1. A seminar on
AN AUTOMATED IMAGE PROCESSING SYSTEM FOR
IDENTIFICATION OF FISH SPECIES LABEO BATA
FRM-591
2. Contents of my seminar:
Introduction
Need of fish identification
Challenges of identification
History of an image recognition
Objective of my thesis
Systematic position
Methodology of an automated image recognition system
Data collection
Image preprocessing
Features extraction
Features extraction technique
Classification
Fish identification
Advantage of an automated image recognition system
Future scope of the research
Conclusion
References
3. INTRODUCTION:
Fish species identification is traditionally based on external
morphological features.
This process of identification is follow well trained fisherman and
fishery expert.
In case of common people it is very hard work to identified all
kinds of fishes.
Most recently An automated Image Recognition System was
developed to identify fishes.
In this method user provided a photograph of a fish as input and
the software identifies the fish to a taxonomic level.
4. NEED OF FISH IDENTIFICATION
World has more than 32,500 fish species. Among them some are edible and
some are extremely poisonous.
Many of people died every year because they do not identify the poisonous
and non-poisonous fish.
So it is essential to identify properly whether it is harmful or not.
Fishes like Puffer fish, Lion fish, and some of the Eels are extremely
dangerous to eat.
To avoid this type of poisonous fishes we need to have proper knowledge of
identification.
There are many others reason where proper indentified of fish is needed.
This are -
1. In fish processing industries the first step is to identify the fishes and
sorted according to their commercial importance. Because the value of the
product depends on quality and types of fishes. If some unknown species is
processed by mistake then the whole product should be damaged.
2. Fish identification is not only helpful to fisheries industries, it also helps
the consumers. There are several fishes in the market which are quite similar on
their appearance and huge confusion among them. Sometimes consumers are
cheated by the seller due to lack of knowledge about fishes.
3. The price of the Ornamental fish depends on sex of the fishes. So proper
identification is highly profitable for the business,
5. Identification of fish is more difficult than any other organism.
Teleost species are the largest category among vertebrate
animals.
Their numbers reached more than 32,500.
Fishes show their diversity among themselves with shape, size
and color. So it is very hard work to recognition the fish species.
Some difficulties are-
Challenges of identification of fish
1) Not easily visible- It is very difficult to identify a fish
under water as it is not possible to see clearly
morphological and anatomical characters.ii) Perishable-Fishes are highly perishable object. After removing
fishes from water, it started to degrade so quickly. The
anatomical and morphological character spoiled. So it becomes
very difficult to identify particular species
iii) Similar shape- There are several fishes in the world which are
similar shape and size. It makes difficult to identify the fishes.iv)Unidentified species- Till now there is a huge unexploited
underwater area and many fishes are totally unknown to the
scientists.
v) Hybridization- Sometime cross breeding is occur among the
underwater species and new types of species with new characters
are come out.
6. History of Image Recognition:-
The evolution of fish began about 530 million years ago.
During this time people caught fishes only for food purposes
without knowing any name, and mistakenly having some poisonous
fishes then accident may occur.
After that people trying to identifying the edible fishes and till
now the identifying process is continue.
Now identifying of fish through computer system is developed
recently around 60 years ago.
Computer scientists are always trying to extract meaning of the
image by machine learning.
7. 1) In 1959 Russell Kirsch and his colleagues develops a system that
transforms visual images into numbers, which machine could understand.
2) In the year 1982 David Man developed a system which can detect edges,
curves, corners of an image.
3) A Japanese computer scientist, Fukushima built a self-organizing
artificial network which recognize the patterns of an image. The network
consists of several convolutional layers which can identify shape of an
image.
4) In 1997 Jitendra Malik tried to convert a images into sensible parts using
a graph theory algorithm.
8. Objective of my thesis:
Labeo bata commonly known as bata or bangon is one of the
most important target species for small scale fisheries.
L. bata is commercially important and great demand in the
market because of its high nutritional value and good taste.
Object of my research tropic is to identify the Label bata
through an automatic image processing system.
10. Morphological description:
Dorsal profile is more convex than that of abdomen. Body elongated, A pair of
small maxillary barbells is hidden inside the labial fold. Dorsal originates midway
between snout tip and anterior base of anal. Pelvic originate slightly nearer to
snout tip than caudal base. Bluish or darkish on upper half, silvery below. scales
on the lateral line is 38 and 40 respectively.
Habit and habitat:
Labeo bata is a freshwater fish found in small rivers, canals, ponds and
ditches. Its food are crustaceous and insect larvae in early stages. In adult stage
rotten plant, algae and plankton are eaten.
Breeding:
Size at first sexual maturity is 14.12 and 14.60 cm in male and female.
Spawning season varies from June to October, average fecundity was 192785.
11. Methodology of an Automated Image
Recognition System:-
There are mainly 5 steps of an automated image recognition
system. This are-
Data collection Image
prepossessing
Image
segmentation
Classification
Feature extraction
12. Data collection:
Data collection is an important part of this system. Here collection
of data means captures the images of fishes. There is no particular
limit of data collection. More data means more accurate result.
However, for a good accuracy minimum 300 picture needed for
each species. There are some process of data collection.
