Deep learning based mobile application for varietal identification and ripeness grading of traditional Indian banana varieties

Abstract

India is the largest producer of bananas, con-tributing to 1/5th of the world production. Traditional Indian banana varieties have their own health benefits and consumer preferences. Grading of ripening stages is essential for quality check during handling and export as well as domestic market consumer acceptance. The proposed work is aimed to develop a smart phone based mobile application capable of identification of various traditional Indian banana varieties along with the grading of its ripening stages. Image processing is the better choice for identification of banana varieties and the determination of colour dependent ripening stages. Combining multiple aspects of deep learning inclusive of Convolution neural network (CNN) and eXtreme Gradient Boosting (XGboost) algorithm (CNN-XGBoost), a varietal identification and ripeness grading model is developed. Images of the banana fruits are fed to the network, where the CNN acts as the trainable feature extractor of the images and XGboost in the last layer of the CNN acts as the identifier of variety and ripening stage. The identification accuracy of the proposed model is 95 % which is higher than other techniques such as Gaussian Naive Bayes classifier (66 %), support vector classifier (83.5 %) and k-nearest neighbourhood algorithm (90 %). The developed model is deployed into smart phone based mobile application to facilitate non-invasive varietal identification of banana fruits. The developed app is capable to identify various unique Indian traditional banana varieties and could provide detailed insights on the ripening stages. The computational complexity of the developed model is also lesser which reduces the computational burden of the mobile application. The developed mobile application could be of great help to the consumers to decide upon the right variety and the optimal stage of ripening to be consumed for their dietary requirement.