Feature detection and classification of pistachio by using image processing

Quality in agricultural product marketing is one of the important factors. The main purpose of this study is to classify it according to its colors, size and quality in order to make quality classification of pistachios. The standardized products will have advantages in price because they will be easier to store and process. Traditionally, pistachios are classified via visual inspection of workers, manually. As a result, the classification process is subjected to poor efficiency in terms of time and cost. The most effective method used in grading machines today is image processing. The scope of the research is to create a fast and real time classification system with high accuracy for pistachios and unwanted materials. The study can be divided into two main parts, segmentation and classification. For image segmentation, local thresholding with average filter were used to isolate the background from the objects in the input images. For image classification, to produce excellent and fast results deep convolutional neural network is used. Deep learning net was trained by 2750 images of 11 types of classification which are fresh red pistachio, fresh green pistachio, split pistachio, non-split pistachio, peeled pistachios kernels, stones, loose kernels, leaves, branch pieces, pistachios shell. The system was tested with a set of real-time images as well as images stored in the computer, by making a segmentation to the image to make the objects separated in individual images, and then classify each object image by the network. High accuracy classification results are obtained. The features of the objects such as area, centroid, length, and width are also calculated.