Optimizing a combination of texture features with partial swarm optimizer method for bulk raisin classification

Grapes are one of the important agricultural products that are consumed both fresh and dried. By drying grapes in different conditions, various raisins are produced. After raisin production in the field, it is delivered to raisin production factories in order to wash and remove bad grains and remove thorns and weeds. Pricing and determining the quality of bulk raisins at this stage is one of the most important challenges between the seller and the buyer, who is the factory owner. In this research, using the machine vision method, 15 different classes of bulk raisins were investigated based on the composition of good and bad seeds and dry wood. The texture features of the images were used for classification, and the best combination of image texture extraction algorithms was evaluated using the particle swarm optimization (PSO) method. Three different classifier by name support vector machine (SVM), linear discriminate analysis (LDA) and K-nearest neighborhood were used for modeling. The results showed that the combination of several texture feature extraction methods using PSO improves the classification accuracy for all classifiers. The best results were achieved using SVM and LDA modeling as 99.33% and 99.73%, respectively. Since the number of algorithms used in LDA modeling was less than SVM, so the LDA model was selected as a best model. Results showed that the machine vision system can be used successfully for quality evaluation of bulk raisin pricing.