A Deep Learning-Based Fruit Quality Assessment System

S. Harini, Parijat Deshpande, Jayita Dutta, Beena Rai

Abstract: Fresh and ideally ripe fruits contribute to nutrition, dietary diversity, and consequently to consumer health and are therefore imperative to ensure their quality. Fruit quality check is conventionally conducted by visual inspection performed by humans, which is time-consuming and may also become inconsistent. Cost-effective and accurate sorting can be achieved by automated sorting. We have proposed a deep learning-based technique that detects the quality of fruits with an accuracy of 89.5%. This process has two major aspects, first is the fruit recognition task, subsequently followed by detecting the quality of the fruit, which is a good/bad classification of the recognized fruit. We have curated a database of the 12 most commonly used fruits in India. However, to operate for fruits outside our database, we have employed a transfer learning approach. To facilitate this, we have used the Siamese network with triplet loss, which finds a similarity between the fruit and gives us the closest match. For each fruit, based on the availability of data, we have developed models by using a deep learning approach and machine learning models on extracting image-based features. This technique presents an automatic approach for the detection of spoilage in fruits efficiently.

External IDs:doi:10.1007/978-3-030-75315-3_22