A Keras implementation of VGG19-SVM model to predict malaria from microscopic images
This implementation is based on a research paper by professors Dr. Rajesh Kanna B, Dr. Vijayalakshmi A., Mr. Dinesh Jackson. The paper is also uploaded in the repo.
fine_tune_model.ipynb contains the code to fine tune the VGG19 model which is trained on imagenet dataset for the malaria dataset. The softmax layer is removed and replaced with another softmax layer with two classes. Either 0 or 1. Whether the given microscopic image of blood sample has or doesnt have malaria. This finetuned model
extract_features_finetune.ipynb contains the code to extract feature vector after the fifth convolution block and before the fully connected layer of the above fine tuned model. The feature size is (7x7x512) which on flattening gives feature vector of size (1x25088) for every image (in both test, validation sets ) and is saved to a pickle file for future use.
svm.ipynb contains the code to train SVM on the features extracted from the finetuned model. This trained model can be used to predict if a image has malaria or not.
SVM_FULL.ipynb contains updated svm.ipynb code. K-Flod CrossValidation and Grid Search are added to the previous code.
data.zip contains the dataset.
Note: - Code is not commented.