Abstract:
Conventional plant disease detection approaches are time consuming and require high skills. Above all, it cannot be scaled down to smallholder farmers in most developing countries. Using low cost IoT sensor technologies that are gas, ultrasound and NPK sensors mounted next to maize varieties for profiling these parameters on a given period. Here we report an experiment performed under controlled environment to learn metabolic and pathologic behavioral patterns on healthy and NLB inoculated maize plants by generating time series dataset on profiled Volatile Organic Compounds (VOC), Ultrasound and Nitrogen, Phosphorus, Potassium (NPK). Dataset has been preprocessed with pandas and analyzed using machine learning models which are dickey fuller test and python additive statsmodel and visualized using matplotlib library to enable the inference of an occurrence of a disease a few days post inoculation without subjecting a plant to an invasive procedure. This enabled a deployment and implementation of noninvasive plant disease detection prior to visual symptoms that can be applied on other plants. With analyzed data, the IoT technology in this experiment has enabled the detection of NLB disease on maize disease within seven days post inoculation because of monitoring VOC and ultrasound emission.