Small Infrared Target Detection Based on Fully Convolutional Network
-
摘要: 在小天体探测、导弹制导和战场侦察等航空航天领域，由于目标信号较弱，占有像素数少，缺少目标形状和纹理信息，使用手工特征提取的传统算法容易出现大量虚警，而拥有强大特征提取能力的深度学习算法无法对微小且缺乏轮廓信息的目标训练。本文采用了滑动窗口取样训练，它源自基于人类视觉特性的传统目标检测算法中嵌套结构的思想，设计了一种使用递归卷积层的全卷积网络，在不增加额外训练参数的情况下，扩展了模型的网络深度，该网络的并行卷积结构的多个分支网络模拟了传统算法的多尺度操作，有利于在复杂环境中增强目标和背景之间的对比度，并且设计使用了多种损失函数的组合，以对抗正负样本严重不平衡的问题。实验结果表明：该方法实现了比传统方法更好的检测效果，为此领域的研究者们提供了一个新的思路和解决途径。Abstract: In the field of aerospace research, such as in small celestial body detection, missile guidance, and battlefield reconnaissance, because the target signal is weak, the number of pixels occupied is small, and the target lacks shape structure and texture information, traditional algorithms with manual feature extraction are prone to false alarms, whereas deep learning methods with powerful feature extraction capabilities cannot train tiny targets that lack contour information. In this context, a sliding window sampling training method is adopted, which originates from the idea of nested structures in traditional algorithms based on human visual characteristics. A fully convolutional network using recursive convolutional layers is designed to extend the depth of the network without increasing the training parameters. The multi-branch structure of the network's parallel convolution structure simulates the multi-scale operation of the traditional algorithm, which can enhance the contrast between the target and the background. Additionally, various loss functions are designed to combat the serious imbalance between positive and negative samples. The results show that the algorithm achieves a better detection performance than the traditional algorithms.
-
-
![]()
图 1 滑动窗口为3，滑动步长为2的取样窗口示意图
Figure 1. The illustration of sampling window with size 3 and sliding step with 2
![]()
图 3 递归模块的展开结构，相同卷积参数应用不同递归层
Figure 3. Unfolding recursive module, the same filter W is applied to feature maps recursively
![]()
图 6 不同检测方法在5张测试图像上的滤波结果，矩形框表示目标，圆圈表示滤波后的噪声
Figure 6. The representative results of different methods on five test images, the rectangles denote the targets and the circles are representative examples of noise
![]()
图 7 不同检测方法在红外图像上的滤波结果，矩形框表示目标，圆圈表示滤波后的噪声
Figure 7. Different methods on infrared images, the rectangles denote the targets and the circles are representative examples of noise
表 1 本文使用的F-CNN网络结构
Table 1 F-CNN architecture for semantic segmentation
Layers Output size Layer configurations Feature extraction 48×48 $\begin{gathered} \left[ {\begin{array}{*{20}{c}} {3 \times 3, 32} \\ {3 \times 3, 16} \end{array}} \right] \times 2 \\ \left[ {3 \times 3, 32} \right] \times 1 \\ \end{gathered} $ Recursive block 48×48 $\left[ {3 \times 3, 32} \right] \times 4$ Reconstruction module 48×48 $\left[ {\begin{array}{*{20}{c}} {3 \times 3, 32} \\ {3 \times 3, 32} \\ {3 \times 3, 16} \\ {3 \times 3, 2} \end{array}} \right] \times 1$ 
下载: 导出CSV
表 2 无人机测试图像的SCR值和目标像素数
Table 2 The SCR and target size of UAV test images.
Test images Image 1 Image 2 Image 3 Image 4 Image 5 SCR 4.009 2.337 8.378 4.411 2.976 Target size/Pixel 15 25 12 13 28
下载: 导出CSV
表 3 不同方法对图 6第一列测试图像滤波结果的SCRG和BSF值
Table 3 The evaluation results of SCRG and BSF of different methods for images in the first column in Fig. 6
Methods Image 1 Image 2 Image 3 Image 4 Image 5 BSF SCRG BSF SCRG BSF SCRG BSF SCRG BSF SCRG MS-AAGD 0.596 6.908 0.639 5.272 0.298 3.440 0.690 6.764 0.952 6.178 LoG 0.193 1.526 0.773 4.517 0.099 1.030 0.243 1.791 0.306 2.379 MPCM 0.793 8.577 1.395 8.202 0.358 3.547 0.732 7.201 0.985 6.401 F-CNN 0.875 8.655 1.419 13.684 0.305 3.694 0.764 8.821 1.195 11.137
下载: 导出CSV
表 4 不同方法对图 7中红外图像滤波结果的SCRG和BSF值
Table 4 The evaluation results of SCRG and BSF of different methods for infrared images in Fig. 7
Methods Image 1 Image 2 Image 3 Image 4 BSF SCRG BSF SCRG BSF SCRG BSF SCRG BSF SCRG MS-AAGD 2.110 11.483 0.712 18.757 1.872 3.998 2.56 6.298 1.091 16.575 LoG 0.853 4.967 0.473 15.559 1.279 3.042 1.358 3.408 0.241 4.298 MPCM 1.758 12.399 2.011 96.279 1.984 4.088 4.142 8.333 0.966 18.781 F-CNN 2.485 16.337 3.314 63.100 1.710 5.407 2.465 10.528 2.404 15.440
下载: 导出CSV
-
[1] DESHPANDE S D, ER M H, VENKATESWARLU R, et al. Max-mean and max-median filters for detection of small targets[C]//Proceedings of SPIE, 1999, 3809: 74-83.
