Classification of Hazelnut Kernels by Impact Acoustics

An automated hazelnut classification system is developed using sub-band information of impact acoustic signal taken from hazelnut kernels. It is observed that hazelnuts emit different acoustic signals when they impact on a metal plate. Impact acoustic signal of nuts are decomposed with undecimated wavelet transform. Each sub-band is divided into non-overlapping time segments and a feature vector is constructed using the energy values calculated for each segment. A maximum likelihood classifier is used to classify the hazelnut kernels into three groups: i) empty or undeveloped kernels ii) fully developed nuts with regular shell and iii) fully developed nuts with cracked shell. A two stage classification scheme is developed in this study. Hazelnuts kernels are first classified into two classes i) empty or undeveloped and ii or iii) fully developed classes with 98.20% accuracy. The fully developed hazelnuts are then classified into cracked shell and regular shell classes at the second stage. The developed algorithm detected 95.26% of cracked shells at the second stage.