Abstract: Efficient and accurate extraction of plant phenotypic parameters is an important guide for the management of plants during all growth stages. To improve the efficiency and accuracy of morphological parameter extraction in plant leaf images, a model framework and key functions were optimized based on a fullconvolutional neural network. Additionally, the plant leaf image segmentation effect was achieved through supervised learning methods. The mean recall of the model on the test set was 0.95 for r and 0.94 for MIoU. The morphological parameters of extracted plant leaves in the segmentation results were highly correlated with the manual extraction results whereby r2>0.96. In this study, high throughput segmentation of plant leaf images was achieved, and the plant leaf morphological parameters extracted from the segmentation results can be used in related studies such as crop growth monitoring.

Key words: image segmentation, deep learning, fully neural network, morphological parameter analysis

摘要： 为提高植物叶片图像中形态参数提取的效率和准确率,以全卷积神经网络为基础,对模型构架和关键函数进行优化,通过有监督的学习方法实现植物叶片图像分割效果。模型在测试集上的平均召回率r为0.95,MIoU为0.94。在分割结果中提取植物叶片的形态学参数与人工提取结果高度相关,r~2>0.96。该研究实现了植物叶片图像高通量地分割,并且在分割结果中提取的植物叶片形态参数可以用于作物长势监测等相关研究。

关键词: 图像分割, 深度学习, 全卷积神经网络, 形态参数解析