Wheat Breeding Across the Full Lifecycle with Multimodal Deep Learning: Practical Review and Future Strategies

This review explores the use of multimodal deep learning in wheat breeding to tackle issues like extended cycles, inefficiency, and environmental adaptability. It integrates diverse data and deep learning to monitor wheat's growth, stress resistance, and yield. The study employs models like LSTM, ResNet50, and GNN to extract phenotypic features, creating a new approach to full lifecycle breeding. A multimodal dataset with hyperspectral and RGB images and genotypic data is used to extract features, ensuring data analysis accuracy. The research also includes stress resistance analysis and genotypic association mining to develop precise breeding strategies. A data platform is built for efficient data storage, analysis, and sharing in wheat breeding. The study supports the advancement of smart agriculture through multimodal deep learning, enabling monitoring and precise breeding across wheat's growth. Future work will optimize algorithms, datasets, and model generalization for broader application in wheat breeding.