Classification of Hazelnut Varieties Based on Shell Characteristics: An Image Processing and Machine Learning Approach

Abstract

The use of hazelnut varieties in different applications directly affects processing efficiency and product quality, and this is of significant importance in the food and agricultural industries. Currently, however, there are no machines designed specifically to differentiate hazelnut varieties based on their shell characteristics; instead, existing machinery predominantly separates hazelnuts into categories such as full, empty, or broken. The ability to distinguish between hazelnut varieties is essential for optimizing processing workflows and improving the quality of the final product. In this context, the proposed study seeks to classify hazelnuts by analyzing their physical attributes, including color and texture, to determine their suitability for specific applications. After capturing images of hazelnut shells on the conveyor belt, preliminary image processing steps such as cropping, background removal, and measurement standardization were performed. Subsequently, feature extraction was carried out on the processed images. To ensure high classification accuracy, the feature extraction methods were designed to capture variations in size, color, and texture. The techniques employed for feature extraction included Dimension, Color, Haralick, Local Binary Patterns (LBP), and Histogram of Oriented Gradients (HOG). These extracted features were then applied to various classifiers, including K-Nearest Neighbor (k-NN), Decision Tree (DT), Support Vector Machine (SVM), Feedforward Neural Networks (FFNN), Recurrent Neural Networks (RNN), and Cascade Forward Neural Network (CFNN). The classification results revealed that the CFNN classifier outperformed the others. In conclusion, the machine learning-based system developed for classifying hazelnuts by their outer shell is poised to significantly enhance quality in industrial processes while introducing a major innovation by elevating efficiency to a new level.

Affiliated Institutions

• Düzce Üniversitesi TR

Related Publications