基于计算机视觉的植物叶片分割技术对植物分类具有重要意义,监测植物生长,精准农业,和其他科学研究。在本文中,YOLOv8-seg模型用于图像中单个叶片的自动分割。为了提高分割性能,我们进一步在标准Yolov8模型中引入了Ghost模块和双向特征金字塔网络(BiFPN)模块,并提出了两个修改版本。Ghost模块可以通过廉价的转换操作生成几个内在特征图,BiFPN模块可以融合多尺度特征,提高小叶的分割性能。实验结果表明,Yolov8在叶片分割任务中表现良好,和Ghost模块和BiFPN模块可以进一步提高性能。我们提出的方法在植物表型(CVPPP)叶片分割挑战中的计算机视觉问题的所有五个测试数据集上实现了86.4%的叶片分割得分(最佳骰子)。表现优于其他报告的方法。
Computer-vision-based plant leaf segmentation technology is of great significance for plant classification, monitoring of plant growth, precision agriculture, and other scientific research. In this paper, the YOLOv8-seg model was used for the automated segmentation of individual leaves in images. In order to improve the segmentation performance, we further introduced a Ghost module and a Bidirectional Feature Pyramid Network (BiFPN) module into the standard Yolov8 model and proposed two modified versions. The Ghost module can generate several intrinsic feature maps with cheap transformation operations, and the BiFPN module can fuse multi-scale features to improve the segmentation performance of small leaves. The experiment results show that Yolov8 performs well in the leaf segmentation task, and the Ghost module and BiFPN module can further improve the performance. Our proposed approach achieves a 86.4% leaf segmentation score (best Dice) over all five test datasets of the Computer Vision Problems in Plant Phenotyping (CVPPP) Leaf Segmentation Challenge, outperforming other reported approaches.