针对红外弱小目标探测在边缘端部署时探测精度低、算力开销大及资源调度失衡等问题，提出一种面向可重构边缘端部署的红外目标检测算法与软硬协同加速方法。在算法层面，构建轻量化探测网络SCAD-SSD，利用高分辨率自适应与先验框特化（HAAS）策略重构采样密度，缓解微弱信号湮灭；设计稀疏上下文感知模块（SCB）挖掘环境线索，强化信号响应；引入自适应辨识拒绝模块（ADR）构建特征过滤机制，抑制背景虚警。在硬件层面，构建基于SoC的并行加速架构：设计多维并行访存单元（DPAU）优化片内数据复用与流式传输；提出参数化重构引擎（UPRE）并配合全局任务调度机制，通过算力资源动态重组与双引擎负载均衡，有效掩蔽非规整计算导致的流水线阻塞。实验结果表明：SCAD-SSD在IRSTD-UAV官方基准下， 达97.74%，召回率达96.49%，在GPU端的推理速度达104.0 FPS，并具备良好的跨场景泛化潜力；在仅有3.69M参数量与6.64 GFLOPs计算量的极低开销下，硬件加速器在Zynq-7020平台上实现了71.05 GOPS的算力吞吐与10.7 FPS的实际运行帧率，能效比达21.02 GOPS/W。所提方法在提升探测灵敏度的同时，满足边缘端实时探测与可靠防御的工程需求。
To address the issues of low detection accuracy, high computational cost, and resource scheduling imbalance in the edge deployment of infrared small target detection, a hardware-software co-design acceleration method for reconfigurable edge deployment is proposed. At the algorithmic level, a lightweight network, SCAD-SSD, is constructed: the High-resolution Adaptation and Anchor Specialization (HAAS) strategy is utilized to reconstruct sampling density and alleviate weak signal annihilation; the Sparse Contextual awareness Block (SCB) is designed to mine environmental cues for signal enhancement; and the Adaptive Discriminative Rejection (ADR) module is integrated to build a feature filtering mechanism for background suppression. At the hardware level, a SoC-based parallel acceleration architecture is developed: a Dimensional Parallel-Access Unit (DPAU) is designed to optimize on-chip data reuse; a Universal Parameterized Reconfigurable Engine (UPRE) with a global task scheduling mechanism is proposed to effectively mask pipeline stalls through dynamic resource reorganization and dual-engine load balancing. Experimental results show that SCAD-SSD achieves 97.74% AP50 and 96.49% recall on the IRSTD-UAV benchmark, with a inference speed of 104.0 FPS on the GPU, and shows good cross-scenario generalization potential. With a minimal overhead of 3.69 M parameters and 6.64 GFLOPs, the hardware accelerator achieves a throughput of 71.05 GOPS and an actual running frame rate of 10.7 FPS on the Zynq-7020 platform, yielding an energy efficiency of 21.02 GOPS/W. The proposed method improves detection sensitivity while meeting the requirements of real-time detection and reliable defense at the edge.
[1]王恒慧, 曹东, 赵杨, 等.基于深度学习的红外弱小目标检测算法研究综述[J].激光与红外, 2022, 52(9):1274-1279
[2]WANG H H, CAO D, ZHAO Y, et al.Survey of infrared dim small target detection algorithm based on deep learning[J].Laser & Infrared, 2022, 52(9):1274-1279
[3]LI R H, SHEN Y.YOLOSR-IST: A deep learning method for small target detection in infrared remote sensing images based on super-resolution and YOLO[J].[J].Signal Processing, 2023, 208(无):108962-108962
[4]DENG H M, ZHANG Y.FMR-YOLO: Infrared ship rotating target detection based on synthetic fog and multiscale weighted feature fusion[J].[J].IEEE Transactions on Instrumentation and Measurement, 2024, 73(无):5001717-5001717
[5]FENG Y F, HUANG J, DU S Y, et al.Hyper-YOLO: When visual object detection meets hypergraph computation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(4):2388-2401
[6]LYU H T, QIU F, AN L, et al.Deer survey from drone thermal imagery using enhanced faster R-CNN based on ResNets and FPN[J][J].Ecological Informatics, 2024, 79(无):102383-102383
[7]OU J H, WANG J G, XUE J, et al.Infrared image target detection of substation electrical equipment using an improved Faster R-CNN[J].IEEE Transactions on Power Delivery, 2023, 38(1):387-396
[8]WU X, HONG D F, CHANUSSOT J.UIU-Net: U-Net in U-Net for infrared small object detection[J][J].IEEE Transactions on Image Processing, 2023, 32(无):364-376
[9]YUAN S, QIN H L, YAN X, et al.SCTransNet: Spatial-channel cross transformer network for infrared small target detection[J][J].IEEE Transactions on Geoscience and Remote Sensing, 2024, 62(无):5002615-5002615
[10]ZHANG M J, OUYANG Y J, GAO F, et al.MOCID: Motion context and displacement information learning for moving infrared small target detection[C]∥Proceedings of the 39th AAAI Conference on Artificial Intelligence. Philadelphia: AAAI Press, 2025: 10022-10030.
