Cross-variety seed vigor detection using new spectral analysis techniques and ensemble learning methods

Abstract

During the storage process, seeds deteriorate with aging, and their vigor gradually decreases, affecting the field emergence rate, uniformity, and crop yield. Therefore, it is critical to accurately and quickly identify seed vigor. Spectroscopic techniques are widely used in research to evaluate seed vigor rapidly. However, current research predominantly focuses on single varieties, with limited exploration of generalized detection methodologies for multi-variety seeds. In this paper, autofluorescence and reflectance spectroscopy techniques were used to detect several varieties of wheat seeds and analyze the changing pattern of their spectra after artificially accelerated aging treatment. The analysis compares the effectiveness of feature modeling of autofluorescence and reflectance spectra in cross-variety detection. By integrating these two spectral features through ensemble learning, a seed vigor detection model capable of cross-variety identification was developed and validated with an optimal accuracy of more than 87.5 %. Furthermore, for seed vigor group detection, the detection strategy was designed to discriminate small batches of seeds, and the final detection accuracy reached 93.9 % on average. In summary, this study shows that ensemble learning can combine the advantages of autofluorescence and reflectance spectroscopy, enhance the discrimination ability of combined datasets, and provide an effective reference for cross-variety seed vigor detection.

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.