Advanced-Lettuce: An integrated lettuce growth and yield prediction software

Lettuce is a globally important economic crop, and accurate prediction of its growth and yield is of great importance for its breeding and commercialization. In this study, we developed an image-based software for lettuce growth and yield prediction. Using a dataset of 15,187 background-free images of 237 lettuce varieties representing six major types, we trained a high-accuracy classification model (98.7%). Based on this classification model, using a generative adversarial network (GAN), we developed the first single image-based stepwise growth prediction model which could generate the images of lettuce in the coming 5 days during growth period ( SSIM = 0.948 ). Based on these developed models, we proposed three methods of image-based yield prediction, achieving a maximum prediction accuracy of 0.80. Finally, we integrated these models and methods into a user-friendly desktop application and named it the Advanced Lettuce for lettuce segmentation, classification, growth prediction, and yield prediction. Case study using an independently collected dataset further demonstrated the practical applicability of the software under realistic field conditions. Overall, this study developed an integrated lettuce growth and yield prediction software that may facilitate lettuce phenotyping, breeding, and production research.