Intelligent Classification Of Charleston Gray Watermelon Variety Based On Fruit Ripeness Using Acoustic Signal Processing

According to the water crisis in the country and watermelon traditional irrigation process, it is possible to reduce planting and consequently increase the price of this product in the coming years, which highlights the necessity of indices for choosing high-quality watermelons. The purpose of this study is classification of the Charleston Gray watermelon variety into unripe, ripe and overripe classes, in this regard acoustic signals processing, data mining algorithms, and artificial intelligence techniques have been used for this purpose. After preparing the samples, firs through a capacitive microphone, signals acquired from different positions of watermelon using a solenoid and then, samples classes were determined by performing sensory evaluations. Signal processing techniques in time, frequency, processing domains and wavelet transformation were used for extraction of important features from acoustic signals of the watermelons, then some of the features that were significant in classification were selected using the t-test. Support Vector Machines and K Nearest Neighbor algorithms were used for sample classification. Totally 52% of the samples were classified correctly by experts. For metal ball, SVM algorithm with cubic kernel function resulted 78% correctly classification for acoustic signals obtained from middle position and Gaussian kernel function resulted 75% correctly classification for signals obtained from stem side position. K Nearest Neighbor algorithm equipped with the cosine distance resulted highest samples classification with a precision of 79% for the metal ball and the position of the stem side.