Crop Yield Prediction Using Deep Neural Networks
The world’s population is on the rise and in order to feed the world in 2050, food production will need to increase by 70% . As a result, it is of great importance to construct powerful predictive models for phenotype prediction based on Genotype and Environment data (so-called G by E problem). The objective of the G by E analysis is to understand how genotype and the environment jointly determine the phenotype (such as crop yield and disease resistance) of plant or animal species. In this research, deep neural networks are trained and used as predictive models. Deep neural networks have become a popular tool in supervise learning due to considerable ability in training nonlinear features . Recent articles have stated that the network depth is a vital factor in decreasing classification or regression error. But, deeper networks have a so-called vanishing/exploding gradients problem which makes the training and optimizing deeper networks difficult. He et al. proposed residual learning method which alleviates this problem very well and showed that deep residual networks are significantly better and more efficient than previous typical networks . As a result, residual training has been used in this research to prevent gradient degradation and ease the optimization process. Finally, since it is difficult to predict the yield difference directly, two separate residual neural networks have been trained to predict yield and check yield. After training the networks, the RMSE for check yield and yield are 8.23 and 10.52, respectively, which are very good because of considerable amount of missing values, uncertainty, and complexity in the datasets.
KeywordsG-by-E interaction analysis Supervised learning Machine learning Deep neural networks
- 2.A. Krizhevsky, I. Sutskever, G.E. Hinton, Imagenet classification with deep convolutional neural networks. Adv. Neural Inf. Proc. Syst. 1097–1105 (2012)Google Scholar
- 3.C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, A. Rabinovich, Going deeper with convolutions, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1–9 (2015)Google Scholar
- 5.K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778 (2016)Google Scholar