基于雙重YOLOv8-pose模型的探地雷達(dá)雙曲線關(guān)鍵點(diǎn)檢測與目標(biāo)定位

侯斐斐; 彭應(yīng)昊; 董健; 銀雪

doi:10.11999/JEIT240242

基于雙重YOLOv8-pose模型的探地雷達(dá)雙曲線關(guān)鍵點(diǎn)檢測與目標(biāo)定位

doi: 10.11999/JEIT240242

1.
中南大學(xué)自動(dòng)化學(xué)院長沙 410083
2.
中南大學(xué)電子信息學(xué)院長沙 410004

基金項(xiàng)目: 國家自然科學(xué)基金(62406346)，湖南省自然科學(xué)基金(2022JJ30052)，長沙市自然科學(xué)基金(kq2208285)

詳細(xì)信息

作者簡介:
侯斐斐：女，講師，研究方向?yàn)樘降乩走_(dá)圖像解譯、無損探測、深度學(xué)習(xí)、目標(biāo)識(shí)別

彭應(yīng)昊：男，本科生，研究方向?yàn)閳D像識(shí)別與深度學(xué)習(xí)

董健：男，教授，研究方向?yàn)樘炀€、雷達(dá)信號(hào)處理、機(jī)器學(xué)習(xí)算法及其電磁應(yīng)用

銀雪：女，本科生，研究方向?yàn)橛?jì)算機(jī)視覺與目標(biāo)識(shí)別

通訊作者:
董健　dongjian@csu.edu.cn

中圖分類號(hào): TN957.51; TP391
計(jì)量
- 文章訪問數(shù): 611
- HTML全文瀏覽量: 260
- PDF下載量: 86
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2024-04-08
- 修回日期: 2024-09-12
- 網(wǎng)絡(luò)出版日期: 2024-09-19
- 刊出日期: 2024-11-10

Ground Penetrating Radar Hyperbolic Keypoint Detection and Object Localization Based on Dual YOLOv8-pose Model

1.
School of Automation, Central South University, Changsha 410083, China
2.
School of Electronic Information, Central South University, Changsha 410004, China

Funds: The National Natural Science Foundation of China (62406346), Hunan Provincial Natural Science Foundation (2022J30052), Changsha Natural Science Foundation (kq2208285)

摘要

摘要: 探地雷達(dá)(GPR)是一種可用于地下目標(biāo)識(shí)別的無損檢測方法。針對現(xiàn)有方法存在不同尺度目標(biāo)兼容性差、復(fù)雜圖像識(shí)別難度大、無法精確定位等問題，該文提出一種基于雙重YOLO姿態(tài)模型(YOLOv8-pose)的GPR雙曲線關(guān)鍵點(diǎn)檢測與目標(biāo)定位，命名為雙重YOLO關(guān)鍵點(diǎn)定位方法(DYKL)，用于地下目標(biāo)的檢測與精確定位。所提模型架構(gòu)包含兩個(gè)階段：首先，第1階段是基于YOLOv8-pose模型的GPR目標(biāo)檢測，以確定候選目標(biāo)的位置；接著，第1階段的部分訓(xùn)練權(quán)重被共享并傳遞到第2階段，后者以此為基礎(chǔ)繼續(xù)訓(xùn)練YOLOv8-pose網(wǎng)絡(luò)，用于候選目標(biāo)特征的關(guān)鍵點(diǎn)檢測及獲取，從而實(shí)現(xiàn)地下目標(biāo)的自動(dòng)化定位。通過與級(jí)聯(lián)區(qū)域卷積網(wǎng)絡(luò)(Cascade R-CNN)、更快的區(qū)域卷積網(wǎng)絡(luò)(Faster R-CNN)、實(shí)時(shí)對象檢測模型(RTMDet)以及“你只看一次”人臉模型(YOLOv7-face)4種先進(jìn)的深度模型進(jìn)行比較，所提模型平均識(shí)別準(zhǔn)確率達(dá)到98.8%，性能優(yōu)于其他模型。結(jié)果表明所提DYKL模型具有較高的識(shí)別準(zhǔn)確性與較強(qiáng)的魯棒性，可以為地下目標(biāo)的精確定位提供參考。