[11]FANG H, GUO S, CHEN Q, et al.Spatio-temporal context learning with temporal difference convolution for moving infrared small target detection[C]∥Proceedings of the 40th AAAI Conference on Artificial Intelligence. Philadelphia: AAAI Press, 2026.
[12]武宏程, 黄家明, 张冰逸, 等.一种红外目标检测低比特量化神经网络方法及其 实现[J].上海航天中英文, 2023, 40(1):35-43
[13]WU H C, HUANG J M, ZHANG B Y, et al.A low-bit quantization neural network method for infrared object detection and its FPGA implementation[J].Aerospace Shanghai (Chinese & English), 2023, 40(1):35-43
[14]孙小坚, 林瑞全, 方子卿, 等.基于 加速的低功耗的 网络识别系统[J].计算机测量与控制, 2023, 31(5):221-227
[15]SUN X J, LIN R Q, FANG Z Q, et al.Low power MobileNetV2 network recognition system based on FPGA acceleration[J].Computer Measurement & Control, 2023, 31(5):221-227
[16]CHENG Y B, LAI X F, XIA Y C.Design and optimization of an FPGA-based infrared dim small target detection network under a sky cloud background[J].Applied Sciences, 2025, 15(9):4634-4634
[17]LIU Y Y, DU H, WU Y, et al.FPGA accelerated deep learning for industrial and engineering applications: Optimal design under resource constraints[J].Electronics, 2025, 14(4):703-703
[18]ADIONO T, PRADANA A S, HARIMURTI S, et al.Low latency YOLOv3-tiny accelerator for low-cost FPGA using general matrix multiplication principle[J][J].IEEE Access, 2020, 8(无):196843-196856
[19]PAN Y J, ZHANG L, ZHANG Y J.Rapid detection of PCB defects based on YOLOx-Plus and FPGA[J][J].IEEE Access, 2024, 12(无):61343-61358
[20]LI Y D, DONG H, LI H G, et al.Multi-block SSD based on small object detection for UAV railway scene surveillance[J].Chinese Journal of Aeronautics, 2020, 33(6):1747-1755
[21]SUH H S, MENG J, NGUYEN T, et al.Algorithm-hardware co-optimization for energy-efficient drone detection on resource-constrained FPGA[J].ACM Transactions on Reconfigurable Technology and Systems, 2023, 16(2):33-33
[22]JOCHER G, CHABIUK A, TOMAS J, et al.Ultralytics YOLOv8[CP/OL][J].https://github.com/ultralytics/ultralytics, 2023, 无(无):无-无
[23]SUN H, BAI J X, YANG F, et al.direction induced attention network for infrared dim small target detection with a large-scale dataset IRDST[J][J].IEEE Transactions on Geoscience and Remote Sensing, 2023, 61(无):5000513-5000513
[24]SUN H, BAI J X, YANG F, et al.Receptive-field and direction induced attention network for infrared dim small target detection with a large-scale dataset IRDST[J][J].IEEE Transactions on Geoscience and Remote Sensing, 2023, 61(无):5000513-5000513
CJA