Bangladeshi Flower Identification Using Computer

Vision and Machine Learning Techniques

Prepared for
Thesis defense committee
Department of Computer Science and Engineering
IUBAT- International University of Business Agriculture and Technology

Supervisor: Prepared by:

Md. Sakibul Islam S. M. Jamil Sagar ID: 18103188
Lecturer Md. Imtiaj Hossain Rafi ID: 18103081
 2

Content
• Introduction

• Problem Statement

• Motivation

• Literature Review

• Methodology

• Experimental Details

• Datasets

• Research Question

• Conclusion

• Reference
 3

Introduction
 There have almost 250,000 named species of flowering plants in the world.
 It’s a part of our life.
 The number of flower species is so big.
 Image classification by MKL – SVM.
 4

Problem Statement
• The applicability of symbolic representation schemes along with different classification
strategies for effective multi-label classification of flower spices.
• With the help of MKL – SVM, we can identify the newly input flower image with
optimum accuracy.
 5

Motivation
• Challenges:
1. Identify native Bangladeshi flowers.
2. To make high accurate result of flowers recognition.
3. To identify same flower of different specific color.
• Solution: We can use MKL – SVM model. Basic and morphology features including
color, size, texture of flower and flower image are extracted to increase the identification
accuracy.
 6

Literature Review
1. Digital Image Classification for Malaysan Blooming Flower (Fadzilah Siraj et al.)
- They showed the usage of NN and a picture method for flower image options. The prediction
accuracy of provision regression is 26.8%.
Weakness: The accuracy rate is very low.

2. Automatic recognition of wild flowers (T. Saitoh et al.)

- They developed an automatic recognition system for wild flowers. There are two input
images utilized; flower and leaf.
Weakness: It is not convenient to take pictures of both flower and leaf.
Limitation: The system may require much time and space complexity.
 7

Literature Review (cont’d)

3. Flower Recognition System Based on Image Processing (Tanakorn Tiay et al.)
- They used edge and color features of flower pictures for the KNN to classify flowers.
Weakness: The accuracy of this method is around 56%.

4. A visual vocabulary for flower classification (M. E. Nilsback et al.)

- They extracted two main flower features, color and shape. For color dependent conditional
random forest algorithms were used to obtain color features. For shape features, generic shape
fitting was used to detect petals.
Weakness: Some flowers have overlapping petals.
Limitation: There is a problem of finding correct center of the flower.
 8

Methodology
Our proposed model works in seven steps.
• Data Collection
• Flower Preprocessing
• Flower Segmentation
• Morphology Features of Flower
• Multi-Label Classification
• Training Model for Data Accuracy
• Testing Model for Final Output
 9

Experimental Details
Dataset Pre-Processing Flower Identification
SVM, KNN, ANN,
Resizing Random Forest
Flower image

Training
dataset MKL+SVM+ML

Noise filtering

IDENTIFICATIO

Testing
RGB N
Feature Extraction

Testing
COLOUR
Flower
Image Segmentation
TEXTURE Identification
Output
Otsu Thresholding
SIZE
 10

Experimental Details (cont’d)

Multi
labelling
of flower
species

Flower
Feature
Dataset Dual Formulation of
Linear SVM

Predicting Flower name

 11

Datasets
Flower image dataset:
- Here we have a dataset for experiment consists with 25000 flower images with 102
different species.
Link: https://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz
- We also have another dataset with 4000 unlabeled flower image in it. It need
prepressing before it can be used.
Link: https://www.kaggle.com/alxmamaev/flowers-recognition
 12

Dataset Images
 13

Research Question
• Can it identify native Bangladeshi flower?
• Can a system recognize different color of same flower using machine learning?
• Is Multi-Label Classification technique is perfect to identify flower accurately?
 14

Conclusion
A prediction model using machine learning supervised algorithm i.e. SVM is built with the
integration of MKL and multi-labeling which can be helpful for the non-botanist’s to
classify different flower species easily, and the process becomes very fast, helping them in
further research and study. It is observed the proposed approach achieved relatively a good
classification accuracy with optimum possible extracted features.
 15

References
1. Fadzilah Siraj, Muhammad Ashraq Salahuddin and Shahrul Azmi Mohd Yusof,” Digital Image
Classification for Malaysan Blooming Flower” IEEE-2010.
2. T. Saitoh and T. Kaneko, “Automatic recognition of wild flowers”, System and Computer in
Japan, vol. 34, no. 10, 2013.
3. M. E. Nilsback, and A. Zisserman, “A visual vocabulary for flower classification”, In CVPR,
2016.
4. Tanakorn Tiay, Pipimphorn Benyaphaichit, and Panomkhawn Riyamongkol, “Flower
Recognition System Based on Image Processing” IEEE-2014.
5. Jana Wäldchen, Michael Rzanny, Marco Seeland, Patrick Mäder, “Automated plant species
identification—Trends and future directions”, Pols, Computational Biology, April 5, 2018.
16