- 探地雷達(dá) /
- 目標(biāo)檢測 /
- 關(guān)鍵點(diǎn)檢測 /
- YOLOv8
Abstract: Ground Penetrating Radar (GPR) is identified as a non-destructive method usable for the identification of underground targets. Existing methods often struggle with variable target sizes, complex image recognition, and precise target localization. To address these challenges, an innovative method is introduced that leverages a dual YOLOv8-pose model for the detection and precise localization of hyperbolic keypoint. This method, termed Dual YOLOv8-pose Keypoint Localization (DYKL), offers a sophisticated solution to the challenges inherent in GPR-based target identification and positioning. The proposed model architecture includes two stages: firstly, the YOLOv8-pose model is employed for the preliminary detection of GPR targets, adeptly identifying regions that are likely to contain these targets. Secondly, building upon the training weights established in the first phase, the model further hones the YOLOv8-pose network. This refinement is geared towards the precise detection of keypoints within the candidate target features, thereby facilitating the automated identification and exact localization of underground targets with enhanced accuracy. Through comparison with four advanced deep-learning models— Cascade Region-based Convolutional Neural Networks (Cascade R-CNN), Faster Region-based Convolutional Neural Networks (Faster R-CNN), Real-Time Models for object Detection (RTMDet), and You Only Look Once v7(YOLOv7-face), the proposed DYKL model exhibits an average recognition accuracy of 98.8%, surpassing these models. The results demonstrate the DYKL model’s high recognition accuracy and robustness, serving as a benchmark for the precise localization of subterranean targets.
- Ground penetrating radar /
- Target detection /
- Keypoint detection /
- YOLOv8

