Image-Classification-with-Deep-CNN

A Deep Convolutional Neural Network with drop out layer architecture to decrease over-fitting and to increase accuracy.

Requirements

• Python 3.6 and above
• Tensoflow 1.6.0 and above
• NumPy
• Pandas
• Matplotlib
• PIL
• Also be using a utility library to understand the outputs better.

Data

Using MNIST dataset for this to test our model architechture's accuracy against best algorithm. MNIST is a database of handwritten digits that has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image.

We can import the dataset using TensorFlow built-in feature, as shown below-

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

Input has 784 pixels, 28x28 (widthxheight)

Output, 10 possible classes (0, 1, 2, 3, 4, 5, 6, 7, 8, 9)

Model Architechture

1. Input - MNIST dataset
2. Convolutional, ReLU and Max-Pooling
3. Convolutional, ReLU and Max-Pooling
4. Fully Connected Layer
5. Processing - Dropout and ReLU
6. Readout layer - Fully Connected
7. Outputs - Classified digits

Padding used for convolution is SAME padding

Optimizer is AdamOptimizer with learning rate of 1e-4. Using the complete dataset produces better results but can lead to overfitting of data. Test Accuracy of 99.57% (0.43% Error rate).