Abstract

Modern machine learning algorithms need a substantial quantity of high-quality annotated figures to guarantee effective performance. Meanwhile, manually performing the data collection and explanation methods takes a lot of time and resources. Obtaining enough training data is frequently only practicable in some real-world application settings. The best approach to addressing this complexity right now is data augmentation. Data augmentation primarily aims to enhance training data's quantity, quality, and diversity. The proposed idea thoroughly explores data augmentation procedures that apply to computer vision provinces. Newer, more cutting-edge methods of data augmentation are highlighted. This paper provides an in-depth analysis of data augmentation techniques utilized in computer vision, emphasizing the use of modern, advanced methods. The survey covers various methods for augmentation, including deep learning strategies, extracted features and Machine Learning based techniques, data synthesis approaches, and dataset processing for 3D graphics modeling, neural rendering, and generative adversarial networks. It covers modern techniques and applications, examining the effectiveness of various augmentation methods and their performance on different datasets and tasks. Deep learning-Neutral Networks (DLNN) architecture for medical images applications are an attractive area of research for rapid examination and accurate diagnosis, but require numerous datasets and training stages. Due to factors such as insufficient patients, reluctance to share images, lack of medical apparatus, or the inability to acquire images that satisfy particular criteria, medical images are scarce. This causes overfitting, bias in datasets, and inaccurate results. A variety of techniques have been implemented on distinct image types as part of the standard data augmentation procedure for addressing this issue. However, it is crucial to determine which data augmentation technique provides more efficient results for each image type, as different diseases, network architectures, and datasets are used. This study examines augmentation techniques used in deep learning-based disease diagnosis using different imaging modalities, including PET. US, MRI and mammography. The effectiveness of these techniques in classification with a deep network is discussed, and experiments suggest that augmentation methods are suitable has been chosen for the proposed work. Based on obtained results, the proposed work has produced better in terms of mean error, median error and max error with improvement of 23%.