Wheat yield prediction through artificial bee colony-enhanced convolutional neural network

Crop improvement programs aim to develop high-yielding varieties, coupled with resistance to biotic and abiotic stresses with nutritional superiority. Grain yield, being a complex trait is governed by genotypes, environment, and their interaction. Growing of large number of genotypes under multiple environments and measuring grain yield and its components are tedious and resource-consuming tasks. Therefore, there is a great need for novel, cost-effective techniques to evaluate the performance of crops at the field scale through indirect selection of easily scorable traits using sound algorithms based on comprehensive data. Convolutional neural networks (CNN) are one of the most promising deep learning methods for dealing with several complex tasks including crop yield prediction, but their performance is affected by manually set hyper-parameters. To address this, we proposed the artificial bee colony optimizer to efficiently search the hyper-parameters of CNN models for predicting the wheat yield on the basis of normalized difference vegetation indices, canopy temperature and plant height. Models are developed on crop yield data using 3350 germplasm of wheat planted in two growing environments as well as two different locations during the winter season of 2020-21. When compared to other popular optimization algorithms, such as genetic algorithms and particle swarm optimizers, the proposed model is proven to be superior for predicting wheat yield in terms of root mean square error (RMSE) (66.44–80.68 g/m²) and R² (0.88–0.91) and at the same time greatly reduced the computational time. In addition, crop yield prediction using the proposed model can support different management decisions, including timing and amount of fertilization and selective breeding.