File size: 2,992 Bytes
0287c29
1ca5362
 
 
 
 
 
0287c29
fe65e01
 
0287c29
 
1ca5362
0287c29
1ca5362
0287c29
1ca5362
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0224f0d
0c6ac88
0224f0d
0c6ac88
c27bdc0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
---
title: FishClassifier_CNN
emoji: 🐠
colorFrom: blue
colorTo: green
sdk: streamlit
app_file: src/streamlit_app.py
pinned: false
short_description: CNN from scratch for 9-class fish classification.
license: mit
---

# 🐟 Fish Classification (CNN From Scratch)

This project builds a **Convolutional Neural Network (CNN) from scratch** to classify images from the **"A Large Scale Fish Dataset"**, containing 9 different fish species.

This model is part of a larger deep learning assignment where the goal is to build:
1. A CNN **without transfer learning**
2. A CNN **with transfer learning** (e.g., ResNet50)
3. A Streamlit app for each model
4. A technical PDF report for each dataset

---

## πŸ“‚ Dataset
**A Large Scale Fish Dataset**  
Kaggle link: https://www.kaggle.com/datasets/crowww/a-large-scale-fish-dataset

Dataset structure (after download):

Fish_Dataset/
β”œβ”€β”€ Black Sea Sprat
β”œβ”€β”€ Gilt-Head Bream
β”œβ”€β”€ Horse Mackerel
β”œβ”€β”€ Red Mullet
β”œβ”€β”€ Red Sea Bream
β”œβ”€β”€ Sea Bass
β”œβ”€β”€ Shrimp
β”œβ”€β”€ Striped Red Mullet
└── Trout

yaml
Code kopieren

---

## 🧠 Model Overview

The CNN was built using Keras Sequential API with:

- **5 Convolutional Layers**
- **3 MaxPooling Layers**
- **1 Dense Layer (256 units)**
- **Dropout (0.5)**
- **Softmax output layer (9 classes)**

**Training Setup:**

- Loss: `sparse_categorical_crossentropy`
- Optimizer: `Adam(lr=0.0005)`
- Metrics:  
  - `accuracy`  
  - `top-3 accuracy`  
- Callbacks:
  - EarlyStopping  
  - ModelCheckpoint (`best_fish_cnn.keras`)
  - ReduceLROnPlateau  

---

## πŸ“ˆ Results

Final model performance on validation data:

- **Accuracy:** ~84%
- **Top-3 Accuracy:** ~99%
- **Validation Loss:** ~0.45

The model performed very well on distinct species (Shrimp, Red Mullet, Trout).  
Classes that look visually similar (Sea Bass, Bream types) caused some confusion.

The technical report (PDF) contains:
- Classification Report
- Confusion Matrix
- Full explanation of methods & reflection

---

## πŸš€ Usage

### How to Use the Model (CNN From Scratch)

__Load the model__

import tensorflow as tf

model = tf.keras.models.load_model("best_fish_cnn.keras")

_Predict on a new image_
import numpy as np
from tensorflow.keras.preprocessing import image

img = image.load_img("your_image.jpg", target_size=(224, 224))
img_array = np.expand_dims(image.img_to_array(img), axis=0) / 255.0   # normalize

pred = model.predict(img_array)
print(pred)

__Streamlit App__

This HuggingFace Space includes an interactive Streamlit app where you can:

Upload a fish image

See the model’s prediction

View confidence scores for all 9 fish classes

__Project Files__
File	Description
best_fish_cnn.keras	Saved CNN model (from scratch, no transfer learning)
streamlit_app.py	Streamlit app code
requirements.txt	Required Python packages
Fish_CNN_Report_English.pdf	Technical report (simple English)

## πŸ‘€ Author

**Enes Yamac**

Machine Learning & Computer Vision Project (2025)