Abstract
In situ real-time monitoring of physiological information during crop growth (such as leaf chlorophyll values and water content) is crucial for enhancing agricultural production efficiency and crop management practices. In traditional agricultural monitoring, commonly used measurement methods, such as chemical analysis for determining leaf chlorophyll values and drying methods for measuring water content, are all non-in situ measurement techniques. These methods not only risk damaging the plants but may also impact plant growth and health. Furthermore, the complex setup of traditional spectrometers complicates the data collection process, which limits their practical application in plant monitoring. Therefore, there is an urgent need to develop a novel, user friendly, and plant-safe monitoring technology to improve agricultural management efficiency. To this end, this study proposes a novel wearable flexible sensor designed for in situ real-time monitoring of leaf chlorophyll values and water content. This sensor is lightweight, portable, and allows for flexible placement, enabling continuous monitoring by conforming to plant surfaces. Its spectral response covers multiple bands from near ultraviolet to near infrared, and it is equipped with an active light source ranging from ultraviolet to infrared to enable efficient measurements under various environmental conditions. In addition, the sensor is securely attached to the underside of the leaf using a magnetic suction method, ensuring long-term stable in situ monitoring, thus continuously collecting important physiological information throughout the crop growth cycle. Analysis of the sensor-collected data reveals that for leaf chlorophyll, Gaussian process regression shows the best prediction performance during multi-spectral scattering correction, with R(c) (2) of 0.8261 and RMSEc of 1.7444 on the training set; the performance on the test set is Rp² of 0.7155 and RMSE(p) of 2.0374. Meanwhile, for leaf water content, across various data preprocessing scenarios, gradient boosting regression can effectively predict it, yielding Rc² of 0.9401 and RMSEc of 0.0028 on the training set; the performance on the test set is R(c) (2) of 0.6667 and RMSE(p) of 0.0067.