High Power Laser and Particle Beams, Volume. 37, Issue 5, 051001(2025)
A review of multispectral target detection algorithms and related datasets
Figures & Tables(10)
![Schematic diagram of complementarity between visible light and infrared images[15-16]](/Images/icon/loading.gif){kind=icon}
Fig. 2. Flowchart of the multispectral object detection dataset, subfigure is the classification of imaging spectrometers
Fig. 4. Evolutionary process of object detection algorithms, subfigures present the evolution process of key technologies
Fig. 8. Schematic diagram of classification of multispectral object detection algorithms
Table 1. Commonly used multispectral object detection data sets
View tableView in ArticleTable 1. Commonly used multispectral object detection data sets
establishment time brief introduction application domain image acquisition equipment Indian Pines[ 18 ]1992 Covering 400 - 2500 nm with a total of 224 bands.The annotations include 16 categories such as Oats and Woods. agricultural detection, environmental monitoring, etc. Avris (whiskbroom) Salinas[ 27 ]2001 Covering a range of 400 - 2500 nm with a total of 224 bands.The annotations include 16 categories such as fallow land and celery. Agricultural detection, environmental monitoring, etc. Avris (whiskbroom) FLIR[ 15 ]2003 9711 thermal imaging pictures and9233 RGBpictures, covering 15 common object categories. Security monitoring, autonomous driving. FLIR (filter) Pavia Centre & University[ 17 ]2010 The spectrum covers visible light to near-infrared, with a total of 103 bands. The annotations include 9 categories such as water and trees. Land use and cover detection, environmental detection ROSIS (pushbroom) VEDAI[ 20 ]2014 A dataset for vehicle detection captured by drones, providing different resolutions for each scene. traffic monitoring, urban planning / KAIST[ 21 ]2015 A total of 95 328 pictures, each picture contains two versions of RGBand infrared images, and there are a total of 103 128 densely annotated ones. pedestrian monitoring PointGrey Flea3 and FLIR (filter) EuroSAT[ 29 ]2019 The dataset is based on Sentinel-2 satellite images covering 13 spectral bands and consisting out of 10 classes with in total 27 000 labeled and geo-referenced images. land use and cover classification MSI on Sentinel-2 satellite (pushbroom) LLVIP[ 16 ]2021 A pedestrian detection dataset in low-light environments, with 15488 pairs of pixel-aligned RGB-IR images.pedestrian monitoring Hikvision Binocular Photography Platform (filter) MMShip[ 52 ]2023 The dataset encompasses 5 016 sets of ship target images, with each scene consisting of a total of four bands, including the visible light and the near infrared. ship monitoring MSI on Sentinel-2 satellite (pushbroom) SMOD[ 53 ]2024 The dataset comprises 5 378 pairs of daytime scene images captured at 3 p.m. and 3 298 pairs of nighttime ones at 7 p.m. pedestrian monitoring, traffic monitoring Asens FV6 Binocular Photography Platform (filter)
Table 2. A summary of multispectral object detection algorithms
View tableView in ArticleTable 2. A summary of multispectral object detection algorithms
establishment time performance/% KAIST(MR)* LLVIP(mAP) VEDAI(mAP50) KAIST(MR)*：The data selects the ‘all’ subset under the KAIST dataset. IATDNN + IAMSS[ 11 ]2018.2 26.37 \ \ IAF R-CNN[ 12 ]2018.8 15.73 \ \ MBNet[ 13 ]2020.8 8.13 \ \ CFB(CFT)[ 95 ]2022.1 \ 63.6 85.3 SuperYOLO[ 19 ]2023.4 \ \ 75.09 CFMW[ 10 ]2024.4 \ 64.8 \ RSDet[ 99 ]2024.5 26.02 61.3 \
Tools
Get Citation
Copy Citation Text
Tianlong Zhang, Yuanchao Geng, Yuzhen Liao, Dangpeng Xu. A review of multispectral target detection algorithms and related datasets[J]. High Power Laser and Particle Beams, 2025, 37(5): 051001
Paper Information
Category:
Received: Oct. 22, 2024
Accepted: Mar. 3, 2025
Published Online: May. 22, 2025
The Author Email: Dangpeng Xu (xudp803810@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20