AUTHOR=Du Jianjun , Lu Xianju , Fan Jiangchuan , Qin Yajuan , Yang Xiaozeng , Guo Xinyu TITLE=Image-Based High-Throughput Detection and Phenotype Evaluation Method for Multiple Lettuce Varieties JOURNAL=Frontiers in Plant Science VOLUME=Volume 11 - 2020 YEAR=2020 URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2020.563386 DOI=10.3389/fpls.2020.563386 ISSN=1664-462X ABSTRACT=The yield and quality of fresh lettuces are determined by the growth rate and color status of vegetables. Manual investigation and phenotype evaluation for hundreds of varieties of lettuce is very time-consuming and labor-intensive. In this study, we utilized a greenhouse phenotyping platform to periodically capture top-view images of lettuce in the "Sensor-to-Plant" work mode, and established a dataset containing 2000 plants of 500 lettuce varieties at 8 time points during vegetative growth. We then presented a novel object detection - semantic segmentation - phenotyping method based on Convolutional Neural Networks (CNNs) to carry out noninvasive and high-throughput phenotype evaluation for the growth and development status of different lettuce varieties. Multistage CNN models for object detection and semantic segmentation were integrated to bridge the image sequence and plant phenotyping. Object detection model was used to detect and identify each POT from image sequence with 99.82% accuracy, semantic segmentation model was utilized to segment and identify each lettuce with 97.65% F1 score, and phenotyping pipeline was performed to extract a total of 15 static traits (geometry and color) of each lettuce. Furthermore, the dynamic traits (growth and accumulation rates) were calculated based on the changing curves of static traits. The correlation and descriptive abilities of these static and dynamic traits were carefully evaluated for the interpretability of traits to growth and quality of lettuce. Finally, we validated the application of image-based high-throughput phenotyping in geometric measurement and color grading for large-scale lettuce varieties. The proposed method can be extended to potted crops such as maize, wheat and soybean for the noninvasive phenotype evaluation and identification.