Fish disease classification — spotting 6 diseases from a single photo

About the project

Fish diseases spread fast in ponds and farms, and many look alike to the eye. This thesis trains a deep learning model with transfer learning to read a single photo and sort it into 7 classes: bacterial red disease, aeromoniasis, bacterial gill disease, saprolegniasis (fungal), parasitic disease, white tail disease (viral) or a healthy fish.

With data augmentation, layer freezing and careful tuning of batch size and learning rate, the model classified 682 of 697 test photos correctly (about 98%), with every class at 93% or higher. The student received the code, the trained model and a 41-page thesis written in LaTeX.

Key features

• Transfer learning

  A pre-trained CNN fine-tuned on fish photos for faster, better training.

• 7 classes

  Six bacterial, fungal, parasitic and viral diseases, plus healthy fish.

• Data augmentation

  Resizing, rescaling and augmentation make the model robust to new photos.

• Careful fine-tuning

  Layer freezing, batch size and learning-rate tuning, all explained.

• About 98% correct

  682 of 697 test photos right; every class at 93% or higher.

• 41-page thesis

  Literature review, method, results and discussion, written in LaTeX.

How it works

1. 1Fish photo
3. 2Resize & augment
5. 3Fine-tuned CNN
7. 4Disease predicted
9. 5Confidence shown

Tech stack

Deep learning

PythonCNNTransfer learningFine-tuning

Data

Fish disease image datasetData augmentationGoogle Colab

Write-up

LaTeXOverleaf41-page thesis

What you get with a project like this

• Colab notebooks & full source code
• Trained model, ready to use
• Confusion matrix, charts & sample predictions
• Thesis in your university format
• Presentation slides
• Viva prep & one-to-one walkthrough