Arabidopsis Tracker: A Centroid-Based Vegetation Localization Model for Automatic Leaf Canopy Phenotyping in Multiple-Pot Cultivation System

Abstract

Isolation of individual crop in a multiple cropping agricultural system exhibits a challenging issue of mispredictions for each crop specially when plant plots are too near with each other. Likewise, manual phenotyping of numerous crops is time and labor intensive. In this study, phenotype signatures of 24 Arabidopsis thaliana weeds with rosette leaves horticultured in pot-based configuration were nondestructively tracked and extracted from germination to head development stage (27 days) to quantify its growth. It is employed using two major feature engineering processes, namely generation of centroid-based Arabidopsis localization using Raspberry Pi-captured top-view image and growth signature analysis of localized Arabidopsis. To filter annotated images, mask size was modeled using cubic regression. ImageJ platform was configured to generate ground truth images and measurements. Arabidopsis localized raw spectro-morphological signatures namely RGB reflectances, canopy area, convex-hull area, canopy diameter, and perimeter were extracted using blob analysis. Stockiness, relative growth rate, and compactness relatively increases by 28.2×10−3, 0.46×10−3 and 220.1×10−3 per day. Stockiness was observed to be a strong indicator that a weed is growing on its basal vegetative stage. This developed model with sensitivity of 98% is a recommendable approach using computer vision for both field and indoor individual crop analysis such as in lettuce and mustard farms.