(English) Sử dụng xử lý ảnh bằng Opencv ước lượng chiều dài và cân nặng của cá dẹt Hàn Quốc

Real-time estimation of the
length and body weight of
olive flounder in Indoor
Aquaculture Farms Using
Visual Information
• Nguyen Thi Phuong Hang
• Company: Lucas


CONTENTS
I

Project description

II

Environmental system

III Estimation method
IV Result
V

Application

VI Discussion


Project description
Company partner: Lucas
Motivation:
+ For saving feed and labor, the cost

+ Reducing water pollution during starvation, and improving
feeding activity and feed efficiency shortly after refeeding
+ Achieving compensatory growth of fish
Þ Support the harvest, and optimize the farm management
Purpose:
+ Develop image-based olive flounder length measurement
+ Estimate the body weight based on the length with the
length-weight relationship


II.
Experimental environment
and system


Environmental system
A water tank
(2300 × 1195 ×
1200 mm, 3.3 m3)

A 5×5 cm
grid

2 cameras

Experimental conditions for image data acquisition (top).
And Left-side and right-side phone captures of the training environments (bottom)


Data acquisition

The system includes:
ã A water tank (2300 ì 1195 × 1200 mm, 3.3 m3) with a 5 × 5
cm grid bottom created by the black vertical lines and the
black horizontal lines
•

2 cameras at the top of the tank to capture the image

•

The experiment was performed with 90 olive flounders

•

First, the fish were anesthetized in a solution to losing
consciousness then they were put on an electronic scale and a
measuring board to acquire the accurate weight and length as
the actual length (cm) and weight (g)


Dataset
+

The dataset
(1600x900)

includes

180

images

taken

+ After we get the reference object, we can use the
system to estimate length without the 5x5 grid


Estimation
method


Estimation algorithm
Captured
images

Detect
Contours

Detect extreme
points

Calculate the
distance
between
extreme points
Find the max
distance – pixel
length

Cameras

Training
environment
(reference object)

Lagrange's
interpolation
formula
Length in cm


Lagrange’s interpolating formula
 
Lagrange’s
interpolating polynomial is an algorithm that can
construct a polynomial that passes through any desired set
of points and is represented as follows:
 

Here, polynomial of degree passes through points (x1, y1 =
f(x1)),(x2, y2 = f(x2)), …, (xn, yn = f(xn)).
 
𝑛
𝑘
𝑗
¿𝑘=1
𝑗
𝑘
¿𝑘 ≠ 𝑗

𝑃𝑗 ( 𝑥 ) = 𝑦

∏

𝑥−𝑥
𝑥 −𝑥


Results


Results of Reference environment (reference
object to convert from pixel to real length (cm)


The result of the estimated fish length and
t he table summarize the mean, SD (standard
deviation), R-square, and MSE (Mean squared
error) for evaluating estimated results.


Result of length-weight relationship
Mass estimation models were created using two basic
mathematical relationships: Weight = a(length)b + c

Result of the estimated regression equation between the length and the
body weight of fish (left) and the comparison of estimated data with
surveyed data (right)

Application
Build with Streamlit framework



DEMO VIDEO


DISCUSSION AND CONCLUSION
+ The algorithm proposed with high accuracy (R-squared =
0.994)
+ The length-weight relationship generate to support
estimating the body weight of fish, and the length-weight
model with high accuracy (R-squared = 0.91)
+ The overlapping is a limitation
+ Future work: object detection with a Deep learning model
rotated the bounding box will increase the accuracy of length


THANK YOU
FOR
LISTENING