Data is the foundation of artificial intelligence,and the size and quality of the dataset greatly affect the training ef-fect of the model.In the practice of deep learning,it is often faced with insufficient data sample size and uneven class distribution,which leads to problems such as model convergence,overfitting,and falling into local optimum,and cannot achieve the expected training effect.To solve this problem,data augmentation technology has emerged.Aiming at the image data augmentation technology in the field of object detection,this paper classifies and organizes the traditional data augmentation technology and the data augmen-tation technology based on deep learning,summarizes the concept and significance of data augmentation technology,summarizes the data set size estimation methods and influencing factors,systematically analyzes the implementation methods of various augmen-tation technologies,and provides research ideas for researchers to augment data in different modes.
万方数据
维普
