["Established a database containing over 20,000 rice field images and yield data to train a convolutional neural network model.", "The model explains 68-69% of yield variation, highlighting the importance of panicles in yield estimation.", "The model predicts accurately, is cost-effective, and does not require tedious crop sampling, demonstrating the potential for regional yield monitoring."]