HTML全文

圖 1 算法整體框架

下載: 全尺寸圖片幻燈片

圖 2 YOLOv8-pose網(wǎng)絡(luò)結(jié)構(gòu)圖

下載: 全尺寸圖片幻燈片

圖 3 關(guān)鍵點(diǎn)標(biāo)注

下載: 全尺寸圖片幻燈片

圖 4 2維坐標(biāo)系轉(zhuǎn)換

下載: 全尺寸圖片幻燈片

圖 5 不同階段下模型訓(xùn)練過程的損失曲線

下載: 全尺寸圖片幻燈片

圖 6 DYKL模型檢測結(jié)果

下載: 全尺寸圖片幻燈片

圖 7 不同關(guān)鍵點(diǎn)檢測算法在仿真圖像上的性能對比

下載: 全尺寸圖片幻燈片

圖 8 不同關(guān)鍵點(diǎn)檢測算法在實(shí)測圖像數(shù)據(jù)上的對比結(jié)果

下載: 全尺寸圖片幻燈片

圖 9 與傳統(tǒng)圖像處理算法在仿真圖像上的性能對比

下載: 全尺寸圖片幻燈片

表 1 GPR目標(biāo)自動(dòng)識(shí)別算法

文獻(xiàn)	模型類別	算法	檢測目標(biāo)	數(shù)據(jù)集	性能指標(biāo)(%)
蘭天等人^[34]	圖像學(xué)目標(biāo)檢測	基于擬合誤差消除雙曲線識(shí)別模型	/	實(shí)測圖像和仿真圖像	/
Pham等人^[16]	兩階段目標(biāo)檢測	Faster R-CNN	/	50張仿真圖像和100張實(shí)測圖像	/
Cui等人^[18]	兩階段目標(biāo)檢測	Faster R-CNN	地層界面	/	檢測框Pre= 98.3
Zhao等人^[20]	兩階段實(shí)例分割	Mask R-CNN	隧道襯砌上缺陷位置	/	檢測框Pre=96.1 掩模預(yù)測Pre=95.6
Hou等人^[21]	兩階段實(shí)例分割	基于MS—RCNN定制錨定方案	樹根	95張實(shí)測圖像	掩膜預(yù)測AP50=38.6
Wang等人^[26]	單階段目標(biāo)檢測	SSD引入FPN特征融合層網(wǎng)絡(luò)和廣義交并集損失	/	1 170張仿真圖像	檢測框Pre=92.0 檢測框Rec=90.3
Qiu等人^[28]	單階段目標(biāo)檢測	YOLOv5	鐵實(shí)驗(yàn)材料	412張實(shí)測圖像	檢測框Pre=82.6 檢測框Rec=68.0
Hu等人^[29]	單階段目標(biāo)檢測	YOLOv5引入注意力機(jī)制	地下缺陷	3 256張圖像包含實(shí)測圖像和仿真圖像	檢測框mAP=85.4
胡榮明等人^[30]	單階段目標(biāo)檢測	YOLOv7	隧道襯砌病害	506張仿真圖像和84張實(shí)測圖像	檢測框Pre=97.9 檢測框Rec=90.6
Wang等人^[31]	單階段目標(biāo)檢測	YOLOv8引入CBAM注意力機(jī)制	地下缺陷	837張實(shí)測圖像	檢測框mAP50=90.8 檢測框F1=88.3
Li等人^[32]	單階段關(guān)鍵點(diǎn)檢測	YOLOv4引入關(guān)鍵點(diǎn)回歸并增加Wing損失函數(shù)	植物根	1 000張仿真圖像和2 320張實(shí)測圖像	檢測框Pre=94.0 檢測框Rec=96.7
方濤濤等人^[33]	單階段關(guān)鍵點(diǎn)檢測	YOLOv8引入 CBAM 注意力機(jī)制和關(guān)鍵點(diǎn)回歸	地下管線	545張實(shí)測圖像和795張仿真圖像	定位水平誤差8.6 定位深度誤差1.8

下載: 導(dǎo)出CSV

表 2 模型輸入圖像的數(shù)量與尺寸

參數(shù)	目標(biāo)檢測階段	關(guān)鍵點(diǎn)檢測階段
批量大小	32	64
圖像尺寸	640×640	128×128

下載: 導(dǎo)出CSV

表 3 各種目標(biāo)檢測算法的平均精度與平均召回率

模型	mAP50↑	mAP50-95↑	Av_Recall↑
DYKL	0.988	0.642	0.960
Cascade^[40]	0.962	0.597	0.675
Faster R-CNN^[17]	0.940	0.546	0.638
RTMDet^[41]	0.905	0.535	0.719

下載: 導(dǎo)出CSV

參考文獻(xiàn)(42)

