Non-Destructive Biomass Estimation Based on 3D Reconstruction From A Handheld Camera

Jaspreet Singh, Grzegorz Cielniak

Abstract: With the recent advancements in sensing technology and machine learning, it has become possible to develop agricultural technologies that can keep pace with the shrinking agriculture workforce and growing population needs. Aboveground biomass (AGB) is a key trait for crop growth monitoring, crop breeding and yield prediction for agricultural scientists as well as for farmers. It is essential to develop an accurate non-destructive method because the destructive measurement of AGB is manual, time-consuming and expensive. In this paper, we propose and validate a new pipeline based on the latest low-cost technologies to accurately acquire 3D point clouds of the crop plots using a consumer camera and an off-the-shelf structure-from-motion reconstruction algorithm. Unlike the previous methods, the proposed non-destructive AGB pipeline does not rely on large amount of training data and high-cost field robots to capture the 3D data. The proposed pipeline for estimating AGB consists of three steps: i) 3D reconstruction of the crop plot, ii) estimating the volume of the crop plot, and iii) estimating the biomass from the volume. The experimental results showed a strong correlation between the plot volume and biomass with the minimum error in the final estimated biomass as compared to the recorded ground truth biomass.