Enhancing the estimation of cadmium content in rice leaves by integrating vegetation indices and color indices using machine learning.

Abstract

Cadmium (Cd) is a heavy metal recognized for its notable biotoxicity. Excessive Cd levels can have detrimental effects on crop growth, development, and yield. Real-time, rapid, and nondestructive monitoring of Cd content in leaves (LCd) is essential for food security. Previous research has primarily utilized traditional statistical methods and heavy metal-related vegetation indices (VIs) to develop models for estimating LCd, often resulting in a lack of generalizability. Herein, 252 sets of leaf samples with varying Cd contents were collected under six Cd concentration gradients in hydroponic and soil cultivation conditions. An LCd estimation model was developed by integrating VIs, color indices (CIs), and machine learning (ML) algorithms. Results indicate that VIs and CIs were strongly correlated with LCd, exhibiting correlation coefficients (r) of 0.73 and 0.57, respectively. The ML estimation model, which integrated both indices, was more effective than the single-parameter model developed using traditional statistical methods. Notably, the LCd estimation model developed using the random forest method exhibited the highest accuracy, with a coefficient of determination (R²) of 0.81 and a root-mean-square error of 0.120. These results indicate that multisource index data based on ML algorithms can effectively estimate LCd. This study presents an accurate, reliable, and generalized method to estimate LCd, providing valuable insights for assessing the large-scale heavy metal pollution status of rice using unmanned aerial vehicle remote sensing technology.