Pomegranate Grading Using Different Feature Extraction and Machine Learning Approaches

In the preparation and packing of pomegranates, sorting is an essential stage. Pomegranates are currently hand sorted into quality groups. Manual sorting, however, takes a lot of time and is not precise. Furthermore, it isn't recommended to sort fruits of export quality manually. These problems can be resolved by automated sorting devices more precisely and effectively than by hand sorting. The objective of this study is to use digital image processing and machine intelligence to remove these problems with manual quality evaluation. As a result, the sorting procedure requires a computing facility with a machine vision system. This paper makes use of spatial domain features, frequency domain features and Histogram of Oriented Gradients (HOG) features to create datasets from the images by extracting their attributes, using which various Machine Learning classifiers are trained to classify pomegranate photos. It was found that HOG features outperformed spatial domain and frequency domain features, as an accuracy of 93.51% was obtained by training an ANN model with a dataset containing HOG features.