[1]	LI Jing, LIU Cai, ZENG Zhaofa, et al. Gpr signal denoising and target extraction with the ceemd method[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12(8): 1615–1619. doi: 10.1109/lgrs.2015.2415736.
[2]	李洪麗, 鹿琪. 探地雷達(dá)在LNAPL污染土壤探測中的應(yīng)用進(jìn)展研究[J]. 地球物理學(xué)進(jìn)展, 2020, 35(3): 1141–1148. doi: 10.6038/pg2020DD0414. LI Hongli and LU Qi. Progress in application of ground penetrating radar in LNAPL contaminated soil detection of LNAPL[J]. Progress in Geophysics, 2020, 35(3): 1141–1148. doi: 10.6038/pg2020DD0414.
[3]	尹德, 葉盛波, 張經(jīng)緯, 等. 公路結(jié)構(gòu)和介電特性對探地雷達(dá)反射回波的影響研究[J]. 電子測量技術(shù), 2018, 41(5): 51–56. doi: 10.19651/j.cnki.emt.1700792. YIN De, YE Shengbo, ZHANG Jingwei, et al. Study on the effect of highway structure and dielectric properties on GPR echo[J]. Electronic Measurement Technology, 2018, 41(5): 51–56. doi: 10.19651/j.cnki.emt.1700792.
[4]	王韻, 王紅雨, 常留成, 等. 基于探地雷達(dá)的水庫壩前淤積土沉積規(guī)律研究[J]. 水土保持學(xué)報(bào), 2021, 35(4): 152–158. doi: 10.13870/j.cnki.stbcxb.2021.04.021. WANG Yun, WANG Hongyu, CHANG Liucheng, et al. Study on sedimentation regulation of silted soil in the front of reservoir dam based on GPR[J]. Journal of Soil and Water Conservation, 2021, 35(4): 152–158. doi: 10.13870/j.cnki.stbcxb.2021.04.021.
[5]	覃譚, 趙永輝, 林國聰, 等. 探地雷達(dá)在上林湖越窯遺址水下考古中的應(yīng)用[J]. 物探與化探, 2018, 42(3): 624–630. doi: 10.11720/wtyht.2018.1290. QIN Tan, ZHAO Yonghui, LIN Guocong, et al. The application of GPR to underwater archaeological investigation of Shanglinhu Yue kiln relics[J]. Geophysical and Geochemical Exploration, 2018, 42(3): 624–630. doi: 10.11720/wtyht.2018.1290.
[6]	胡進(jìn)峰, 周正歐. 淺地層探地雷達(dá)目標(biāo)探測和定位新方法[J]. 儀器儀表學(xué)報(bào), 2006, 27(4): 371–375. doi: 10.3321/j.issn:0254-3087.2006.04.010. HU Jinfeng and ZHOU Zhengou. Target detection and orientation in sub-surface penetrating radar data[J]. Chinese Journal of Scientific Instrument, 2006, 27(4): 371–375. doi: 10.3321/j.issn:0254-3087.2006.04.010.
[7]	LIU Yayu, WANG Meiqing, and CAI Qiurong. The target detection for GPR images based on curve fitting[C]. The 2010 3rd International Congress on Image and Signal Processing, Yantai, China, 2010: 2876–2879. doi: 10.1109/CISP.2010.5646818.
[8]	LI Wentao, CUI Xihong, GUO Li, et al. Tree root automatic recognition in ground penetrating radar profiles based on randomized hough transform[J]. Remote Sensing, 2016, 8(5): 430. doi: 10.3390/rs8050430.
[9]	MAAS C and SCHMALZL J. Using pattern recognition to automatically localize reflection hyperbolas in data from ground penetrating radar[J]. Computers & Geosciences, 2013, 58: 116–125. doi: 10.1016/j.cageo.2013.04.012.
[10]	DOU Qingxu, WEI Lijun, MAGEE D R, et al. Real-time hyperbola recognition and fitting in GPR data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(1): 51–62. doi: 10.1109/tgrs.2016.2592679.
[11]	PATSIA O, GIANNOPOULOS A, and GIANNAKIS I. Background removal, velocity estimation, and reverse-time migration: A complete GPR processing pipeline based on machine learning[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 2003311. doi: 10.1109/tgrs.2023.3300276.
[12]	ZHANG Xiaowei, XUE Fangxiu, WANG Zepeng, et al. A novel method of hyperbola recognition in ground penetrating radar (GPR) B-scan image for tree roots detection[J]. Forests, 2021, 12(8): 1019. doi: 10.3390/f12081019.
[13]	LIU Bin, ZHANG Jiaqi, LEI Ming, et al. Simultaneous tunnel defects and lining thickness identification based on multi-tasks deep neural network from ground penetrating radar images[J]. Automation in Construction, 2023, 145: 104633. doi: 10.1016/j.autcon.2022.104633.
[14]	XIANG Deliang, PAN Xiaoyu, DING Huaiyue, et al. Two-stage registration of SAR images with large distortion based on superpixel segmentation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5211115. doi: 10.1109/tgrs.2024.3392971.
