Application of Image Processing and Artificial Neural Networks for Detection of Adulteration in Persian Black Cumin

Persian cumin plays a significant role in both exports and domestic industries of Iran. Today, due to the widespread availability of counterfeit cumin in the market, identifying authentic Persian cumin from its counterfeit counterparts has become increasingly essential. Among the various criteria for identification, color and texture indices are particularly notable. Traditional methods, such as manual and visual inspections, are not only time-consuming but also prone to a high degree of human error. In this study, in order to propose a new, precise, and rapid method, machine vision technology was utilized to extract the color and texture features of cumin from its images. Subsequently, a multi-layer perceptron artificial neural network with a backpropagation algorithm and a hidden layer was employed, evaluating different neurons in this layer to perform the process of distinguishing authentic Persian cumin from counterfeit varieties in the market. In this study, five samples of authentic Persian cumin and four samples of counterfeit cumin, were evaluated. The results showed that the highest average classification and identification accuracy of authentic cumin from counterfeit cumin, using a neural network with one hidden layer and employing a sigmoid transfer function in this layer and a linear function in the output layer with the Levenberg-Marquardt learning algorithm, were 93.51% for color features, 95.86% for texture features, and 95.59% for the combination of these two features (color-texture). However, the findings showed that machine vision technology and artificial neural networks have a high capability in accurately identifying authentic Persian cumin from counterfeit samples.