TY - GEN
T1 - Remote Sensing Infrared Weak and Small Target Detection Method Based on Improved YOLOv5 and Data Augmentation
AU - Zhang, Meixin
AU - Liu, Zhonghua
AU - Zhang, Peng
AU - Yu, Qian
AU - Li, Zhiyuan
AU - Li, Yi
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.
PY - 2025
Y1 - 2025
N2 - 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.
AB - 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.
KW - Infrared weak targets
KW - Object detection
KW - YOLO
UR - https://www.scopus.com/pages/publications/85218441325
U2 - 10.1007/978-981-96-0789-1_23
DO - 10.1007/978-981-96-0789-1_23
M3 - 会议稿件
AN - SCOPUS:85218441325
SN - 9789819607884
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 312
EP - 324
BT - Intelligent Robotics and Applications - 17th International Conference, ICIRA 2024, Proceedings
A2 - Lan, Xuguang
A2 - Mei, Xuesong
A2 - Jiang, Caigui
A2 - Zhao, Fei
A2 - Tian, Zhiqiang
PB - Springer Science and Business Media Deutschland GmbH
T2 - 17th International Conference on Intelligent Robotics and Applications, ICIRA 2024
Y2 - 31 July 2024 through 2 August 2024
ER -