Date Fruit Classification using K-Nearest Neighbor with Principal Component Analysis and Binary Particle Swarm Optimization

Abstract

Various cultivars of date fruits distributed throughout exhibit diverse complexity and unique attributes, including color, flavor, shape, and texture. These distinctive characteristics and appearance occasionally lack variability in date fruits, since various kinds of date fruit may have subtle differences in color, shape, and texture. To overcome the difficulty of sorting and classifying multiple types of date fruit, a classification model was developed to categorize date fruit according to their visual appearances and digital characteristics. This study proposes a classification system that categorizes date fruit into five distinct types. The system achieves this by extracting features related to date fruit images' color, shape, and texture. Specifically, color moments, HOG descriptors, and circularity are used for feature extraction. The resulting high-quality training data is then used to train a K-Nearest-Neighbor (KNN) classifier. Considering the parameters applied to develop the proposed classification model is essential. Therefore, the proposed KNN model will be optimized by Principal Component Analysis (PCA) and Binary Particle Swarm Optimization (BPSO). PCA is employed for dimensionality reduction, whereas BPSO is implemented to discover the optimal neighbors. The experimental results demonstrated that the classification model achieved an accuracy of 93.85%, a considerable improvement of 12% over barebone KNN.