LeNet-5 Explained: The Neural Network That Revolutionized Image Recognition

LeNet-5 Explained Simply (Step-by-Step CNN Guide)

📚 Table of Contents

• What is LeNet-5?
• Core Intuition
• Layers Explained
• How It Works
• Code Example
• CLI Output
• Why It Matters
• Key Takeaways
• Related Articles

---

📖 What is LeNet-5?

LeNet-5 is one of the first successful Convolutional Neural Networks (CNNs). It was designed to recognize handwritten digits like 0–9.

💡 Simple idea: Teach a computer to “see” patterns in images, just like humans do.

---

🧠 Core Intuition

LeNet-5 works in 3 simple stages:

• 👀 Look for basic patterns (edges, lines)
• 🧩 Combine patterns into shapes
• 🔢 Decide which digit it is

💡 Think like a human: first see lines → then shapes → then numbers

---

🔍 Layers of LeNet-5

1️⃣ Input Layer

Image size: 32 × 32 (grayscale)

Original images are 28x28. Padding adds borders so edges are not lost.

2️⃣ Convolution Layer (C1)

- 6 filters (5×5) - Output: 28 × 28 × 6

Filters scan the image and detect edges, corners, and simple patterns.

3️⃣ Pooling Layer (S2)

- Average pooling (2×2) - Output: 14 × 14 × 6

Reduces size → faster computation → keeps important info.

4️⃣ Convolution Layer (C3)

- 16 filters - Output: 10 × 10 × 16

Now the network detects more complex shapes (like curves and parts of digits).

5️⃣ Pooling Layer (S4)

- Output: 5 × 5 × 16

6️⃣ Fully Connected Layer (F5)

- 120 neurons

All features are combined to understand the full image.

7️⃣ Output Layer

- 10 neurons → digits (0–9)

---

🔄 How LeNet-5 Works (Step-by-Step)

1. Input image enters
2. Find edges using filters
3. Reduce size using pooling
4. Detect complex shapes
5. Flatten into vector
6. Predict digit

💡 It slowly turns pixels → patterns → meaning

---

💻 Code Example

import tensorflow as tf
from tensorflow.keras import layers

model = tf.keras.Sequential([
 layers.Conv2D(6, (5,5), activation='relu', input_shape=(32,32,1)),
 layers.AveragePooling2D(),
 layers.Conv2D(16, (5,5), activation='relu'),
 layers.AveragePooling2D(),
 layers.Flatten(),
 layers.Dense(120, activation='relu'),
 layers.Dense(84, activation='relu'),
 layers.Dense(10, activation='softmax')
])

model.summary()

---

🖥 CLI Output

Layer (type)         Output Shape
Conv2D               (None, 28, 28, 6)
AvgPooling           (None, 14, 14, 6)
Conv2D               (None, 10, 10, 16)
AvgPooling           (None, 5, 5, 16)
Dense                (None, 120)
Dense                (None, 84)
Dense                (None, 10)

---

🚀 Why LeNet-5 Matters

• First successful CNN
• Showed machines can learn features automatically
• Foundation of modern AI (face recognition, self-driving)

💡 Without LeNet-5, modern deep learning wouldn’t exist

---

🎯 Key Takeaways

✔ CNN learns patterns automatically ✔ Layers go from simple → complex ✔ Pooling reduces size ✔ Final layer makes decision ✔ LeNet-5 started modern AI

---

📚 Related Articles

• Hidden Physics of Neural Network Failure
• ZFNet
• CNN for Image Segmentation
• Leaky ReLU

---

🧠 Final Thought

LeNet-5 teaches us a powerful idea: Computers don’t need rules — they can learn patterns.