Estimating above ground biomass of winter wheat at early growth stages using digital images and deep convolutional neural network

Abstract

Abstract Above ground biomass (AGB) is a critical trait indicating the growth of winter wheat. Currently, non-destructive methods for measuring AGB heavily depend on tools such as Remote Sensing and LiDAR, which is subject to specialized knowledge and high-cost. Low-cost solutions appear therefore to be a necessary supplement. In this study, an easy-to-use AGB estimation method for winter wheat at early growth stages was proposed by using digital images captured under field conditions and Deep Convolutional Neural Network (DCNN). Using canopy images as input, the DCNN was trained to learn the relationship between the canopy and the corresponding AGB. To compare the results of the DCNN, conventionally adopted methods for estimating AGB in conjunction with some color and texture feature extraction techniques were used. Results showed strong correlations could be observed between the actual measurements of AGB to those estimated by the DCNN, with high coefficient of determination (R2 = 0.808) and low Root-Mean-Square-Error (RMSE = 0.8913 kg/plot, NRMSE = 24.95%). Factors may influence the accuracy of the DCNN were evaluated. Results showed selecting suitable values of these factors for the DCNN was the guarantee to accurate estimation results. Plant density was proved to be an influence of factor to all the estimation methods based on digital images. The performances of all the methods were influenced to varying degrees while the DCNN achieved the best robustness, indicating the DCNN with RGB images could be an efficient and robust tool for estimating AGB of winter wheat at early growth stages.