Chlorophyll dynamic fusion based on high-throughput remote sensing and machine learning algorithms for cotton yield prediction

Abstract

In the development of precision agriculture, accurate monitoring of growth dynamics and yield prediction are important for optimizing the cotton production. However, the reliability of yield prediction for different cotton accessions based on monitoring chlorophyll dynamics using remote sensing techniques remains to be further improved. In this study, field images of 419 cotton accessions were collected at eight time points using an unmanned aerial vehicle (UAV) equipped with RGB and multispectral sensors. Four vegetation indices (VIs) and three color indices (CIs) highly correlated with leaf chlorophyll content (LCC) were screened by a hierarchical segmentation method. Subsequently, four machine learning algorithms were used to construct a chlorophyll prediction model. The random forest (RF) model outperformed the ridge regression (RR), support vector machine (SVM), and partial least squares regression (PLSR) models in predicting the LCC with the best accuracy (R² = 0.827) after fusing the VIs and CIs. Yield prediction was then performed using multiple linear regression, with LCC data as predictors from the squaring, flowering, boll development, and boll opening stages (as well as a combined multi-stage input). The model using multi-stage LCC data achieved the highest accuracy (R² = 0.723). Finally, the K-means algorithm was successfully used to demonstrate a strong association between LCC profiles and yield outcomes of 419 cotton accessions. This study fused RGB and multispectral data to predict multiple-stage LCC data and construct a yield prediction model, which was innovatively applied to 419 cotton accessions with rich genetic diversity for the first time. It is superior to methods relying on a single remote sensing dataset or applied to a few genetic materials, providing an innovative method for large-scale phenotypic analysis of cotton varieties and a scientific basis for promoting cotton germplasm screening and precision breeding.

Plant phenotyping relevance

UAVのRGB・マルチスペクトル画像と機械学習により、綿の葉緑素含量を推定する方法を開発・比較検証し、多数系統への大規模表現型解析に適用しているため、フェノタイピング手法が中心である。

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.