An enhanced stacking algorithm for inversing cotton growth parameters using multi-source remote sensing data

Abstract

Cotton is a significant broadacre crop globally, and monitoring its growth is crucial for improving agricultural productivity. With the development of unmanned aerial vehicle (UAV) remote sensing technology, the inversion of cotton growth parameters (including plant height (PH), and leaf chlorophyll content (LCC), leaf area index (LAI), above-ground biomass (AGB)) from remote sensing data has emerged as a prominent research area. To address the issue of limited accuracy in traditional stacking algorithm for remote sensing inversion, this study proposes an enhanced stacking algorithm (ESA). First, multi-source remote sensing data is acquired using UAV equipment equipped with an integrated payload of a LiDAR sensor and a visual RGB camera and raw features are extracted from the data. Then, principal component analysis (PCA) is used to reduce the dimensionality of these features. The model construction is optimized through the following steps: first, explore all feature combinations and train each using multi-class learners to construct the entire set of base models; second, remove over- or under-fitting models to build a candidate pool; third, introduce iterative screening to the pool—each round incorporates the algorithm with the greatest performance gain and removes those with negative contributions, iterating to construct an efficient subset of base models; finally, RidgeCV is used to fuse base-model outputs. The experimental results show that ESA outperforms other traditional methods in terms of prediction performance for the four growth parameters. Specifically, on the test set, the R2 values for PH, LCC, LAI, and AGB are 0.9320, 0.8015, 0.8638, and 0.8272 , respectively. Compared with the second-best model, the relative improvement is approximately 4.6% (PH), 3.6% (LCC), 7.5% (LAI), and 11.2% (AGB) . ESA offers an effective approach for high-precision inversion of cotton growth parameters, providing new insights for the precision management of other crops.

Plant phenotyping relevance

UAVのLiDAR・RGBデータから綿花の複数生育形質を推定するスタッキングアルゴリズムを開発し、他手法と性能比較しており、表現型取得・推定法が中心です。

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.