Deep Learning-Based Identification of Vegetable Species: System for Specially Abled Persons

Abstract

An automatic recognition system is developed to identify six different vegetable species, Solanum melongena (eggplant), Abelmoschus esculentus (okra), Solanum tuberosum (potato), Raphanus sativus (radish), Solanum lycopersicum (tomato), and Daucus carota (carrot). It helps the specially abled persons to buy these vegetables in the market independently in minimum time. The images of the vegetables were taken from various vegetable markets in Durgapur (West Bengal) at different times of the day. The proposed JUDVLP-BCRP: Vegdb.v1 dataset consists of 1800 images with 300 images in each species. After preprocessing the images, some popular deep learning architectures such as MobileNetV2, DenseNet121, and Xception with transfer learning techniques have been used to classify the species accordingly. Unfreezing the top 20% layers of the pre-trained networks and fine-tuning for 40 epochs, 100% accuracy is achieved using DenseNet121 and Xception network individually. The system produces a stable encouraging result in identifying vegetable species and leads to a viable system design.