CHITTAGONG UNIVERSITY OF ENGINEERING & TECHNOLOGY (CUET)

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

CHITTAGONG – 4349
___________________________________________________________________________
(Project Proposal)

Application for the Approval of B.Sc. Engineering Thesis / Project

(Computer Science & Engineering)
Date: 25.02. 2020

1. Name of Student : Tanvir Anjum Ayman

Roll No. : 1504101 Session: 2018-2019

2. Present Address : Room No: A-305, Bangabandhu Hall, Chittagong

University of Engineering & Technology, Raozan,
Chittagong

3. Name of Supervisor : Sharmistha Chanda Tista

Designation : Lecturer
Department of Computer Science & Engineering (CSE)
Chittagong University of Engineering & Technology

4. Name of the Department : Computer Science & Engineering(CSE)

Program : B. Sc. Engineering

5. Date of First Enrollment

in the program : February 25, 2016

6. Tentative Title :

1
7. Introduction

Computer vision comes from modelling image processing using the techniques of machine
learning. Computer vision applies machine learning to recognize patterns for interpretation of
images. Much like the process of visual reasoning of human vision; we can distinguish between
objects, classify them, sort them according to their size, and so forth. Computer vision, like
image processing, takes images as input and gives output in the form of information on size,
color intensity etc. [1]

In other words, Computer vision is a field of artificial intelligence that trains computers to
interpret and understand the visual world. Using digital images from cameras and videos
and deep learning models, machines can accurately identify and classify objects and then react
to what they “see.”[2]

CNN is one of the classification methods based on machine learning theory. Compared with
other methods such as SVM, decision trees and Bayesian Networks, CNN has significant
advantages because of their high accuracy.

As mentioned earlier, computer vision technique can be used for detection and classification of
objects. So, can we use this to detect and classify whether a fruit of same shape like an apple or
a orange is fresh or rotten? Yes, this technique can be applied to classify among fresh and rotten
fruits.

Classification and manual selection of the fresh fruits from a huge amount of fruits is quite
impossible, time consuming and human error rate is high also. Using computer vision for this
task can reduce the human error rate as well as save the valuable time.

8. Background Study

There has been enormous research being carried out to assist in the quality assessment of fruits
and vegetables in agriculture or in the industry. Many applications requiring visual inspection
of fruits and vegetables such as tomatoes [3], dates [4] and mangoes [5] have used machine
vision based systems.

2
Here the classification of ripe and unripe tomatoes are done by CNN but using a very small
dataset [3]. But for getting a higher accuracy, we have to use a comparatively big dataset for
CNN. And this is also used for only a single type of fruit, it can’t be applied for various fruits
of same shape.

The surface bruise of the fruits are detected here but it didn’t classify the fruit [6]. It just detects
the bruise on the surface but as we are trying to use our model for industry, we have to classify
the state of fruits whether to select them or not.

Here a bruise detection system is developed using thermal images [7] but it is expensive to buy
a thermal camera and collect some thermal image. Besides it is not classifying the fruits that
can be used in a industry.

In this work, surface bruise of fruits are detected using Biospeckle technique but again, it is not
classifying the fruits [8]. So it can’t be used in a industry for selection the fruits.

Both SVM and CNN are used as classifier but the accuracy level of the algorithm depends on
the size of dataset. SVM is a popular method for classification of object using computer vision
but the limitation of SVM is that it works perfectly only for a comparatively small dataset. For
a huge dataset of various states of fruits or multiple fruits of same shape like apples and oranges,
CNN is the best method to apply for classifying among them.

A fruit recognition system is implemented using SVM to recognize the fruit but it is not the
classification among fresh and rotten fruit [9], it just show the names of the fruits.

A fruit classifier system is developed using image processing and computer vision. The
classifier system was implemented by using Fuzzy Logic [10]. But the accuracy is level of CNN
is high over Fuzzy Logic. So use of CNN for a huge amount of data will be more logical than
using of Fuzzy Logic.

3
9. Objective with Specific Aims and Possible Outcomes

The main objective, motivation and possible outcomes of my work is illustrated in the
following:

9.1 Motivation
 Implementing a computer vision system to detect whether an apple or a orange
is defective or non-defective.
 Making the system work accurately for a large amount of dataset.

9.2 Objective
 To develop a system that can classify the apple or orange just using simple
images taken by a normal digital camera.
 To maximize to accuracy rate of the output.

9.3 Possible outcome

 The system may find a solution for detecting a fresh fruit.
 The system may become more cost efficient and accurate.

