Small Object Detection Based on Deep Learning for Remote Sensing: A Comprehensive Review

Abstract

With the accelerated development of artificial intelligence, remote-sensing image technologies have gained widespread attention in smart cities. In recent years, remote sensing object detection research has focused on detecting and counting small dense objects in large remote sensing scenes. Small object detection, as a branch of object detection, remains a significant challenge in research due to the image resolution, size, number, and orientation of objects, among other factors. This paper examines object detection based on deep learning and its applications for small object detection in remote sensing. This paper aims to provide readers with a thorough comprehension of the research objectives. Specifically, we aggregate the principal datasets and evaluation methods extensively employed in recent remote sensing object detection techniques. We also discuss the irregularity problem of remote sensing image object detection and overview the small object detection methods in remote sensing images. In addition, we select small target detection methods with excellent performance in recent years for experiments and analysis. Finally, the challenges and future work related to small object detection in remote sensing are highlighted.

Adaptation of YOLOv7 and YOLOv7_tiny for Soccer-Ball Multi-Detection with DeepSORT for Tracking by Semi-Supervised System

“…The objective is to know the trajectory of the ball so that in subsequent work, a correlation can be made with the performance of the players in a context of sports video analysis. The object of study that you want to detect in this work is framed within small object detection (SOD) [34][35][36][37][38], in addition to the problem that the soccer ball moves at an average speed from 120 to 130 km/h in professional matches, so the challenges to be faced focus on the detection and location of an object that has a small dimension compared to the rest of the scene in a soccer match, which brings the following problems: the object is lost within the context, the color of the object can be confused with walls on the field since the camera shows scenes in 2D, the speed of movement of the ball generates motion blur which causes deformation of the object within the scene, in addition, a ghosting effect is generated, which causes blurring of the ball. These are the reasons why using the YOLO v7 architecture is proposed, which is recognized for its effectiveness in detecting objects in real time.…”

“…The technique was analyzed in the detection of balloons, where accuracies of 85% or less were obtained, and in cases of being higher than that percentage, was as a result of being based on datasets of a single specific balloon with shots in controlled environments. In the literature focused on small object detection (SOD) [35], interesting studies have been carried out to compare the performance of the current models [36][37][38], from which it is generally concluded that the best results are obtained by the convolutional models: Faster R-CNN, SSD, and YOLO.…”

Accurate Recognition of Jujube Tree Trunks Based on Contrast Limited Adaptive Histogram Equalization Image Enhancement and Improved YOLOv8

“…However, boosting computational resources is not a preferred option for field computing platforms. In this paper, we start from the algorithm, discarding the redundant information [10], adopting the key frame extraction technique to quickly extract effective crop sample information, and combining it with a suitable image quality enhancement technique to efficiently and accurately train an object detection model [11][12][13][14]. By improving the efficiency of the computerized information transfer, we can achieve refined management and assisted decision-making, ultimately improving the efficiency and accuracy of agricultural production [15][16][17].…”

A Lightweight Network FLA-Detect for Steel Surface Defect Detection

“…The other type is the single-stage network based on direct feature extraction, such as the YOLO [7,8] series. With the development of deep learning [9], object detection has made significant progress in the field of computer vision, yielding promising results across various application domains such as autonomous driving [10], underwater operations [11], facial detection [12], medical imaging [13] and remote sensing detection [14,15]. As a result, defect detection is also becoming increasingly popularly employed by deep learning-based target detection techniques.…”