外文摘要：The identification of different plant genotypes, particularly during various developmental stages, is challenging due to the high genetic diversity of agricultural products. In this study, the efficacy of visible and infrared leaf spectroscopy in identifying five wheat genotypes was assessed across three main growth stages: leaf opening, tillering, and flowering. Soft independent modelling of class analogy (SIMCA) and artificial neural networks (ANN) were employed to develop growth -stage dependent and independent classification models based on spectral reflection. To handle the large amount of spectral data generated, a deep learning technique was first used to extract advanced features from the spectral data using the stacked autoencoder (SAE) approach. The results revealed that the tillering stage yielded the highest accuracy in predicting wheat genotypes when using the linear SIMCA approach. Furthermore, employing deep learning features extracted by SAE networks in the ANN classifier resulted in superior performance compared to the SIMCA classifier. During the leaf opening and tillering stages, the ANN classifier achieved perfect classification, with a weighted F1 -score of 100 %, while during the flowering stage, it achieved a weighted F1 -score of 98.02 %. In the combined dataset, the ANN classifier demonstrated impressive performance, with a weighted F1 -score of 95.98 %. These findings suggest that in -field leaf spectroscopy techniques using deep learning feature extraction methods can be a suitable approach for identifying wheat plant genotypes during different growth stages.
外文关键词：Flowering;artificial neural networks (ANN);Wheat genetic diversity;Soft independent modelling of class analogy (SIMCA);Stacked autoencoder (SAE);Leaf opening;Tillering
作者：Mireei, Seyed Ahmad;Salehi, Bakhtiyar;Jafari, Mehrnoosh;Hemmat, Abbas;Majidi, Mohammad Mahdi
作者单位：Isfahan Univ Technol
期刊名称：BIOSYSTEMS ENGINEERING
期刊影响因子：0.0
出版年份：2024
出版刊次：238
原文传递申请：江苏省科技资源（工程技术文献）统筹服务平台