[2] ZENG M, LI J X, PENG Z, et al. The design of top-hat morphoLoGical filter and application to infrared target detection[J]. Infrared Physics and Technology, 2006, 48(1): 67-76. DOI: 10.1016/j.infrared.2005.04.006
[3] YANG L, YANG J, YANG K, et al. Adaptive detection for infrared small target under sea–sky complex background[J]. Electronics Letters, 2004, 40(17): 1083-1085. DOI: 10.1049/el:20045204
[4] HILLIARD C I. Selection of a clutter rejection algorithm for real-time target detection from an airborne platform[C]//Proceedings of SPIE, 2000, 4048: 74-84.
[5] REED I, GAGLIARDI R, STOTTS L, et al. Optical Moving Target Detection with 3-D Matched Filtering[J]. IEEE Transactions on Aerospace and Electronic Systems, 1988, 24(4): 327–336. DOI: 10.1109/7.7174
[6] WEI Y T, YOU X G, LI H, et al. Multiscale patch-based contrast measure for small infrared target detection[J]. Pattern Recognition, 2016, 58: 216-226. DOI: 10.1016/j.patcog.2016.04.002
[7] DENG H, SUN X P, LIU M L, et al. Infrared small-target detection using multiscale graydifference weighted image entropy[J]. IEEE Transactions on Aerospace and Electronic Systems, 2016, 52(1): 60-72. DOI: 10.1109/TAES.2015.140878
[8] KIM S H, LEE J Y. Scale invariant small target detection by optimizing signal-to-clutter ratio in heterogeneous background for infrared search and track[J]. Pattern Recognition, 2012, 45: 393-406. DOI: 10.1016/j.patcog.2011.06.009
[9] 沈旭, 程小辉, 王新政. 结合视觉注意力机制基于尺度自适应局部对比度增强的红外弱小目标检测算法[J]. 红外技术, 2019, 41(8): 764-771. http://hwjs.nvir.cn/article/id/hwjs201908012 SHEN X, CHENG X H, WANG X ZH. Infrared dim-small object detection algorithm based on adaptive scale local constrast enhancement combined with visual attention mechanism[J]. Infrared Technology, 2019, 41(8): 764-771. http://hwjs.nvir.cn/article/id/hwjs201908012
[10] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, New York: 2016: 770-778.
[11] KIM J, LEE J K, Lee K M, et al. Deeply-recursive convolutional network for image super-resolution[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, New York: 2016: 1637-1645.
[12] LI J, YE D H, CHUNG T, et al. Multi-target detection and tracking from a single camera in unmanned aerial vehicles (UAVs)[C] //Proceedings of 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2016: 4992-4997.
[13] GAO C, ZHANG T, LI Q, et al. Small infrared target detection using sparsering representation[J]. IEEE Aerospace and Electronic Systems Magazine, 2012, 27(3): 21-30. DOI: 10.1109/MAES.2012.6196254
[14] QU X J, CHEN H, PENG G H, et al. Novel detection method for infrared smalltargets using weighted information entropy[J]. Journal of Systems Engineering and Electronics, 2012, 23(6): 838-842. DOI: 10.1109/JSEE.2012.00102
-
期刊类型引用(8)
1. 马鑫，李丰睿，倪亮，李明涛，徐赤东，王楷铎，周炳红，李国浩. 激光烧蚀偏转小行星轨道低成本在轨试验方案设计. 空间科学与试验学报. 2024(04): 47-54 . 
百度学术
2. 张睿，李允臣，王家宝，李阳，苗壮. 基于深度学习的红外目标检测综述. 计算机技术与发展. 2023(11): 1-8 . 百度学术
3. 代牮，赵旭，李连鹏，刘文，褚昕悦. 基于改进YOLOv5的复杂背景红外弱小目标检测算法. 红外技术. 2022(05): 504-512 . 本站查看
4. 刘衍会，张宏武. 强杂波下微弱目标检测方法研究. 现代雷达. 2022(07): 31-37 . 百度学术
5. 陈世琼. 基于非线性抗噪声估计的视觉显著性弱小目标检测. 红外与激光工程. 2022(09): 435-440 . 百度学术
6. 胡焱，胡皓冰，赵宇航，原子昊，司成可. 红外热成像中低分辨率行人小目标检测方法. 红外技术. 2022(11): 1146-1153 . 本站查看
7. 田栋，刘坤，郝博. 基于LIG的红外视频小目标检测方法. 机械制造与自动化. 2022(06): 226-230 . 百度学术
8. 蔡云泽，张彦军. 基于双通道特征增强集成注意力网络的红外弱小目标检测方法. 空天防御. 2021(04): 14-22 . 百度学术
其他类型引用(3)
计量
- 文章访问数: 631
- HTML全文浏览量: 88
- PDF下载量: 119
- 被引次数: 11
