Early prediction of the growth and development of plants is important for the intelligent breeding process, yet accurate prediction and simulation of plant phenotypes is difficult. In this work, a prediction model of plant growth and development based on spatiotemporal long short-term memory (ST-LSTM) and memory in memory network (MIM) was proposed to predict the image sequences of future growth and development including plant organs such as ears. A novel dataset of wheat growth and development was also compiled. The performance of the prediction model of plant growth and development was evaluated by calculating structural similarity index measure (SSIM), mean square error (MSE), and peak signal to noise ratio (PSNR) between the predicted and real plant images. Moreover, the optimal number of time steps and the optimal time interval between steps were determined for the proposed model on the wheat growth and development dataset. Under the optimal setting, the SSIM values surpassed 84% for all time steps. The mean of MSE values was 46.11 and the MSE values were below 68 for all time steps. The mean of PSNR values was 30.67. When the number of prediction steps was set to eight, the prediction model had the best prediction performance on the public Panicoid Phenomap-1 dataset. The SSIM values surpassed 78% for all time steps. The mean of MSE values was 77.78 and the MSE values were below 118 for all time steps. The mean of PSNR values was 29.03. The results showed a high degree of similarity between the predicted images and the real images of plant growth and development and verified the validity, reliability, and feasibility of the proposed model. The study shows the potential to provide the plant phenotyping community with an efficient tool that can perform high-throughput phenotyping and predict future plant growth.

Wheat plays a crucial role in ensuring food security. However, its complex genetic structure and trait variation pose significant challenges for breeding superior varieties. In this study, a genomic prediction method for wheat (WheatGP) is proposed. WheatGP is designed to improve the phenotype prediction accuracy by modeling both additive genetic effects and epistatic genetic effects. It is primarily composed of a convolutional neural network (CNN) module and a long short-term memory (LSTM) module. The multilayer CNNs within the CNN module focus on capturing short-range dependencies within the genomic sequence. Meanwhile, the LSTM module, with its unique gating mechanism, is designed to retain long-distance dependency relationships between gene loci in the features. Therefore, WheatGP could comprehensively extract multilevel features from genomic inputs. Compared to ridge regression best linear unbiased prediction (rrBLUP), extreme gradient boosting (XGBoost), support vector regression (SVR), and deep neural network genomic prediction (DNNGP), WheatGP demonstrates a clear advantage in terms of prediction accuracy. The prediction accuracy for wheat yield reaches 0.73, while the prediction accuracies for various agronomic traits range between 0.62 and 0.78. It also exhibits robust performance across other crop types and multi-omics datasets. In addition, SHapley Additive exPlanations (SHAP) is employed to evaluate the contributions of inputs to the predictive model. As a high-performance tool for genomic prediction in wheat, WheatGP opens up new possibilities for achieving efficient and optimized wheat breeding.