Diagnostic study of defoliation and boll opening effects on machine-harvested cotton using multi-source UAV remote sensing data

Abstract

Accurate assessment of cotton defoliation (DF) and boll opening (BO) is essential for optimizing yield and fiber quality during mechanized harvesting, as improper timing can reduce yield and impair fiber quality. Unmanned aerial vehicle (UAV)-based remote sensing has become an effective tool for monitoring these indicators, but most current methods rely on single-sensor data, limiting diagnostic accuracy and generalizability. To address this limitation, we propose a multi-source data fusion framework integrating RGB, multi-spectral (MS), and thermal infrared (TIR) sensors for comprehensive canopy information. The fused dataset includes vegetation indices (VIs), color indices (CIs), texture features (Tex), and canopy temperature (TC). Feature selection was performed using pearson correlation coefficients (PCCs), recursive feature elimination with cross-validation (RFECV), and the Boruta algorithm to identify key variables. Three machine learning models—partial least-squares regression (PLSR), random forest regression (RFR), and extreme gradient boosting regression (XGBR)—were developed and compared. The RFECV-selected RGB+MS+TIR features in the XGBR model achieved the highest predictive accuracy, with R² values of 0.918 for defoliation rate and 0.867 for boll opening rate, improving by 1.9 % and 4.3 %, respectively, over single-sensor models. Root mean square error (RMSE) and relative RMSE (rRMSE) were reduced by 1.11 %-1.99 % and 1.66 %-2.28 %, respectively. These findings demonstrate that multi-source UAV data fusion, combined with advanced machine learning techniques, significantly enhances the accuracy and robustness of cotton defoliation and boll opening diagnosis. This approach offers a practical solution for precision agriculture to improve harvest scheduling and defoliant management.

Plant phenotyping relevance

UAVのRGB・マルチスペクトル・熱赤外データを融合し、綿花の落葉率と綿花開絮率という植物状態を推定する手法を開発・比較しており、フェノタイピング手法が研究の中心である。

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.