Identification of worm-damaged chestnuts using impact acoustics and support vector machine

Abstract

Chestnut has both economically and nutritional values, and its production in the World is about 2 Mt. Turkey is one of the important chestnut producers with a production amount of about 60,000 t. Worm damage is one of the reasons which may reduce economical value of chestnut. Aim of this study was to reveal possibilities of distinguishing of worm-damaged chestnuts from healthy ones using impact acoustics and sound analysis methods. A Turkish local variety called ‘Osmanoglu’ was chosen for the study. A sound acquisition station was comprised, and acoustic emissions of worm-damaged and healthy nuts were acquired at a sampling quality of 192 kHz and 16 bit. Each sample was labelled according to worminess situation by shattering the nut after acoustic measurements. A band-pass filter between cutoff frequencies of 70 Hz and 100 kHz was designed and applied to sound samples to alleviate negative effects of unwanted noise. Various signal features such as variance, standard deviation, kurtosis, zero crossing rate, and spectral centroid were calculated. A relevant feature subset was determined using feature selection technics. An identification model was trained using Support Vector Machine and cross-validation rules. Performance of the classification system was measured on a test set. In this study, reporting the preliminary results of an ongoing and comprehensive research project, promising results were obtained for identification of wormdamaged chestnuts with proposed system.