates sequentially connected CBAM and GAM modules along with sub-pixel convolution, achieving a mean mIoU of 85.34% and an F1-score of 91.53%. A repair network based on Pix2Pix was designed and combined with the segmentation network. Additionally, the Grad-CAM algorithm is used to compare heatmaps to explain the segmentation focus of each model and verify the advantages of the proposed method. Finally, the distribution of lateral root angles in in situ root images under different environments is counted using ImageJ software. The data results show that the angle of cotton lateral roots under stress conditions is larger than that under normal conditions. In conclusion, the improved segmentation model and repair model proposed in this paper can accurately segment root systems and repair broken roots, providing new ideas and methods for root phenotypic research. The code can be obtained on GitHub: https://github.com/jiwd123/improved-unet-and-p2p/tree/main . • A U-Net network model based on a hybrid attention mechanism was proposed, and its performance was visually validated using heatmaps. • An improved Pix2Pix model was designed to address and repair broken roots, specifically targeting fractures in slender roots within the segmented images. • By incorporating root phenotypic data, the feasibility of integrating the aforementioned two networks for broken root repair was verified.
Authors: Liuli Wang, Meng Zhang, Minye Li, Xingyun Liu, Xingyun Liu, Qiushi Yu, Lingxiao Zhu, Liantao Liu, Xiuwei Liu, Xiuwei Liu, Nan Wang
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: gold
Estimated APC paid: $2,530
OpenAlex ID: W7116074816