Robust Detection of Plant Features with Overhead Imaging in a Range of Crop and Weed Scenarios

Abstract

Overhead imaging from UAVs and UGVs can help detect features of interest for farm scenarios with crop-specific features such as flowers, fruits, and surrounding weeds. These features are important for yield assessment and de-weeding. However, identifying them can be challenging due to their morphological similarity and occlusion. We use deep learning to process high-resolution images of crop fields and classify the features at different complexity levels, for example, to detect green-coloured fruits that are partly concealed beneath leaves. We compare different approaches for multiple seasonal crops and weeds in the Kharif season and discuss their performance and limitations. The feature recognition outputs are illustrated at various complexity levels, which should help pave the way forward in creating holistic insights that drive action on the ground to enhance management efficiency and profitability.

Plant phenotyping relevance

作物の花・果実などの植物器官をオーバーヘッド画像と深層学習で検出・分類し、収量評価に利用する画像解析手法が研究の中心であるため。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.