1) Data should be collected randomly.
2) Resolution should be same for each image.
3) Picture must be taken from 900 angle.
4) Full body of the fish have to be captured.
5) Caudal fin arranged in relaxed position.
6) Fishes must be photographed sideway.
13. Image preprocessing:
Image pre-processing is nothing but a image is prepared
and ready for next step. Image preprocessing can be
done by-
1. Sorting and Labeling of the images.
2. Colour intensity normalization
3. Enhances the edges of an image.
4. Reduces the blur of an image..
5. Rotation of an images into same direction.
6. Resizes the images into a normalized range.
7. Horizontal and vertical brightness normalization.
14. Feature extraction:-
Features define the behavior of an image and Feature
extraction is the main part of an image recognition system.
The main purpose of feature extraction is to detect largest set
of features of a species which are same for a similar species but
different from another species.
In this process relevant features were extracted from object’s
image which is form ‘feature vectors.’
Then these ‘feature vectors’ were used by classifiers to
recognize the input data for target output data.
The classifier is classify between different classes by looking
at these features and make easy to distinguish between two
classes.
15. Some feature extraction technique.
1. Texture feature extraction using LBP:- The term texture generally refers to
texels, which contains several pixels. LBP described the texture of an image.
In this process the image is dividing into several small regions from which the
features are extracted and considers the result as a binary number.
2. Geometric parameter using contour feature :- Contours detection is a
process can be explained simply as a curve joining to all continuous points.
3. Colour feature using Colour Histogram:- Colour Histogram is the most
widely used technique for extracting the colour feature of an image. It
represents the frequency distribution of colour bins of an image
16. Classification:
The main purpose of the classification is to categorize the input
images depends on their features.
It is machine learning approach which the computer program learns
from the input datasets.
On the basis of features, classifier categorized the input data to
specific type of category.. At first, the classifier was trained with
training dataset. Accuracy of the classifier depended on the quality of
training data set.
17. Some classifier are
Artificial Neural Network (ANN )- Biologically inspired computer
programs. ANNs gather their knowledge by detecting the patterns
and relationships in data and learn through experience, not from
programming.
K-Nearest Neighbor (KNN-) K nearest neighbors is a simple
algorithm that stores all available classes and classifies new classes
based on a similarity measure.
Support Vector Machine (SVM) : It is a classifier formally defined
by a separating hyperplane. Hyper plane is a one dimensional or
two-dimensional boundary which separate 2 classes.
18. Fish identification:
The application was built on MATLAB 2018a APP designer.
It was mainly graphical user interface model.
The application has a button of ‘Load Image’ to input an unknown
image.
The image was displayed on the upper portion of the application.
It has a ‘Test Image’ button to identify the image of fish. After
pressing the ‘Test Button’ the predicted result will be shown below
the ‘Test Button’.
The accuracy rate or quality of prediction will be shown on bottom
left corner of the system.
19. Advantages of Image Recognition through
automatic image recognition System:
Response Time- Response time defines how quickly a result is obtained.
Traditional tools require more time to identify fish because it is processed through
several steps. But Automatic image identification system works on very low
response time. Therefore, it is used in practical field.
Accuracy-Accuracy measures the error rate of an identification system.
Sometimes manual identification is not so accurate because the chance of mistake.
But automated computer based recognition system has more accuracy than any
other identification tools.
Parts of body: In case of manual identification it is not possible to identify a
fish through its body part. Whereas through automatic image processing system it
is possible to identify species by its body parts or photographs.
20. The present study on “An automated image recognition system for identification
of Indian minor carps will be helpful to all label of person who engaged fishery
sector. The major focus in this field-
Identification of live underwater fishes through video footage.
Automatic Sorting of fishes according to the species, sex and size.
Implementations of the system into mobile device so that anyone can access
the application at anywhere anytime like QR barcode.
The system needs to be implemented in a world-wide-web based system so
that people can share global fish information.
FUTURE SCOPE OF RESEARCH
21. The development of an automated image
recognition system (software) may be a stepping
stone in the field of fish taxonomy.
But it has to cross a long journey to fulfill the
requirement of fish folks of the world. Then it will be
helpful for the people who are ignorant about the fish
species, its’ variety and its identification.
CONCLUSION
22. References:-
www.google.com
A Brief History of Computer Vision (and Convolutional Neural Networks).
Retrieved 31 July (2019), from https://hackernoon.com/a-brief-history-of-
computer-vision-and-convolutional-neural-networks-8fe8aacc79f3
Texture Feature. Retrieved 31 July (2019), from
https://support.echoview.com/WebHelp/Windows_and_Dialog_Boxes/Dialog_Box
es/Variable_propert
Identification of Fish Species based on Image Processing and Statistical
Analysis Research(Lian Li , Jinqi Hong)
Shape-Based Fish Recognition Using Neural Network.(Purti Singh,Deepti
Pandey)BBD University, Lucknow, India.
SVM (Support Vector Machine) — Theory. Retrieved 31 July (2019), from
https://medium.com/machine-learning-101/chapter-2-svm-support-vector-machine-
theory-f0812effc72