CSCE822 Data mining

CSCE822 Data mining

Homework 4
classification
Deep learning application for microscopy image

Install Keras deep learning framework. (If you can use Tensorflow or Pytorch, it is also
ok)
Pip3 install tensorflow keras
Step1 : Study one of the following tutorials on how to apply basic convolutional neural
network for image classification using the Keras deep learning framework
https://towardsdatascience.com/building-a-convolutional-neural-network-cnn-in-keras-
329fbbadc5f5
https://adventuresinmachinelearning.com/keras-tutorial-cnn-11-lines/
https://keras.io/examples/cifar10_cnn/
or the simplest CNN code for digit number image classification
https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py
Download sample code, and run the code, and report your training and test
performance.
Step2:
Modify the sample code for digital number recognition and develop a deep learning
algorithm based on convolutional neural networks for classifying the following 3 classes
of Scanning Electron Microscopy images: fibres, sponge, and power. (See attached image
files). Note that the image dimension of this image set is much larger (1024×768) than
the mnist digit number recognition dataset.
You should use the first the 75% images as training, next 5% as validation set and
remaining as test dataset. Try to compare the performance with or without max-pooling,
batch-normalization, varying the number of convolution layers and convolution layers.
CSCE822 Data mining

Leave a Reply

Your email address will not be published. Required fields are marked *