GAN-based image prediction of maize growth across varieties and developmental stages

Plant growth prediction assists physiologists and botanists in analyzing future development trends, thereby shortening experimental cycles and reducing costs. Traditional growth prediction methods mainly focused on phenotypic traits instead of images, which leads to limited visual interpretability. This article proposed a visualized growth prediction method based on an improved Pix2PixHD network, incorporating spatial attention mechanisms, an improved loss function, and a modified dropout strategy to enhance prediction accuracy and visual fidelity. The proposed method can employ maize images from early time points to predict the images of later stages. The prediction results are presented in the form of side-view growth images with a resolution of 1024 × 1024 pixels, enabling the capture of detailed, organ-level growth information. This study conducted experiments on 696 varieties, a highly genetically diverse maize population derived from the crossbreeding of 24 foundational Chinese inbred lines. The results showed that Fréchet Inception Distance, Peak Signal-to-Noise Ratio and structural similarity between the predicted images and the actual images reached 20.27, 23.23 and 0.899, respectively. The model achieved a mean Pearson correlation coefficient of 0.939 between predicted and actual phenotypic traits, while maintaining robust performance across different time intervals. It was also demonstrated that the model outperformed the existing related studies. The code is available online. The results showed that the method can make realistic predictions of multi-variety maize growth based on high-resolution generation. Furthermore, it can achieve prediction of maize growth throughout the entire growth cycle with high accuracy. In conclusion, this article provided a novel solution for visualized growth prediction of large plants with complex physiological structures throughout the entire growth cycle. A primary limitation of this study is its focus on modeling and predicting crop growth under uniform environmental conditions, without considering environmental variability. Future work will aim to incorporate diverse environmental factors into the model to enhance its robustness and predictive accuracy.