Transfer Learning-based Feature Fusion of White Blood Cell Image Classification
Shivani Parayil, J Aravinth
Abstract
White Blood Cells (WBCs) are a crucial part of the human immune system that helps shield the human body from foreign bodies and other infections. Any change in the count, increase or decrease of WBC can be fatal to the human body. This leads to the need for proper identification and classification of WBCs. Automating the classification of WBCs can lead to earlier predictions of abnormalities. This paper makes an effort to devise a methodology for automation by using feature fusion. For feature extraction, various fusion techniques using transfer-learning approaches such as Densely connected convoluted neural networks (DenseNet201) and VGG16 (Visual Geometry Group 2016) were proposed. The classification results are compared using various performance metrics such as Accuracy, Precision, Recall, and F1-Score. The maximum accuracy of 89.75% was obtained with the help of feature fusion combined with the Convolutional Neural Network (CNN) classifier.