Design and development of a machine vision system to determine the apparent apple imperfections

The machine vision system is one of the newest systems for identifying the quality of agricultural products. Apple is one of the fruits whose apparent quality used by customer to select this product at the market. Automatic detection of faulty apples through the machine's visual system is difficult due to the non-uniform of distribution of on the surface and the similarity between actual defects with the color changes of the fruit peel. For this purpose, in this study, a new method for detecting apparent defects of apple using a machine vision system with a combination of auto-correction of light was presented. In order to classify the samples, the histogram of the taken images was corrected based on the RGB method, then three-color and 11 textural features were extracted. Based on the results of the feature selection, the best features for the highest accuracy in the classification were respectively entropy, energy, correlation and local smooth. Finally, for categorization of data, two classifiers namely relevance vector machine (RVM) and support vector machine (SVM) were used. Based on the classification results, the accuracy of the RVM classification was 95% in the sound group, 82% in the unsound group and 88.5% in for total accuracy; but the accuracy of the SVM classification was 100% in the sound group, 94.23% in the unsound group and 97.11% for total accuracy. Therefore, in order to detect sound samples from unsound ones the SVM classification is more suitable than the RVM, due to the greater accuracy and less error.