Poplar seedling varieties and drought stress classification based on multi-source, time-series data and deep learning

Abstract

Drought is a main abiotic stress facing agriculture and forestry production and its impacts are exacerbated by climate change. Accurately and effectively monitoring the drought stress levels of crop and tree species is crucial for their efficient management and the selection of drought-resistant varieties. This study used four poplar seedlings with differing drought tolerance and conducted five drought stress level tests. A self-propelled phenotyping platform was constructed and equipped with an Intel RealSense D435i RGB-D (Red-Green- Blue-Depth) camera and a RedEdge-MX multispectral camera. The side-view RGB and depth images and five-channel top-view multispectral images of poplar seedlings were collected by this platform; and the color, texture, depth, and spectral features were extracted through image processing. In addition, plant height, ground diameter, leafstalk angle, chlorophyll content and water content were collected from the poplar seedlings. Using long short-term memory (LSTM), a multi-output classification was performed on the varieties and drought stress levels of the four poplar seedlings based on the 50 parameters obtained through manual and image processing. By combining ResNet18 and the improved ResNet18 embedded in the convolutional block attention module (CBAM) with the LSTM model, the resulting ResNet18-LSTM and ResNet18-CBAM-LSTM models were used to perform multi-output grading on the varieties and drought stress levels. As for varieties, the classification accuracy of LSTM, ResNet18-LSTM, and ResNet18-CBAM-LSTM models were 96.56 %, 98.12 %, and 99.69 %, respectively. As for drought stress levels, the classification accuracy of the three models were 83.44 %, 85.62 %, and 90.94 %, respectively. The ResNet18-CBAM-LSTM model performed the best in the classification of the two parameters. This study comprehensively and continuously monitored the dynamic response of multiple varieties of poplar seedlings under different drought conditions, and a new perspective for the classification of drought stress levels and the breeding of better varieties are provided.

Plant phenotyping relevance

自走式フェノタイピングプラットフォーム、RGB-D・マルチスペクトル画像、画像処理による形質抽出、深層学習分類が研究の中心であり、ポプラの品種と干ばつストレス状態を評価する実質的なフェノタイピング手法研究である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.