Identify of Potato Genotypes by Combining Machine Vision Techniques and Neural Networks after the Storage Period

Potato storage is essential to ensure the continued supplying of potatoes to consumers and the potato processing industry. During storage, physiological changes and water loss lead to changes in color, shape, size, and texture of potatoes. Therefore, there is a need for a quick and accurate method to measure the quality of the product. In this study, machine vision and neural network methods were used in classification and modeling of two stored potato samples (Agria and Clone 8-397009)  under constant and variable conditions. Among 29 measured parameters relating to color, texture and morphological features of potato, some features were selected as the main parameters to monitor the chnges in product during storage period: Major Axis Length, Compactness, and area (morphological features), L* and b* (color features) and Average contrast (Ac) and Average gray level (Agl) (texture features). Among the training algorithms, Levenberg–Marquardt (LM)  training algorithm with the lowest root mean square error (RMSE=0.012) and the highest coefficient of determination (R2=95.01) were considered as an optimal model for classification of two samples stored in non-technical and technical storage. The accuracy of identification of the Agria genotype was 89.2% and 87.6%, and the accuracy of the genotype Clone 8-397009 was 92.4% and 90.3%, in non-technical and technical storage respectively.