Scribd
Upload
10 views

Image Fish

Research paper

Uploaded by

Shohanur Rahman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
10 views

Image Fish

Research paper

Uploaded by

Shohanur Rahman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Fish Image

Classification Project
Introduction
This project aims to develop a machine-learning model for Bangladeshi Fish classifying
using deep-learning techniques. The project is divided into three sections, each
focusing on different aspects of machine learning workflows, including data
preprocessing, model building, and comparison of models with and without pre-trained
weights.

Dataset Source

● Datalink :Fish Images

Or

Use this data : Fish Images dataset 2 (if the upper one gives bad result)

Preparatory Instructional Materials


Before starting the project, participants must watch specific videos and read articles to
understand the methodologies and tools involved:

CNN Conceptual Video: Video Tuitorial

● For Sections 1 and 2:


● Video Tutorial
● For Sections 2 and 3:
● CNN Multiclass Image Classification Python / Confusion Matrix
● Precision, Recall, F1 score, True Positive
● Article on Evaluation Metrics for Classification Models
● VGG16 Transfer Learning (for your implementation, you have to use
layers.trainable=True)

Note: For computational resources, participants can use their own PC's GPU or cloud
platforms like Google Colab or Kaggle. In Google Colab, select the GPU as the runtime
type for enhanced performance.

Section 1: Data Preprocessing


Objectives:

● Download and upload the dataset to Google Drive.


● Mount Google Drive in the Colab environment.
● Load the dataset into the notebook.
● Resize images to 224x224x3, convert them into NumPy arrays, normalize by
dividing by 255.0, and add them to a list named images.
● Create labels for the dataset and use encoders to format them appropriately.
● Plot a bar graph to visualize the distribution of samples across different classes.
● Split the dataset into training (70%), testing (20%) & validation (10%) sets, using
stratification.

References:

● Data Splitting and Stratification: Stratify in Train/Test Split

Section 2: Building a Custom VGG19 Model


Objectives:

● Construct a custom VGG19 model without using pre-trained weights.


● Compile the model using the Adam optimizer and an appropriate loss function.
● Fit the model to the dataset for 10 epochs, incorporating early stopping and
CSVLogger for bonus points.
● Save the trained model.
● Evaluate and display training and testing accuracy.
● Plot graphs for training and testing accuracy and loss, and analyze overfitting or
underfitting.
● Generate a classification report and a confusion matrix using the Seaborn library.

References:

● Early Stopping and CSVLogger: Guide to Early Stopping and CSVLogger

Section 3: Implementing Pretrained VGG19 Model


Objectives:

● Create a VGG19 architecture with ImageNet pre-trained weights, ensuring all


layers are trainable.
● Compile and train the model for 10 epochs.
● Save the trained model.
● Evaluate and display training and testing accuracy.
● Plot graphs for training and testing accuracy and loss, and analyze overfitting or
underfitting.
● Generate a classification report and a confusion matrix.

References:

● VGG19 Pretrained Model: Keras VGG19 Model

Comparison and Conclusion


Tasks:

● Compare the performance of the models from Sections 2 and 3.


● Discuss which model performed better and why.
● Suggest methods for increasing model accuracy and potential future works on
the dataset.

You might also like