• Tidak ada hasil yang ditemukan

fresh and rotten fruits classification using vgg16 and resnet50

N/A
N/A
Protected

Academic year: 2023

Membagikan "fresh and rotten fruits classification using vgg16 and resnet50"

Copied!
36
0
0

Teks penuh

This project/internship titled “CLASSIFICATION OF FRESH AND ROTTEN FRUIT USING VGG16 ALGORITHM AND RESNET50” submitted by Saidur Rahman Saeed, ID No, to Department of Computer Science and Engineering, Daffodil International University has been accepted as satisfactory for partial completion requirements for the B.Sc. We hereby declare that this thesis has been produced by us under the guidance of Nusrat Jahan, Senior Lecturer, CSE Department and co-supervised by Ms. We also declare that neither this thesis nor any part of this thesis has been submitted elsewhere for the award of any degree or diploma.

First of all, we would like to thank the Almighty Allah for the immense blessing that enables us to complete the final thesis successfully. We are truly grateful and express our serious indebtedness to Nusrat Jahan, Assistant Professor, Department of CSE Daffodil International University, Dhaka, Bangladesh. Touhid Bhuiyan, Professor and Head, Department of CSE, for his kind help in completing our thesis and also to other faculty members and the staff of CSE Department of Daffodil International University.

We would like to express our gratitude to the fellow student from Daffodil International University, who participated in this discussion during the completion of this work. We would like to express our enormous gratitude to the people who provided us with the necessary raw data to make our work possible. Here we use the red fruit and clean fruit CNN architecture for vgg16 and resnet50.

With a large farm and a savory factory to observe rotten fruit, this gadget can make our artwork easier.

Introduction

  • Introduction
  • Motivation
  • Rationale of the Study
  • Research Questions
  • Expected Outcome
  • Project Management and Finance
  • Report Layout

Vgg16 and resnet50 also known as image classifier can be used in many types of fields to solve different problems like - find cancer cells or can detect disease in leaves. We also note that the latest advanced tensorflow library improves the way vgg16 and resnet50 are used. AI is now ubiquitous and the problem can even be shown via the categorization of fresh produce.

We can use the latest advances in artificial intelligence and image processing to build these solutions right from the start. This problem is minimized and a system is created that allows users to quickly scan and retrieve findings. To prevent this problem, we tried to develop an intelligence system based on deep learning technology.

To create these solutions from the ground up, we can take advantage of the latest advances in AI and image processing. This difficulty is minimized and a method is created to enable users to scan quickly and get the results. To prevent this dilemma, we have sought to create a deep learning technological intelligence system.

Give an overview of this study. The first analysis is a key step in this first part

Background Study

  • Related Works
  • Comparative Analysis and Summary
  • Scope of the Problem
  • Challenges

The accuracy of the categorization using machine learning is mostly dependent on the drawn features and the features chosen for the machine learning method. It is based on surfaces approximated by convex polyhedra with quantized face lines, where each side of the polyhedra represents a single descriptor. A variation that uses color and a variant that does not are tested for two variants of both CTI and SHOT descriptors.

Compared with existing fruit recognition methods, the main benefit of the suggested expert system is its computational efficiency, which is important for its purpose - autonomous fruit picker. Detection was done for rotten or fresh apple based on fruit skin defects introduced by Roy et.al.[5] the decomposed section in the RGB image of the apple is bisected in the deep learning architecture. Daffodil International University 6 validation accuracy of 97.46 percent and 97.54 percent respectively, UNet as the base architecture achieved an accuracy of 95.36 percent were the suggested En-UNet model generating improved results than UNet.

6] analyzed blueberry fruit characteristics including cluster compactness, fruit maturity and berry quantity, a deep segmentation study was implemented to assist blueberry breeders. To detect symptoms and track disease locations on leaves, we used visualization methods to evaluate the given deep pattern. CNN is trained most accurately, with an accuracy of 99.55% or higher, among other image classification algorithms.

Through our study, we have found that Convolutional Neural Network does such a good job. Judges can simply fulfill their responsibilities by using our technology in many parts of this nation's wholesale marketplaces where red fruit is sold. For correct determination of our dataset or future changes, we have used powerful ML and image processing technologies.

Since Deep Learning requires a large amount of data from multiple marketplaces and from wholesale locations, we were unable to collect that much data. This is why we sourced our data from the Kaggle website from a third party. Even if we try to include part of the data obtained in the field in our data set.

Research Methodology

  • Research Subject and Instrumentation
  • Data Collection Procedure
  • Statistical Analysis
  • Applied Mechanism
  • Implementation Requirements
  • Experimental Setup
  • Experimental Result and Analysis

Daffodil International University 10 Because we were not allowed to touch the fruit sellers except to sell their produce. Therefore, we collected our data from a third-party site, Kaggle. Even if we tried to put some of all the data obtained from the field into our database.

In this picture, we can see all the rotten and fresh fruit among all six classes. Daffodil International University 14 Neural Networks That Use Jump Connections or Shortcuts to Jump Over Specific Levels. There are three different layers of each convolution block and three levels of each identity block.[12]

Daffodil International University 15 VGG16 is a CNN architecture for neural networks used in the 2014 ILSVR(Imagenet) competition (CNN). The most distinctive feature of VGG16 is that they focused on 3x3 step 1 filter layers and always used 2x2 step maxpool and padding filters rather than having a large number of hyper parameters. For project implementation we used different machine learning library all the version are given below.

It is highly recommended for AI-based work programming languages ​​and is very popular among new generation programmers because it is easy to learn. The Implications section should be designed so that the findings are communicated without your knowledge or research. As usual, an experiment consists of the systematic manipulation and effect of one or more independent variables on specific dependent variables[11].

A machine learning experiment therefore requires much more work than just one training under different settings.

Figure 3.2:  Data augmentation and background removal
Figure 3.2: Data augmentation and background removal

VGG16

Discussion

Summary, Conclusion, Recommendation and Implication for Future Research

Summary of the Study

Conclusion

Implication for Future Study

APPENDIX

PLAGIARISM REPORT

Gambar

Figure 3.1: Methodology diagram
Figure 3.2:  Data augmentation and background removal
Figure 3.3:  Dataset representations
Figure 3.4:  Dataset representations
+7

Referensi

Dokumen terkait

ZINCROMATECAT HITAM SEBAGAI TIANG RAILING KAYU SOLID BENGKIRAI T= ±20mm SEBAGAI HANDRAIL PARAPET / TANGGULAN PLAT LANTAI BETON JEMBATAN T=130MM GARIS PROYEKSI PLAT LANTAIAREA AOS