Data Augmentation and its Application in Histopathological Oral Cell Image Classification
Madhusmita Das, Rasmita Dash · 2022
Deep learning and Deep Neural Networks (DNN) is one of the most popular fields of Machine Learning. The working principle of DNN and deep learning is based on the Artificial Neural Network (ANN). The fundamental of ANN is grounded on human brain, so as DNN also tries to mimic human brain. From various DNN structure, Convolutional Neural Network (CNN) is one of the leading DNN structure used in the area of image classification problems. Although the performance of CNN in image classification problem is outstanding as compare to other traditional machine learning algorithm, but CNN has certain limitation. One of the most challenging issues of CNN is, insufficient data set problem. To learn a CNN model, huge amount of training data set is required, but in real time scenario, huge amount of data set is not practically available. Data augmentation is one of the solutions for this insufficient data set problem in CNN. In this paper author has discuss various approaches of data augmentation technique. In data augmentation technique, new data set is generated from the existing one so as to increase the size of the data set. In this work to analyze the performance of data augmentation, author has proposed a 14 layered CNN model, and observe the role of data augmentation, by using the proposed model for the classification of histopathological oral cell image data set. The performance analysis of the proposed CNN model is also compared with the state-of-the-art model like VGG16 and VGG19 and the proposed model outperform with 90.01% of testing accuracy using augmented data set whereas, using the original data set the proposed model got testing accuracy of 80.07%.