4
10. Outline of Methodology/Experimental Design

Fresh and rotten fruit’s image collection

Image Pre processing

Suitable CNN model selection for dataset

Train the model using the dataset

Evaluation of the model

Hyper Parameter Tuning for maximum

accuracy

Classification of the fruits

Defective Non-
Defective

Figure 01: Steps for proposed methodology

5
The proposed methodology is described here:
 The first step is to collect dataset of fresh and rotten fruits of same shape like apples and
oranges.
 Then we need to get into the step called image preprocessing. Resizing of sample images
may be needed and might need to reduce the noises from the images. This reduction of
noise can be done using “Gausian Filter”.
 We are going to use Convolutional Neural Networks (CNN) for classification. Selection
of a CNN model is one of the most important task in our project. There are various CNN
models like Inception, Xception and VGG16 etc. We need to try these CNN
architectures on our dataset and see which one suits our model the best and gives us the
most accuracy.
 Training the model with our collected dataset.
 Evaluation of the model upon it’s Accuracy, Precision, Recall and F1 Score.
 Hyper Parameter tuning of the model will be needed for getting a higher accuracy rate.
We need to check the performance of the model by changing the nodes and layers of
the CNN.
 The final step is the classification steps. If we train our model using collected dataset
properly, then it will be automatically capable of classifying an unseen image of apple
or orange whether it is fresh or rotten.

11. Resources Required Completing the Work

The resources that are required by our proposed system are specified below:
 A personal Computer for creating the model.
 A digital camera for capturing images of fruits.
 A large amount of dataset to train the CNN model.

6
12. Cost Estimate

The costs that will occur to implement out proposed system are given below:

a) Typing, Drafting, Binding :

Internet Browsing and Downloads Tk. 2000.00
Transportation Cost for collecting Samples Tk. 500.00
Printing & Binding Tk. 200.00
Miscellaneous Tk. 200.00
Tk. 2900.00

13. References

1. “What Is The Difference Between Computer Vision And Image ...” [Online].
Available: https://analyticsindiamag.com/what-is-the-difference-between-computer-
vision-and-image-processing/. [Accessed: 24-Feb-2020].
2. “Computer Vision: What it is and why it matters,” SAS. [Online]. Available:
https://www.sas.com/en_us/insights/analytics/computer-vision.html. [Accessed: 24-
Feb-2020].
3. “Figure 2f from: Irimia R, Gottschling M (2016) Taxonomic revision of Rochefortia
Sw. (Ehretiaceae, Boraginales). Biodiversity Data Journal 4: e7720.
https://doi.org/10.3897/BDJ.4.e7720.”
4. M. Arakeri and Lakshmana, “Computer Vision Based Fruit Grading System for
Quality Evaluation of Tomato in Agriculture industry,” Procedia Computer Science,
vol. 79, pp. 426–433, 2016.
5. D. Unay, B. Gosselin, O. Kleynen, V. Leemans, M.-F. Destain, and O. Debeir,
“Automatic grading of Bi-colored apples by multispectral machine vision,” Computers
and Electronics in Agriculture, vol. 75, no. 1, pp. 204–212, 2011.
6. C. S. Nandi, B. Tudu, and C. Koley, “Machine Vision Based Techniques for
Automatic Mango Fruit Sorting and Grading Based on Maturity Level and
Size,” Sensing Technology: Current Status and Future Trends II Smart Sensors,
Measurement and Instrumentation, pp. 27–46, Oct. 2013.

7
 7. M. Satone, S. Diwakar, and V. Joshi, “Automatic Bruise Detection in Fruits Using
Thermal Images,” International Journal of Advanced Research in Computer Science
and Software Engineering, vol. 7, no. 5, pp. 727–732, 2017.

8. F. Vega and M. C. Torres, “Automatic detection of bruises in fruit using Biospeckle

techniques,” Symposium of Signals, Images and Artificial Vision - 2013: STSIVA - 2013,
2013.
9. Y. Zhang and L. Wu, “Classification of Fruits Using Computer Vision and a Multiclass
Support Vector Machine,” Sensors, vol. 12, no. 9, pp. 12489–12505, 2012.
10. Seema, A. Kumar, and G. Gill, “Automatic Fruit Grading and Classification System
Using Computer Vision: A Review,” 2015 Second International Conference on
Advances in Computing and Communication Engineering, 2015.

14. CSE Undergraduate Student (CUGS) Committee reference:

Meeting No: Resolution No: Date:

15. Number of Under-Graduate Student(s) working with the Supervisor at

Present:

Signature of the student

________________________
Signature of the Supervisor

Signature of the Head of the Department