Pistachio Classification Based on Acoustic Systems and Machine Learning

Abstract

An acoustic emission and machine learning based pistachio classification system has been developed. This system performs feature extraction using Mel frequency cepstral coefficients (MFCC) and classification using support vector machine (SVM). This study revealed that when closed-shelled pistachios hit a steel plate, they have different frequency components compared to open-shelled pistachios. The audio signals of the samples selected for the classification process were recorded using a high sensitivity carbon microphone and MATLAB Analog Input Recorder. These recorded sounds were processed by applying a hamming window to remove ambient noise and make them more clearly analyzable.

MFCC is one of the leading methods used to extract features representing audio signals. In this study, MFCCs are used to distinguish between open and closed shelled pistachios. By analyzing the frequency components in audio signals, this feature extraction method helps to identify the distinctive features of the signals. These features are given as input to a support vector machine algorithm called FITCSVM for classification. FITCSVM is an algorithm that can perform one-class and two-class (binary) classification on low or medium-sized prediction datasets.

In this study, open and closed shelled pistachios were classified with high accuracy. The results show that the acoustic emission and machine learning based classification system has the potential to be used in the pistachio industry. In particular, distinguishing between open and closed shelled pistachios is of great importance for increasing product quality and improving processing processes.

As a result, this research shows that MFCC and SVM algorithms can be used effectively in pistachio classification. The sound signals obtained by acoustic emission method were analyzed with MFCC to extract the features required for classification and FITCSVM was used to classify with high accuracy. Such innovative approaches can contribute to the development of more efficient and effective methods for processing agricultural products.