基于海思Hi3531部署的红外小目标检测算法研究

doi:10.3969/j.issn.1000-5641.2025.01.012

摘要/Abstract

摘要：

针对现有算法计算量大、实时性差、部署困难等问题, 同时为满足红外探测系统对实时性及准确率的高要求, 提出了一种部署于国产嵌入式芯片的轻量化算法, 即YOLOv5-TinyHisi. YOLOv5-TinyHisi算法根据红外小目标特点对主干网络结构进行轻量化改造, 并使用SIoU优化损失函数中的边界误差, 提高了红外小目标定位的准确性. 将YOLOv5-TinyHisi算法模型部署到海思Hi3531DV200嵌入式开发板上, 利用芯片集成的神经网络加速引擎 (neural network inference engine, NNIE) 对网络推理进行加速. 在公开数据集上的实验结果表明, 该算法能够大幅度降低参数量和模型大小, 与YOLOv5相比, 在平均精度上的提升了1.52%. 在海思Hi3531DV200嵌入式开发板上对分辨率为 (1280×512)像素的单张图像推理速度可达到35帧/s, 召回率可达到95%, 满足了红外探测系统对实时性和准确率的要求.

关键词: 红外小目标检测, 嵌入式系统, YOLOv5, 神经网络加速引擎

Abstract:

In response to the existing shortcomings of large computational complexity, poor real-time performance, and deployment difficulties in current algorithms, and to meet the high requirements of real-time performance and accuracy for infrared detection systems, proposes a lightweight algorithm deployed on domestically produced embedded chips, termed YOLOv5-TinyHisi. The YOLOv5-TinyHisi algorithm undertakes lightweight modifications to the backbone network structure based on the characteristics of infrared small targets. Additionally, it utilizes SIoU optimized loss function for boundary error, thereby enhancing the accuracy of infrared small target localization. The YOLOv5-TinyHisi algorithm model is deployed on Hi3531DV200, utilizing the chip-integrated neural network inference engine (NNIE) to accelerate network inference. Experimental results on public datasets demonstrate that the algorithm achieves a 1.52% improvement in average precision (mAP) compared to YOLOv5, while significantly reducing parameter count and model size. On the Hi3531DV200, the inference speed for a single image with a resolution of (1280 × 512) pixels reaches 35 frames per second (FPS), with a recall rate of 95%, meeting the real-time and accuracy requirements of the infrared detection system.

Key words: infrared small target detection, embedded system, YOLOv5, neural network inference engine

中图分类号:

TP391

傅晓雪, 黄昶. 基于海思Hi3531部署的红外小目标检测算法研究[J]. 华东师范大学学报（自然科学版）, 2025, 2025(1): 151-164.

Xiaoxue FU, Chang HUANG. Infrared small target detection algorithm deployed on HiSilicon Hi3531[J]. J* E* C* N* U* N* S*, 2025, 2025(1): 151-164.

