Seybold Report ISSN: 1533-9211
Dr Vijay Prakash Singh1, Dr S N Talbar2, Deepali M Bongulwar3
1Research Guide ,Department of Electronics and Communication Engineering, School of Engineering, Sri Satya Sai University of Technology and Medical Sciences, Sehore, MP, India
2Research Co-Guide Department of Electronics and Telecommunication Engineering, SGGSIE&T, SRTMU, Nanded (M.S), India
3Research Scholar, Department of Electronics and Communication Engineering, School of Engineering, Sri Satya Sai University of Technology and Medical Sciences, Sehore, MP, India
1vijaybhabha12@gmail.com, 2sntalbar@sggs.ac.in, 3ddeepali2006@yahoo.com
Vol 17, No 10 ( 2022 ) | DOI: 10.5281/zenodo.7259996 | Licensing: CC 4.0 | Pg no:2059-2072 | Published on: 28-10-2022
Abstract
In the farming business, inspecting the quality of the fruit is a crucial responsibility. The study offers a thorough examination of several fruit images using deep learning for freshness evaluation. Freshness is the most important fruit quality indicator, as it directly affects consumers' physical well-being and purchase intentions. Additionally, it is a significant market pricing component. Therefore, it is important to assess how fresh the fruit is. Convolutional neural networks (CNNs) have illustrated their effectiveness in a number of agricultural applications. We applied the idea of transfer learning to the evaluation of fruit quality. The idea of reusing a previously trained CNN model to address a new issue without the requirement for extensive training datasets is referred to as transfer learning. The present study aims to compare the performance of four distinct CNN architectures, AlexNet, GoogleNet, VGG-19, and ResNet-50, to find the fastest method for separating fresh from rotting fruit. The performance of CNN architectures in the fruit quality rating system is based on accuracy and processing speed. "Fruits Fresh and Rotten," a publicly accessible dataset, is used for experimentation. There are six different fruit classifications, including fresh/rotten oranges, fresh/rotten bananas, and fresh/rotten apples. 7,630 training images, 3,271 validation images, and 2,698 testing images are available. AlexNet outperforms the other three models with a validation accuracy of 99.48% and a training duration of 5 m 30 s. As a result, AlexNet excels in both time and accuracy. Additionally, various metrics like Recall, Precision, and F1 score are employed to assess the efficiency of the models. We come to the conclusion that producers can improve fruit sorting by using the AlexNet model.
Keywords:
Convolutional Neural Network, Deep Learning, Transfer Learning, Fruits, Fruits Fresh and Rotten dataset, Agriculture, Fruit quality