1. Read the images into the training set and the testing set


2. Design the CNN architecture and train

3. 嘗試更改不同的batch_size 最後發現切成1/3 設為180會訓練出比較高的準確度

