A Cost-effective Multispectral Sensor System for Leaf-Level Physiological Traits
With the concern of the global population to reach 9 billion by 2050, ensuring global food security is a prime challenge for the research community. One potential way to tackle this challenge is sustainable intensification; making plant phenotyping a high throughput may go a long way in this respect. Among several other plant phenotyping schemes, leaf-level plant phenotyping needs to be implemented on a large scale using existing technologies. Leaf-level chemical traits, especially macronutrients and water content are important indicators to determine crop’s health. Leaf nitrogen (N) level, is one of the critical macronutrients that carries a lot of worthwhile nutrient information for classifying the plant’s health. Hence, the non-invasive leaf’s N measurement is an innovative technique for monitoring the plant’s health. Several techniques have tried to establish a correlation between the leaf’s chlorophyll content and the N level. However, a recent study showed that the correlation between chlorophyll content and leaf’s N level is profoundly affected by environmental factors. Moreover, it is also mentioned that when the N fertilization is high, chlorophyll becomes saturated. As a result, determining the high levels of N in plants becomes difficult. Moreover, plants need an optimum level of phosphorus (P) for their healthy growth. However, the existing leaf-level P status monitoring methods are expensive, limiting their deployment for the farmers of low resourceful countries. The aim of this thesis is to develop a low-cost, portable, lightweight, multifunctional, and quick-read multispectral sensor system to sense N, P, and water in leaves non-invasively. The proposed system has been developed based on two reflectance-based multispectral sensors (visible and near-infrared (NIR)). In addition, the proposed device can capture the reflectance data at 12 different wavelengths (six for each sensor). By deploying state of the art machine learning algorithms, the spectroscopic information is modeled and validated to predict that nutrient status. A total of five experiments were conducted including four on the greenhouse-controlled environment and one in the field. Within these five, three experiments were dedicated for N sensing, one for water estimation, and one for P status determination. In the first experiment, spectral data were collected from 87 leaves of canola plants, subjected to varying levels of N fertilization. The second experiment was performed on 1008 leaves from 42 canola cultivars, which were subjected to low and high N levels, used in the field experiment. The K-Nearest Neighbors (KNN) algorithm was employed to model the reflectance data. The trained model shows an average accuracy of 88.4% on the test set for the first experiment and 79.2% for the second experiment. In the third and fourth experiments, spectral data were collected from 121 leaves for N and 186 for water experiments respectively; and Rational Quadratic Gaussian Process Regression (GPR) algorithm is applied to correlate the reflectance data with actual N and water content. By performing 5-fold cross-validation, the N estimation shows a coefficient of determination (R^2) of 63.91% for canola, 80.05% for corn, 82.29% for soybean, and 63.21% for wheat. For water content estimation, canola shows an R^2 of 18.02%, corn of 68.41%, soybean of 46.38%, and wheat of 64.58%. Finally, the fifth experiment was conducted on 267 leaf samples subjected to four levels of P treatments, and KNN exhibits the best accuracy, on the test set, of about 71.2%, 73.5%, and 67.7% for corn, soybean, and wheat, respectively. Overall, the result concludes that the proposed cost-effective sensing system can be viable in determining leaf N and P status/content. However, further investigation is needed to improve the water estimation results using the proposed device. Moreover, the utility of the device to estimate other nutrients as well as other crops has great potential for future research.
Plant Phentyping, Multispectral, Sensor, Leaf Nitrogen, Leaf Phosphorus, Leaf Water, Spectroscopy, Machine Learning
Master of Science (M.Sc.)
Electrical and Computer Engineering