Portable Optical Sensors for the Detection of Agricultural and Environmental Contaminants

The contamination of the environment and agricultural products pose major threats to human health and the world’s ecosystems. In order to mitigate the effects of these issues, it is imperative that systems be developed that can detect the presence of contaminants at an early stage. The objectives of this thesis were to evaluate portable methods for detecting plant stress, bacterial contamination on produce, and types of oil in oceanwater oil spills. The first research project consisted of a smartphone-based system that used fluorescence imaging data to predict the species and concentration of bacteria on spinach leaves. The system utilized a smartphone attachment that contained a 405nm LED to produce the fluorescent excitation light and 7 different optical filters to capture the fluorescence emission at various wavelengths. Four different bacteria species and three different concentrations of the bacteria were imaged with this attachment. The images captured were then processed to produce an average intensity value for each image, which was used as a gauge for the intensity of the fluorescence emission. The intensity data was then analyzed using an analysis of variance (ANOVA) procedure for a two-factor factorial design with statistical analysis software. Although the system shows promise in differentiating between bacteria species, the study is currently inconclusive and more data needs to be taken and evaluated. The second project aimed to predict the saturate, aromatic, resin, and asphaltene (SARA) contents of oil samples mixed with ocean water using nonlinear machine learning regression methods. The training data consisted of samples diluted with dichloromethane (DCM). The first phase of the analysis involved randomly splitting this dataset into testing and training components and evaluating the performance of various scaling and regression methods on these random splits. The combinations that performed the best were selected to undergo an independent validation test with separate samples diluted in ocean water. The system performed better with random splitting of the training data as opposed to independent validation, but this was due to inconsistencies with the solvent used to dilute the oil samples