Skip to main navigation Skip to search Skip to main content

Remote Sensing Infrared Weak and Small Target Detection Method Based on Improved YOLOv5 and Data Augmentation

  • Meixin Zhang
  • , Zhonghua Liu
  • , Peng Zhang
  • , Qian Yu
  • , Zhiyuan Li
  • , Yi Li*
  • *Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Remote sensing infrared satellite images have the characteristics of weak targets, insufficient contrast, and easy to be affected by the surrounding environment, such as clouds and fog, so it is a great challenge to detect weak and small targets in remote sensing. In this paper, we propose a detection method based on weak and small target enhancement, which uses a bidirectional histogram to improve the image contrast, and uses the infrared image dehazing algorithm with fog line dark primary color prior to preserve the pixel distribution of the infrared image to the greatest extent while enhancing its contrast and detail. In terms of the model, we introduce a simple and efficient weighted bidirectional feature pyramid network to optimize feature fusion, reduce redundant calculations while maintaining the detection ability of the model, and greatly reduce the memory occupation. The results show that the proposed method has achieved more competitive results than the current mainstream methods in dealing with the problem of infrared weak and small target detection, and in addition, due to the application of the weighted bidirectional feature pyramid network, the video memory is reduced by 43% while maintaining the competitive accuracy, which is of great practical significance.

Original languageEnglish
Title of host publicationIntelligent Robotics and Applications - 17th International Conference, ICIRA 2024, Proceedings
EditorsXuguang Lan, Xuesong Mei, Caigui Jiang, Fei Zhao, Zhiqiang Tian
PublisherSpringer Science and Business Media Deutschland GmbH
Pages312-324
Number of pages13
ISBN (Print)9789819607884
DOIs
StatePublished - 2025
Event17th International Conference on Intelligent Robotics and Applications, ICIRA 2024 - Xi'an, China
Duration: 31 Jul 20242 Aug 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume15209 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference17th International Conference on Intelligent Robotics and Applications, ICIRA 2024
Country/TerritoryChina
CityXi'an
Period31/07/242/08/24

Keywords

  • Infrared weak targets
  • Object detection
  • YOLO

Fingerprint

Dive into the research topics of 'Remote Sensing Infrared Weak and Small Target Detection Method Based on Improved YOLOv5 and Data Augmentation'. Together they form a unique fingerprint.

Cite this