Infrared Small Target Detection Method Based on Multi-Scale Feature Fusion
-
摘要: 红外小目标检测因其探测距离远、抗干扰能力强等特点,在空中目标探测与跟踪系统中得到了广泛的应用。针对目前红外小目标检测算法在复杂背景下检测准确率低、虚警率高等缺点。提出了一种基于多尺度特征融合的端到端红外小目标检测模型(multi-scale feature fusion single shot multibox detecto,MFSSD)。考虑到红外小目标的特点,通过细化和融合特征图的方法提出了一种特征融合模块,通过SP模块提高特征图不同通道的相关性,3种不同序列红外图像的实验结果表明,该算法在红外小目标检测中的平均检测精度高达87.8%。与传统的多尺度目标检测算法相比,准确率和召回率都有显著提高。Abstract: Infrared small target detection is widely used in aerial target detection and tracking systems owing to its long detection range and strong anti-jamming ability. Aiming at to overcome the shortcomings of the current infrared small target detection algorithm, such as a low precision rate and high false alarm rate when dealing with complex backgrounds, we propose an end-to-end infrared small target detection model (called MFSSD) based on multi-scale feature fusion. Considering the traits of the targets, we propose a feature fusion module using a refinement and fusion feature map method and improve the correlation of different channels through the SP module. The experimental results of three different sequences of infrared image detection show that the average detection accuracy of the MFSSD algorithm for infrared small target detection was as high as 87.8%. Compared with those of the traditional multi-scale target detection algorithm, both the precision rate and recall rate have been significantly improved.
-
Key words:
- attention mechanism /
- infrared small target /
- SSD /
- multi-scale feature fusion
-
表 1 红外小目标数据集描述
Table 1. Details of the infrared small target dataset
Name Total number Image resolution Detail Data1 399 256×256 The background is a sky back-ground
with varying degrees of thermal noise and a single targetData2 100 256×256 Background is the intersection of sky
and ground background, a single targetData3 998 256×256 The background is a sky back-ground with two
targets and cross flying表 2 实验中的比较算法
Table 2. The comparison algorithms in the experiment
Model Model description 1 SSD 2 SSD+FFM(FFM module
adopts up-sampling and down-sampling methods for fusion)3 SSD +FFM(FFM module
adopts subpixel convolutional layer and path layer methods for fusion)4 SSD + FFM(FFM module
adopts up-sampling and down-sampling methods for fusion) + SP module5(ours) SSD+FFM(FFM module
adopts subpixel convolutional layer and path layer methods for fusion)+ SP module表 3 不同网络算法的性能比较
Table 3. Comparison of algorithm performance of different networks
Model Input Train Test Map Fps 1 256 1297 200 82.5 29 2 256 1297 200 85.5 23 3 256 1297 200 86.2 25 4 256 1297 200 86.1 15 5 256 1297 200 87.8 17 -
[1] SHAO Xiaopeng, FAN Hua, LU Guangxu, et al. An improved infrared dim and small target detection algorithm based on the contrast mechanism of human visual system[J]. Infrared Physics & Technology, 2012, 55(5): 403-408. [2] Coppo Peter. Simulation of fire detection by infrared imagers from geostationary satellites[J]. Remote Sensing of Environment, 2015, 162: 84-98. doi: 10.1016/j.rse.2015.02.016 [3] 王周春, 崔文楠, 张涛. 基于支持向量机的长波红外目标分类识别算法[J]. 红外技术, 2021, 43(2): 153-161.WANG Z, CUI W, ZHANG T. A long-wave infrared target classification and recognition algorithm based on support vector machine[J]. Infrared Technology, 2021, 43(2): 153-161. [4] ZHANG X, REN K, WAN M, et al. Infrared small target tracking based on sample constrained particle filtering and sparse representation[J]. Infrared Physics & Technology, 2017, 87: 72-82. http://www.sciencedirect.com/science/article/pii/S1350449517302864 [5] QIN Y, LI B. Effective Infrared Small Target Detection Utilizing a Novel Local Contrast Method[J]. IEEE Geoscience & Remote Sensing Letters, 2016: 1890-1894. http://ieeexplore.ieee.org/document/7725517 [6] DENG L, HU Z, QUAN Z, et al. Adaptive top-hat filter based on quantum genetic algorithm for infrared small target detection[J]. Multimedia Tools & Applications, 2017(6): 1-13. doi: 10.1007/s11042-017-4592-2 [7] 王瑞, 朱志宇, 张冰. 