图/表 15

图1

图2

图3

图4

图5

图6

图7

表1

图8

图9

图10

图11

表2

图12

表3

参考文献 27

1	张敏, 韩芳, 康键, 等.. 红外热成像技术在民用领域的应用. 红外, 2019, 40 (6): 35- 43.
2	GAO J, GUO Y, LIN Z, et al.. Robust infrared small target detection using multiscale gray and variance difference measures. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2018, 11 (12): 5039- 5052.
3	ZHANG W, CONG M Y, WANG L P. Algorithms for optical weak small targets detection and tracking: Review [C]// Proceedings of the International Conference on Neural Networks and Signal Processing. 2003: 643-647.
4	DU J, LU H, ZHANG L, et al. Infrared small target detection and tracking method suitable for different scenes [C]// Proceedings of the 2020 IEEE 9th Joint International Information Technology and Artificial Intelligence Conference. 2020: 664-668.
5	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]// Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition. 2014: 580-587.
6	蒋昕昊, 蔡伟, 杨志勇, 等.. 基于YOLO-IDSTD算法的红外弱小目标检测. 红外与激光工程, 2022, 51 (3): 502- 511.
7	TIAN Z, SHEN C, CHEN H, et al. FCOS: Fully convolutional one-stage object detection [C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. 2019: 9626-9635.
8	WANG K, LI S, NIU S, et al.. Detection of infrared small targets using feature fusion convolutional network. IEEE Access, 2019, (7): 146081- 146092.
9	HOU Q, WANG Z, TAN F, et al.. RISTDnet: Robust infrared small target detection network. IEEE Geoscience and Remote Sensing Letters, 2022, 19, 7000805.
10	高蕾, 符永铨, 李东升, 等.. 我国人工智能核心软硬件发展战略研究. 中国工程科学, 2021, 23 (3): 90- 97.
11	高昕. 基于Hi3559的智能相机系统软件研发 [D]. 杭州: 浙江大学, 2022.
12	张思雨. 基于相关方法的自适应目标跟踪算法研究与实现 [D]. 武汉: 华中科技大学, 2022.
13	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 936-944.
14	LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 8759-8768.
15	REN S, HE K, GIRSHICK R, et al.. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39 (6): 1137- 1149.
16	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression [EB/OL]. (2019-11-19)[2023-12-16]. https://arxiv.org/pdf/1911.08287.
17	GEVORGYAN Z. SIoU loss: More powerful learning for bounding box regression [EB/OL]. (2022-05-25)[2023-12-13]. https://arxiv.org/pdf/2205.12740.
18	ABADI M, BARHAM P, CHEN J, et al. TensorFlow: A system for large-scale machine learning [C]// Proceedings of the 12th USENIX Conference on Operating Systems Design and Implementation. 2016: 265-283.
19	PASZKE A, GROSS S, MASSA F, et al. PyTorch: An imperative style, high-performance deep learning library [C]// Proceedings of the 33rd International Conference on Neural Information Processing Systems. 2019: 8026-8037.
20	XIA X, LI J, WU J, et al. TRT-ViT: TensorRT-oriented vision transformer [EB/OL]. (2022-07-12)[2023-12-13]. https://arxiv.org/pdf/2205.09579.
21	AMBROSI J, ANKIT A, ANTUNES R, et al. Hardware-software co-design for an analog-digital accelerator for machine learning [C]// Proceedings of the IEEE International Conference on Rebooting Computing. DOI: 10.1109/ICRC.2018.8638612.
22	JIA Y, SHELHAMER E, DONAHUE J, et al. Caffe: Convolutional architecture for fast Feature embedding [C]// Proceedings of the 22nd ACM International Conference on Multimedia. 2014: 675-678.
23	GUO F, HUANG H, LIU Y, et al. Application of neural network based on Caffe framework for object detection in Hilens [C]// Proceedings of the Chinese Automation Congress. 2019: 4355-4359.
24	NEUBECK A, VAN GOOL L. Efficient non-maximum suppression [C]// Proceedings of the International Conference on Pattern Recognition. 2006: 850-855.
25	WANG H, ZHOU L, WANG L. Miss detection vs. false alarm: Adversarial learning for small object segmentation in infrared images [C]// Proceedings of the IEEE International Conference on Computer Vision. 2019: 8508-8517.
26	LI B, XIAO C, WANG L, et al.. Dense nested attention network for infrared small target detection. IEEE Transactions on Image Processing, 2023, 32, 1745- 1758.
27	范晨亮, 李国庆, 马长啸, 等.. 基于深度学习的风机叶片裂纹检测算法. 科学技术创新, 2020, (13): 72- 75.

数据集	图像数量/张	目标背景	目标类型
NUST-SIRST	10000	城市/海洋/多云	点目标/斑点状目标
NUDT-SIRST	1327	城市/密林/海洋/天空/多云	点目标/斑点状目标/扩展目标

模型名称	在NUST-SIRST上的模型性能指标			在NUDT-SIRST上的模型性能指标
模型名称	mAP/%	模型大小/MB	检测时间/ms	mAP/%	模型大小/MB	检测时间/ms
SSD	81.1	214.6	27.9	73.2	113.8	38.8
Faster R-CNN	92.3	412.1	45.4	90.1	315.0	57.3
YOLOv3	77.9	324.7	19.0	89.2	234.9	27.7
YOLOv4	81.2	88.7	12.3	83.6	24.3	18.7
YOLOv5	95.6	54.0	8.7	92.2	13.6	13.2
YOLOv7	86.0	46.3	6.2	81.6	12.2	10.0
YOLOv5-TinyHisi	97.0	3.7	7.6	93.4	1.1	8.6

算法名称	数据指令文件大小/MB	后级处理时间/ms	识别帧率/帧·s^–1
Faster R-CNN	79.852	1.589	10
SSD	63.124	1.328	13
YOLOv3	59.177	1.032	10
YOLOv5	7.780	0.794	21
YOLOv5-TinyHisi	0.638	0.501	35