Automatic pomegranate quality classification using machine learning and transfer learning

The pomegranate fruit is well known for its phytochemical and therapeutic potential globally. However, the fruit is more prone to number of diseases that are often impacted by the different meteorological condition such as temperature fluctuation, rainfall and humidity, which affect the quality and profitability of the fruit. The pomegranate fruit is more vulnerable to Bacterial blight, Anthracnose, Alternaria fruit and Cercospora fruit spot disease, which can be automatically identified by using advance machine learning techniques. The present study proposed an automatic technique focused on machine and transfer learning approach for categorization of healthy fruit and those affected by Bacterial infection, blight, Anthracnose, Cercospora spot to enable the early detection of fruit health. The datasets consist of a total 5099 images and categorized (into 5 categories) according to disease incidence. The features and classifiers used in the study are Gray Level Co-occurrence Matrix (GLCM), k-Nearest Neighbors (KNN), Support Vector Machine (SVM), Histogram of Oriented Gradients (HOG), Logistic Regression, Random Forest (RF), Decision Tree (DT), and Naive Bayes (NB) respectively. The four pre-trained models (VGG16, ResNet152V2, InceptionV3, and DenseNet201) were used for pomegranate classification. The logistic regression performs better with 88.53% accuracy as compare to other machine learning algorithms. Again, in transfer learning, DenseNet201 showed higher accuracy (96.16%) accuracy for the classification of healthy vs diseased fruit. The research offers an effective solution for the automatic detection and classification of pomegranate fruit, thereby play an important role at farm as well as industrial level.