การทำนายระดับความสุกของทุเรียนจากเสียงเคาะ

Abstract

In this independent research endeavor, a predictive model for assessing the ripeness levels of durian fruits based on sound tapping was developed. The ripeness levels were categorized into three stages: raw, unripe and ripe. To create a highly accurate predictive model, the researchers conducted a comparative study in the data feature extraction phase from the tapping sound. Three methods were explored and compared: Mel-Spectrogram, Short-time Fourier Transform (STFT), and Mel-Frequency Cepstral Coefficients (MFCC). The researchers opted to use a Convolutional Neural Network (CNN) algorithm in developing the model. To evaluate the performance of the developed model, accuracy was measured by comparing the data extracted using all three feature extraction methods. It was found that the Short-time Fourier Transform method yielded the highest accuracy, achieving 99% accuracy when tested with previously unseen data and 99% accuracy on real-world data.