Classification of Philippine Soybean Variety Using Image Processing Technique and Machine Learning Method
John Christian S. Diamante, Isaac Angelo M. Dioses, Jesusimo L. Dioses, John Paul Q. Tomas
Abstract
Together with dairy and wheat, soybeans are a major agricultural import into the Philippines. Historically, imports accounted for 99% of the country's supply of soybeans from 1995 to 2014, with local manufacturers making up the remaining 1%. The emergence of novel technologies has enabled the classification of diverse agricultural commodities, such as soybean cultivars, by merging computer vision and machine learning methodologies that utilize edge detection algorithms. Precise categorization of seed variants is essential for farmers and seed manufacturers to maintain variety purity, which in turn affects crop productivity and the quality of soybeans provided to nearby retailers. Morphological features were retrieved from pre-processed soybean pictures using the regionprops function, which made use of edge detection methods. After the extraction of features, the data was subjected to pre-processing and machine learning analysis. The KNearest Neighbors (KNN) model classified the data using Euclidean distance. A 75:25 split of the dataset was made into training and testing subsets, with five neighbors being used for categorization. The CL1 and PSB SY2 soybean varieties were classified by the KNN model with an accuracy rate of 85%, indicating a first step in variety classification research.