Enhanced Pomegranate Grading Using InceptionResNetV2 with Transformer Integration

Sorting pomegranates based on quality grades is a crucial stage in their export preparation and packing process. Traditionally reliant on manual sorting methods which are time-consuming, prone to inaccuracies and incur very high costs due to the need for a large workforce prove to be very inefficient. Moreover, manual grading of export-quality of foods poses additional challenges. This paper addresses such issues by proposing an automated system employing modern digital image processing and deep learning techniques for pomegranate quality grading and classification. Initially, pomegranate images are subjected to several augmentation techniques like extracting HSV color features, Local Binary Pattern features, horizontal and vertical transformations, etc. Then a novel classification model having InceptionResNetv2 convolutional network as its backbone and integrated with optimally placed transformer blocks is proposed. The augmented dataset is utilized to train six state-of-art deep learning models namely, Vision transformers, XceptionNet, EfficientNetV2B0, DenseNet121, InceptionResNetV2 and NasNet-Large. It is observed that the proposed InceptionResNetV2 model containing transformer blocks exhibited superior performance, achieving an accuracy of 90.56%. This underscores the ability of the proposed deep learning model in automating quality evaluation and sorting procedures for pomegranates.