Salt Body Identification by Convolutional Neural Network

December 11, 2018

by Shihang Feng

Figure 1: (a) Input image, (b) classication results of all imaging points in the input image.

1 Objective

In this lab, we will go through convolutional neural network (CNN) to distinguish salt body in a migration image.

2 Prerequisties

Matlab

3 How to use the CNN lab

Download the codes ./CNN_salt.zip and unzip it. Change your Matlab working directory under this file so you may able to use all necessary sub-functions. The main function is function test_example_salt.m, open it in the matlab script and run.

test_example_salt

4 Procedure of CNN

Build training sets. Pick 2000 crack image in the image and labeled as 1 and then pick 2000 non-crack image in the image and labled as 0.
Convert the image into image patches.
Use CNN algorithms to estimate a optimal set of weights and bias.
Input all the image patches into the trained CNN model to do classification.

5 Convolutional Neural Network

The misfit function of the convolutional neural network is:

J(w,b) = 1-∑ (y(x )- a)2 2n x

where w is the set of all weights and b the set of all biases, n is the number of training inputs, x is the set of all training inputs, y(x) is the expected output of the network for input x, and a is the actual output of the network for input x, with respect to w and b. The architecture of the convolutional neural network is shown in Figure 2.

Figure 2: The convolutional neural network architecture with two convolutional layer and a fully connected layer.

6 Questions

Use more or less trainning examples in the trainning, how the size of trainning data would affect the result?
Pick the training set manually by yourself, how it would affect the result?
Modify the structure of CNN. See how the CNN parameter will change the result.
What the relationship between minibatch, epoch and iteration? Try to change batch size, how the iteration number changes?

7 PS

If there are any errors in this Lab, please contact: shihang.feng@kaust.edu.sa