部分特征重定型和浅层交互用于遥感对象检测
Minh Tai Pham Nguyen1, Quoc Duy Nam Nguyen2, Hoang Viet Anh Le3
1Faculty of Advanced Program, Ho Chi Minh City Open University, Ho Chi Minh City, 700000, Vietnam. 2151013086tai@ou.edu.vn.
这项研究介绍了SORA-DET,它是用于遥感图像的高效物体探测器,通过降低参数和更快的推断速度实现高性能. 它提高了无人机应用的检测准确性.
科学领域:
- 计算机视觉 计算机视觉
- 深度学习 (Deep Learning) 是一种深度学习.
- 遥感 遥感 遥感 遥感
背景情况:
- 在遥感图像中对象检测在平衡性能和计算效率方面存在挑战.
- 现有的深度学习模型往往难以满足实时处理和高精度的需求.
研究的目的:
- 开发一种高效的单阶段物体探测器,用于远程传感图像分析.
- 为了提高遥感对象检测任务的检测性能和计算效率.
主要方法:
- 建议使用PRepConvBlock来降低复杂性,并使用重组参数化卷积扩展受体场.
- 引入了SB-FPN,一个浅层的多尺度融合框架,通过跨尺度交互来增强特征表示.
- 开发了用于无人机应用的浅层优化重构架构探测器 (SORA-DET).
主要成果:
- 在VisDrone2019测试套件中,SORA-DET实现了39.3%的mAP50,在SeaDroneSeeV2验证套件中实现了84.0%的mAP50.
- 该探测器的参数数量少得多 (小于大型模型的88.1%),推断速度快5.4毫秒.
- 在远程传感物体检测方面表现优于许多大型模型和最先进的作品.
结论:
- 索拉-DET为远程传感图像中高效准确的物体检测提供了一个引人注目的解决方案.
- 提出的创新有效地解决了性能和计算成本之间的权衡.
- 索拉-DET展示了现实世界无人机遥感应用的巨大潜力.
更多相关视频
13:35Reefshape: A System for the Efficient Collection and Automated Processing of Time-Series Underwater Photogrammetry Data for Benthic Habitat Monitoring
Published on: June 13, 2025
09:19Measuring the Structure, Composition, and Change of Underwater Environments with Large-area Imaging
Published on: April 18, 2025
相关概念视频
Functional Classification of Joints
The functional classification of joints is determined by the amount of mobility between the adjacent bones. Joints are functionally classified as a synarthrosis or immobile joint, an amphiarthrosis or slightly moveable joint, or as a diarthrosis, a freely moveable joint. Fibrous and cartilaginous joints can be functionally classified as either synarthroses or amphiarthroses, whereas all synovial joints are classified as diarthroses.
Synarthrosis
An immobile...
High-Performance Liquid Chromatography: Types of Detectors
Introduction to Sensory Receptors
The first classification criterion is based on cell type, position, and function. Some receptor cells are neurons with free nerve endings, where their dendrites are embedded in the...
Depth Perception and Spatial Vision
Gestalt Principles of Perception
