Analysis of hyperspectral reflectance for disease classification of soybean frogeye leaf spot using Knime analytics

The feasibility of classifying soybean frogeye leaf spot (FLS) has been investigated with the advance of hyperspectral technology. Hyperspectral reflectance data of healthy and FLS disease soybeans were used. The first step was to smooth out the data by using a filtering technique namely Savitzky-Golay to eliminate the noise of the spectrum. In order to select the most significant wavelengths, genetic algorithm (GA) was used as a forward feature selection technique. This analysis involved the implementation of machine learning (ML) algorithms, including decision trees, random forests, and stacking, to classify soybean FLS severity levels. Preprocessing ML steps including converting class numbers to strings, identifying and removing missing values, partitioning and normalizing data were implemented prior to the development of the model. Overall accuracy and the receiver operating characteristic curve measure were used to assess the performance of this analysis. All of these steps were carried out through KNIME analytical platform. Based on the results of the analysis, GA-stacking and random forest algorithms achieved the best overall accuracy of 85.9 and 84.3, respectively. In terms of reproducibility, data flow control, data exploration, analysis and visualization, KNIME Analytics Platform provided great convenience in connecting tools graphically and ensuring the same results on different operating systems. The rapid implementation of workflow in KNIME Analytics Platform provided the opportunity to process hyperspectral reflectance data to classify crop diseases.

Actions (login required)

View Item