Grain Surface Classification via Machine Learning Methods

TL;DR

This paper presents a machine learning framework that classifies grain surface types using radar backscatter signals processed with various transforms and feature extraction methods, achieving high accuracy with STFT+GLCM+SVM.

Contribution

It introduces a novel combination of signal transforms and texture feature extraction techniques for grain surface classification using radar data.

Findings

STFT+GLCM+SVM achieved the highest classification accuracy.

The framework effectively differentiates grain surface types.

Cross-validation confirms the robustness of the method.

Abstract

In this study, radar signals were analyzed to classify grain surface types by using machine learning methods. Radar backscatter signals were recorded using a vector network analyzer between 18-40 GHz. A total of 5681 measurements of A scan signals were collected. The proposed method framework consists of two parts. First Order Statistical features are obtained by applying Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT) on backscatter signals in the first part of the framework. Classification process of these features was carried out with Support Vector Machine (SVM). In the second part of the proposed framework, two dimensional matrices in complex form were obtained by applying Short Time Fourier Transform (STFT) on the signals. Gray-Level Co-Occurrence Matrix (GLCM) and Gray-Level Run-Length Matrix (GLRLM) were obtained and feature extraction process was completed. Classification process was carried out with DVM. 10-k cross validation was applied. The highest performance was achieved with STFT+GLCM+SVM.