Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning
With the increasing trend towards informatization and intelligence in modern warfare, high-intensity and continuous reconnaissance activities are becoming increasingly common in battlefield environments via airborne, vehicle, UAV, satellite and other platforms. Visible and infrared images are prefer...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-04-01
|
| Series: | Electronics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9292/12/7/1732 |
| _version_ | 1797608090157187072 |
|---|---|
| author | Yuxing Dong Yan Li Zhen Li |
| author_facet | Yuxing Dong Yan Li Zhen Li |
| author_sort | Yuxing Dong |
| collection | DOAJ |
| description | With the increasing trend towards informatization and intelligence in modern warfare, high-intensity and continuous reconnaissance activities are becoming increasingly common in battlefield environments via airborne, vehicle, UAV, satellite and other platforms. Visible and infrared images are preferred due to their high resolution, strong contrast, rich texture details and color features, and strong information expression ability. However, the quality of imaging is easily affected by environmental factors, making it crucial to quickly and accurately filter useful information from massive image data. To this end, super-resolution image preprocessing can improve the detection performance of UAV, and reduce false detection and missed detection of targets. Additionally, super-resolution reconstruction results in high-quality images that can be used to expand UAV datasets and enhance the UAV characteristics, thereby enabling the enhancement of small targets. In response to the challenge of “low-slow small” UAV targets at long distances, we propose a multi-scale fusion super-resolution reconstruction (MFSRCNN) algorithm based on the fast super-resolution reconstruction (FSRCNN) algorithm and multi-scale fusion. Our experiments confirm the feasibility of the algorithm in reconstructing detailed information of the UAV target. On average, the MFSRCNN reconstruction time is 0.028 s, with the average confidence before and after reconstruction being 80.73% and 86.59%, respectively, resulting in an average increase of 6.72%. |
| first_indexed | 2024-03-11T05:38:45Z |
| format | Article |
| id | doaj.art-5b2688ddcc3a480a968d876e7950194f |
| institution | Directory Open Access Journal |
| issn | 2079-9292 |
| language | English |
| last_indexed | 2024-03-11T05:38:45Z |
| publishDate | 2023-04-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Electronics |
| spelling | doaj.art-5b2688ddcc3a480a968d876e7950194f2023-11-17T16:34:51ZengMDPI AGElectronics2079-92922023-04-01127173210.3390/electronics12071732Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep LearningYuxing Dong0Yan Li1Zhen Li2Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaWith the increasing trend towards informatization and intelligence in modern warfare, high-intensity and continuous reconnaissance activities are becoming increasingly common in battlefield environments via airborne, vehicle, UAV, satellite and other platforms. Visible and infrared images are preferred due to their high resolution, strong contrast, rich texture details and color features, and strong information expression ability. However, the quality of imaging is easily affected by environmental factors, making it crucial to quickly and accurately filter useful information from massive image data. To this end, super-resolution image preprocessing can improve the detection performance of UAV, and reduce false detection and missed detection of targets. Additionally, super-resolution reconstruction results in high-quality images that can be used to expand UAV datasets and enhance the UAV characteristics, thereby enabling the enhancement of small targets. In response to the challenge of “low-slow small” UAV targets at long distances, we propose a multi-scale fusion super-resolution reconstruction (MFSRCNN) algorithm based on the fast super-resolution reconstruction (FSRCNN) algorithm and multi-scale fusion. Our experiments confirm the feasibility of the algorithm in reconstructing detailed information of the UAV target. On average, the MFSRCNN reconstruction time is 0.028 s, with the average confidence before and after reconstruction being 80.73% and 86.59%, respectively, resulting in an average increase of 6.72%.https://www.mdpi.com/2079-9292/12/7/1732object detectiondeep learningvisible light targetinfrared target |
| spellingShingle | Yuxing Dong Yan Li Zhen Li Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning Electronics object detection deep learning visible light target infrared target |
| title | Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning |
| title_full | Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning |
| title_fullStr | Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning |
| title_full_unstemmed | Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning |
| title_short | Research on Detection and Recognition Technology of a Visible and Infrared Dim and Small Target Based on Deep Learning |
| title_sort | research on detection and recognition technology of a visible and infrared dim and small target based on deep learning |
| topic | object detection deep learning visible light target infrared target |
| url | https://www.mdpi.com/2079-9292/12/7/1732 |
| work_keys_str_mv | AT yuxingdong researchondetectionandrecognitiontechnologyofavisibleandinfrareddimandsmalltargetbasedondeeplearning AT yanli researchondetectionandrecognitiontechnologyofavisibleandinfrareddimandsmalltargetbasedondeeplearning AT zhenli researchondetectionandrecognitiontechnologyofavisibleandinfrareddimandsmalltargetbasedondeeplearning |