Open Plant Phenotype Database of Common Weeds in Denmark

Simon Leminen Madsen,Mads Dyrmann,Rasmus Nyholm Jørgensen,Morten Stigaard Laursen,Laura-Carlota Paz,Solvejg Kopp Mathiassen

doi:10.3390/rs12081246

Abstract

For decades, significant effort has been put into the development of plant detection and classification algorithms. However, it has been difficult to compare the performance of the different algorithms, due to the lack of a common testbed, such as a public available annotated reference dataset. In this paper, we present the Open Plant Phenotype Database (OPPD), a public dataset for plant detection and plant classification. The dataset contains 7590 RGB images of 47 plant species. Each species is cultivated under three different growth conditions, to provide a high degree of diversity in terms of visual appearance. The images are collected at the semifield area at Aarhus University, Research Centre Flakkebjerg, Denmark, using a customized data acquisition platform that provides well-illuminated images with a ground resolution of ∼6.6 px mm − 1 . All images are annotated with plant species using the EPPO encoding system, bounding box annotations for detection and extraction of individual plants, applied growth conditions and time passed since seeding. Additionally, the individual plants have been tracked temporally and given unique IDs. The dataset is accompanied by two experiments for: (1) plant instance detection and (2) plant species classification. The experiments introduce evaluation metrics and methods for the two tasks and provide baselines for future work on the data.

Highlights

Visual recognition systems are becoming increasingly widespread for farm management in modern agriculture [1,2,3,4]
The high capacity of airborne platforms usually comes at the cost of spatial resolution, which together variable scene illuminations causes a high risk of erroneous plant detections and identifications, which might lead to faulty syllogism/interpretations [9,10]
The full box images are divided into ten splits, based on trail id and box id, so each polystyrene box is only present in a single data split

Summary

Introduction

Visual recognition systems are becoming increasingly widespread for farm management in modern agriculture [1,2,3,4]. The systems are typically used in conjunction with remote sensing technologies to extract knowledge about various field conditions. Airborne sensing platforms, such as satellites and Unmanned Aerial Vehicles (UAVs), have in recent years been demonstrated to be efficient for mapping vegetation coverage [5] and weed infestations (binary distinction between weed or crop) [6,7,8] in fields, due to the relatively high capacity of such systems. Ground-based sensing platforms (proximal sensing) generally provide higher spatial resolution compared to airborne platforms, which can be used to efficiently map the population and composition of plants in fields at species level [10,12,13,14]. Proximal sensing and visual recognition can be applied for other applications, such as phenotyping, diseases and pests detection, grain quality and plant fitness estimation, etc. [2,18]

Objectives

Methods

Results

Discussion

Conclusion