Fang Zhengxin, Fan Qinglu, Tian Luxu, Jiang Haoyu, Wang Chen, Fu Xiuqing, Li Xiaozhong, Li Meng, Zhang Shiyan, Zhang Yaben, Li Yingyue
College of Engineering, Nanjing Agricultural University, Nanjing, China.
Cotton Research Institute, Xinjiang Academy of Agricultural and Reclamation Sciences, Shihezi, China.
Front Plant Sci. 2024 Sep 13;15:1447346. doi: 10.3389/fpls.2024.1447346. eCollection 2024.
Seed germination vigor is one of the important indexes reflecting the quality of seeds, and the level of its germination vigor directly affects the crop yield. The traditional manual determination of seed germination vigor is inefficient, subjective, prone to damage the seed structure, cumbersome and with large errors. We carried out a cucumber seed germination experiment under salt stress based on the seed germination phenotype acquisition platform. We obtained image data of cucumber seed germination under salt stress conditions. On the basis of the YOLOv8-n model, the original loss function CIoU_Loss was replaced by ECIOU_Loss, and the Coordinate Attention(CA) mechanism was added to the head network, which helped the model locate and identify the target. The small-target detection head was added, which enhanced the detection accuracy of the tiny target. The precision P, recall R, and mAP of detection of the model improved from the original values of 91.6%, 85.4%, and 91.8% to 96.9%, 97.3%, and 98.9%, respectively. Based on the improved YOLOv8-ECS model, cucumber seeds under different concentrations of salt stress were detected by target detection, cucumber seed germination rate, germination index and other parameters were calculated, the root length of cucumber seeds during germination was extracted and analyzed, and the change characteristics of root length during cucumber seed germination were obtained, and finally the germination activity of cucumber seeds under different concentrations of salt stress was evaluated. This work provides a simple and efficient method for the selection and breeding of salt-tolerant varieties of cucumber.
种子发芽活力是反映种子质量的重要指标之一,其发芽活力水平直接影响作物产量。传统的人工测定种子发芽活力效率低、主观性强、易破坏种子结构、操作繁琐且误差大。我们基于种子发芽表型采集平台开展了盐胁迫下黄瓜种子发芽实验。获取了盐胁迫条件下黄瓜种子发芽的图像数据。在YOLOv8 - n模型基础上,将原损失函数CIoU_Loss替换为ECIOU_Loss,并在头部网络添加坐标注意力(CA)机制,有助于模型定位和识别目标。添加了小目标检测头,提高了微小目标的检测精度。模型检测的精确率P、召回率R和平均精度均值mAP分别从原来的91.6%、85.4%和91.8%提升至96.9%、97.3%和98.9%。基于改进后的YOLOv8 - ECS模型,通过目标检测对不同浓度盐胁迫下的黄瓜种子进行检测,计算黄瓜种子发芽率、发芽指数等参数,提取并分析黄瓜种子发芽过程中的根长,得到黄瓜种子发芽过程中根长的变化特征,最终评估不同浓度盐胁迫下黄瓜种子的发芽活性。这项工作为黄瓜耐盐品种的选育提供了一种简单高效的方法。