Water Stress Identification in Chickpea Plant Shoot Images using Deep Learning

Abstract

The Stress due to water deficiency in plants can significantly lower the agricultural yield. In recent years, computer vision-based plant phenomics has emerged as a promising tool for plant research. Such techniques have the advantage of being non-destructive, non-evasive and fast. Pulses like chickpeas play an important role towards ensuring food security in poor countries owing to their high protein. In present work, we have compared the performance of traditional Machine Learning with that of deep techniques in classifying water stress using chickpea shoot images. From the experimental result, it is concluded that Deep Learning based methods are superior to the conventional Machine Learning methods. They are superior not only in the performance metrics but also in context that they don't need any handcrafted features or sophisticated feature selection methods. Among all Deep Learning methods, ResNet-18 renders better classification performance than the conventional CNN by attaining 86% and 84% accuracy.

Plant phenotyping relevance

ヒヨコマメのシュート画像から水ストレス状態を分類する画像ベースの計算手法を、機械学習と深層学習で比較評価しており、植物表現型・状態推定が研究の中心である。

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.