Exploring frogeye leaf spot disease severity in soybean through hyperspectral data analysis and machine learning with orange data mining

Importance of the work:With the advancement of hyperspectral remote sensing technology, the potential of categorising frogeye leaf spot (FLS) of soybean has been examined.No previous study has investigated Orange mining tool as visual programming approach in analysing hyperspectral reflectance data, especially in crop disease detection. Objectives:The main objective of the study is to classify the severity level of FLS disease in soybean using hyperspectral reflectance data and machine learning algorithms. Materials and Methods:We used hyperspectral reflectance data from healthy and FLS of soybeans.The first step was to smooth out the data by applying a filtering technique called Savitzky-Golay to remove the spectrum noise.The ReliefF feature selection technique was used to determine the most influential wavelengths for the classification of FLS disease severity in soybean.Next, machine learning (ML) methods (i.e.decision tree, gradient boosting, random forest, stacking, and neural network) were used to classify FLS of soybean.This analysis' performance was evaluated using overall accuracy, F1, precision and the receiver operating characteristic curve metric.All of these steps were conducted using Orange data mining software.Results: Based on the findings, neural network scored the highest overall accuracy of 98.6% after conducting filtering technique.Furthermore, reliefF-Gradient boosting and random forest algorithms achieved promising overall accuracy of 97.4% and 96.9%, respectively after implementing filtering and feature selection techniques.Main finding: Due to the integration of workflow and the specially designed spectroscopic widget in Orange Data Mining Software, it is capable of processing hyperspectral reflectance data in order to determine the severity level of disease affecting crops.