Abstract:
In this paper, an attempt is made to accurately classify vegetables and fruits. This
classification uses a dataset of 6000 photos divided into 15 classes. Convolutional neural
networks, a type of deep learning algorithm, are the most efficient tool in the machine
learning field for classification issues. However, CNN requires huge datasets to perform
well in natural image classification tasks. We undertake an experiment on the performance
of CNN for vegetables and fruits categorization by building a CNN model from scratch.
Furthermore, multiple pre-trained CNN architectures using transfer learning are used to
evaluate accuracy to the standard CNN. This paper suggests the study of such standard
CNNs and their architectures (VGG16, InceptionV3, ResNet, etc.) to create up which
strategy would be most accurate and effective for fresh image datasets. Experimental
findings are presented for all proposed CNN designs. Additionally, a comparison is made
between constructed CNN models and pre-trained CNN architectures. Furthermore, the
paper demonstrates that by leveraging previous information collected from comparable
large-scale studies, the transfer learning technique outperforms classical CNN with a small
dataset. Another addition to this paper is that we created a dataset of 15 vegetables and
fruits groups, total 6000 images. This study examines the use of convolutional neural
networks (CNNs) to categorize and assess the quality of fruits and vegetables. A web
application using deep learning and computer vision, powered by a Python-based CNN
algorithm, marks a significant advancement in combining machine learning with
agricultural technology. The application accurately categorizes and grades quality. The
study used two datasets: 6000 photos in 15 categories for fruits and vegetable
categorization and 6000 images in 15 categories for quality assessment. By using images
data, this gadget can determine how much fruit or vegetables and fruits are rotting and how
fresh it is. Our aim is to reach all the general people for this reason we also build android
platform. Here user click one picture then our system automatically generates and show
the result We believe, it will be advantageous for all farmers, business owners, and common
people.