基于人类视觉机制和粒子滤波的红外小目标跟踪[J]. 江苏科技大学学报: 自然科学版, 2018, 32(4): 74-78. https://www.cnki.com.cn/Article/CJFDTOTAL-HDCB201804013.htmWANG R, ZHU Z, ZHANG B. Infrared small target tracking based on human vision mechanism and particle filtering[J]. Journal of Jiangsu University of Science and Technology: Natural Science Edition, 2018, 32(4): 74-78. https://www.cnki.com.cn/Article/CJFDTOTAL-HDCB201804013.htm [8] Krizhevsky A, Sutskever I, Hinton G. ImageNet Classification with Deep Convolutional Neural Networks[C]//NIPS. Curran Associates Inc., 2012: 84-90. [9] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J/OL]. Computer Science, 2014, https://www.oalib.com/paper/4068791. [10] HE K, ZHANG X, REN S, et al. Identity Mappings in Deep Residual Networks[M]. Springer, Cham, 2016. [11] Szegedy C, WEI L, JIA Y, et al. Going Deeper with Convolutions[J]. IEEE Computer Society, 2015(6): 19. [12] Girshick R, Donahue J, Darrell T, et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[J]. IEEE Computer Society, 2013, 81: 580-587. [13] Girshick R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision (ICCV), 2015: 1440-1448. [14] REN S, HE K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137-1149. http://www.tandfonline.com/servlet/linkout?suffix=CIT0014&dbid=8&doi=10.1080%2F2150704X.2018.1475770&key=27295650 [15] Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-Time Object Detection[C]//Computer Vision & Pattern Recognition. IEEE, 2016: 779-788. [16] LIU W, Anguelov D, Erhan D, et al. SSD: Single Shot MultiBox Detector[J]. European Conference on Computer Vision, 2016: 21-37. doi: 10.1007/978-3-319-46448-0_2 [17] 李慕锴, 张涛, 崔文楠. 基于YOLOv3的红外行人小目标检测技术研究[J]. 红外技术, 2020, 42(2): 176-181. https://www.cnki.com.cn/Article/CJFDTOTAL-HWJS202002014.htmLI M K, ZHANG T, CUI W. Research on Infrared Pedestrian Small Target Detection Based onYOLOv3[J]. Infrared Technology, 2020, 42(2): 176-181. https://www.cnki.com.cn/Article/CJFDTOTAL-HWJS202002014.htm [18] 于周吉. 基于卷积神经网络的红外弱小目标检测算法[J]. 光学与光电技术, 2020, 18(5): 63-67. https://www.cnki.com.cn/Article/CJFDTOTAL-GXGD202005012.htmYU Z J. Infrared dim target detection algorithm based on convolutional neural network[J]. Optics & Optoelectronic Technology, 2020, 18(5): 63-67. https://www.cnki.com.cn/Article/CJFDTOTAL-GXGD202005012.htm [19] SHI M, WANG H, Infrared Dim and Small Target Detection Based on Denoising Autoencoder Network[J]. Mobile Networks and Applications, 2020, 25(4): 1469-1483. doi: 10.1007/s11036-019-01377-6 [20] Ryu J, Kim S. Data Driven Proposal and Deep Learning-based Small Infrared Drone Detection[J]. Journal of Institute of Control, Robotics and Systems, 2018, 24: 1146-1151. doi: 10.5302/J.ICROS.2018.18.0157 [21] Ryu J, Kim S. Heterogeneous Gray-Temperature Fusion-Based Deep Learning Architecture for Far Infrared Small Target Detection[J]. Journal of Sensors, 2019(7): 1-15. http://www.researchgate.net/publication/335356131_Heterogeneous_Gray-Temperature_Fusion-Based_Deep_Learning_Architecture_for_Far_Infrared_Small_Target_Detection [22] Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[J]. IEEE, 2017(11): 6517-6525. http://ieeexplore.ieee.org/document/8100173 [23] SHI W, Caballero J, F Huszár, et al. Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network[J]. IEEE, 2016(12): 1063-6919. http://ieeexplore.ieee.org/document/7780576