Image augmentation approaches for small and tiny object detection in aerial images: a review
Ume Nisa
Abstract
The task of detecting small and tiny objects has not shown significant performance improvement compared to detecting medium and large objects, even with advanced detection methods. Several factors contribute to this, including the small size of the objects themselves, limited availability of small-scale datasets, clustering of objects, and a low object-to-image ratio, etc. However, various image augmentation techniques have been developed to address these challenges. In this study, previous surveys and reviews have been summarized, offering an overview of the augmentation libraries used to implement various augmentation approaches. It also conducted a brief review of both traditional and state-of-the-art augmentation methods, outlining their respective advantages and disadvantages. The focus is on more to explore the effectiveness of traditional and state-of-the-art augmentation approaches for small and tiny object detection tasks in terms of performance improvement. This review paper distinguishes itself from previous papers by covering a thorough study of image augmentation approaches specially designed for detecting small and tiny object detection tasks in aerial images-a topic that has not been extensively covered on this scale in prior literature. This paper also provided a discussion enabling readers to discern which approach or combination of approaches is suitable for addressing specific challenges related to small and tiny object detection. This will help the readers to establish a good knowledge and develop critical and evaluation skills in this domain. Furthermore, it provides present challenges with possible solutions and outlines future directions.