|
|
 |
| |
|
|
|
Deep Transfer Learning and Image Segmentation for Fruit Recognition |
|
|
|
PP: 567-578 |
|
|
|
doi:10.18576/amis/170405
|
|
|
|
Author(s) |
|
|
|
Ahmad Odat,
Mohammad Alodat,
|
|
|
|
Abstract |
|
|
| This paper is structured in such a way that you can able to develop the fruit recognition web app from scratch with Machine Learning & Flask. The primary objective of this research is to identify plant products in order to estimate their pricing without having to rely on recollection to enabling retail market administration and enhancing the shopping experience for customers. Methods and techniques used in this research was machine learning, deep learning, data analysis, processing, and interpretation, depending on Experimental Methods, Survey Research, Meta-Analysis, and Data Mining to provide a comprehensive and robust analysis of the data. Convolutional neural networks (CNNs) based on Deep Transfer Learning (DTL) and using previously trained weights are referred to as pre-trained networks. Pre-training, feature extraction, and fine-tuning steps of pre-trained networks involve plant product recognition. The purpose of this study is to enhance plant product packing services classification during take weight in conventional packaged vending systems. In order to assist and reduce efforts and time for customers instead of waiting while choosing fruits and vegetables that have been sold. Pre-trained ResNet50 network model that has been used in our model has an accuracy that is comparable to actual detection performance. We are using a cloud computing platform called Google Cloud Platform to deploy our model for simple and adaptable market access. This article identifies the barcode of the product accuracy up to 100 % and identifies also the type of fruits and vegetables far away from guesswork or the seller’s experience by object detection. Pre-trained ResNet50 network model achieved the best results at stage fine-tuning by 100 % from the first time.
|
|
|
|
|
|
|
|
