Artificial neural network for estimating leaf fresh weight of rice plant through visual-nir imaging

Abstract

Prediction of fresh biomass is the key for evaluation of the response of crop genotypes to diverse input and stress conditions, and forms basis for calculating net primary production. Hence, accurate and high throughput estimation of fresh biomass is critical for plant phenotyping. As conventional phenotyping approaches for measuring fresh biomass is time consuming, laborious and destructive, image based phenotyping methods are being widely used now in plant phenotyping. However, current approaches for estimating fresh biomass of plants are based on projected shoot area estimated from the visual (VIS) image. These approaches do not consider the water content of the plant tissues which are about 70-80% in leafy vegetation. Since water absorbs radiation in the Near Infra-Red (NIR) (900–1700 nm) region, it has been hypothesized that combined use of VIS and NIR imaging can predict the fresh biomass more accurately that the VIS image alone. In this study, VIS and NIR imaging were captured for rice leaves with different moisture content as a test case. For background subtraction from NIR image, PlantCV v2 NIR imaging algorithm was implemented in MATLAB software (version 2015b). The proposed image derived parameter, viz. Green Leaf Proportion (GLP) from VIS image and mean gray value/intensity (NIR_MGI) from NIR image were used as input to develop Artificial Neural Network (ANN) model to estimate the Leaf Fresh Weight (LFW). This proposed approach significantly enhanced the fresh biomass prediction as compared to the conventional regression technique based on projected shoot area derived from VIS image.

Keywords

• Artificial neural network
• Green leaf proportion
• Image analysis
• LFW
• Non-destructive phenotyping
• Rice