Mango Varieties Classification using EfficientNetB2 Transfer Learning Model

TL;DR: The research employed the EfficientNetB2 model in a simulation, using a constant learning rate of 0.0001 for 18 epochs and a batch size 32, and reported the model's accuracy to be 98%.

Abstract: Mango is widely acknowledged as a popular tropical fruit appreciated for its delightful and indulgent flavor. It is commonly noted that mango holds a special significance as the national fruit of India. It offers diverse options, highlighting unique qualities, flavor profiles, and visual attractiveness. In recent years, deep learning-based applications have demonstrated impressive accuracy in image-based classification. The research paper comprehensively analyzes the diverse factors considered during the study. The research employed the EfficientNetB2 model in a simulation, using a constant learning rate of 0.0001 for 18 epochs and a batch size 32. The dataset consists of 1600 images that have been organized into eight distinct categories such as Anwar Ratool, Chaunsa (Black), Chaunsa (Summer Bahisht), Chaunsa (White), Dosehri, Fajri, Langra, and Sindhri. Every class represents a unique variety of mango. The dataset is divided into three distinct types of data: training data, testing data, and validation data. The training data comprises 1280 images, while the testing and validation data includes 160. The procedure utilized the Adam optimizer. The model's accuracy is assessed using precision, recall, and F1-score. The overall accuracy is reported to be 98%. Mango varieties can enhance culinary experiences, promote agricultural sustainability, and contribute to the long-term conservation of this cherished fruit.