[15]	倪志康, 葉盛波, 史城, 等. 一種深度學(xué)習(xí)輔助的探地雷達(dá)定位方法[J]. 電子與信息學(xué)報(bào), 2022, 44(4): 1265–1273. doi: 10.11999/JEIT211072. NI Zhikang, YE Shengbo, SHI Cheng, et al. A deep learning assisted ground penetrating radar localization method[J]. Journal of Electronics & Information Technology, 2022, 44(4): 1265–1273. doi: 10.11999/JEIT211072.
[16]	PHAM M T and LEFèVRE S. Buried object detection from B-scan ground penetrating radar data using faster-RCNN[C]. IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, Valencia, Spain, 2018: 6804–6807. doi: 10.1109/IGARSS.2018.8517683.
[17]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137–1149. doi: 10.1109/tpami.2016.2577031.
[18]	CUI Fan, NING Muwei, SHEN Jiawei, et al. Automatic recognition and tracking of highway layer-interface using faster R-CNN[J]. Journal of Applied Geophysics, 2022, 196: 104477. doi: 10.1016/j.jappgeo.2021.104477.
[19]	來鵬飛, 李偉, 高堯, 等. 基于改進(jìn)Cascade R-CNN的探地雷達(dá)管線目標(biāo)檢測[J]. 計(jì)算機(jī)系統(tǒng)應(yīng)用, 2023, 32(2): 102–110. doi: 10.15888/j.cnki.csa.008945. LAI Pengfei, LI Wei, GAO Yao, et al. GPR pipeline target detection based on improved Cascade R-CNN[J]. Computer Systems & Applications, 2023, 32(2): 102–110. doi: 10.15888/j.cnki.csa.008945.
[20]	ZHAO Shuai, SHADABFAR M, ZHANG Dongming, et al. Deep learning-based classification and instance segmentation of leakage-area and scaling images of shield tunnel linings[J]. Structural Control and Health Monitoring, 2021, 28(6): e2732. doi: 10.1002/stc.2732.
[21]	HOU Feifei, LEI Wentai, LI Shuai, et al. Deep learning-based subsurface target detection from GPR scans[J]. IEEE Sensors Journal, 2021, 21(6): 8161–8171. doi: 10.1109/jsen.2021.3050262.
[22]	HUANG Jian, YANG Xi, ZHOU Feng, et al. A deep learning framework based on improved self-supervised learning for ground-penetrating radar tunnel lining inspection[J]. Computer-Aided Civil and Infrastructure Engineering, 2024, 39(6): 814–833. doi: 10.1111/mice.13042.
[23]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 779–788. doi: 10.1109/CVPR.2016.91.
[24]	LIU Hai, LIN Chunxu, CUI Jie, et al. Detection and localization of rebar in concrete by deep learning using ground penetrating radar[J]. Automation in Construction, 2020, 118: 103279. doi: 10.1016/j.autcon.2020.103279.
[25]	孫學(xué)超, 張其道, 尹達(dá), 等. 基于YOLOv5s的探地雷達(dá)圖像目標(biāo)檢測研究[J]. 交通世界, 2023(7): 3–6. doi: 10.16248/j.cnki.11-3723/u.2023.07.036. SUN Xuechao, ZHANG Qidao, YIN Da, et al. Research on ground penetrating radar image target detection based on YOLOv5s[J]. World of Transportation, 2023(7): 3–6. doi: 10.16248/j.cnki.11-3723/u.2023.07.036.
[26]	WANG Zhen, LAN Tian, QU Xiaodong, et al. Improved SSD framework for automatic subsurface object indentification for gpr data processing[C]. The 2021 CIE International Conference on Radar (Radar), Haikou, China, 2021: 2078–2081. doi: 10.1109/Radar53847.2021.10028347.
[27]	覃紫馨, 姜彥南, 徐立, 等. 基于YOLO算法的探地雷達(dá)道路圖像異常自動(dòng)檢測[J]. 科學(xué)技術(shù)與工程, 2023, 23(27): 11505–11512. doi: 10.3969/j.issn.1671-1815.2023.27.004. QIN Zixin, JIANG Yannan, XU Li, et al. Automatic detection of anomalies in GPR images based on YOLO algorithm[J]. Science Technology and Engineering, 2023, 23(27): 11505–11512. doi: 10.3969/j.issn.1671-1815.2023.27.004.
[28]	QIU Zhi, ZHAO Zuoxi, CHEN Shaoji, et al. Application of an improved YOLOv5 algorithm in real-time detection of foreign objects by ground penetrating radar[J]. Remote Sensing, 2022, 14(8): 1895. doi: 10.3390/rs14081895.
[29]	HU Haobang, FANG Hongyuan, WANG Niannian, et al. Defects identification and location of underground space for ground penetrating radar based on deep learning[J]. Tunnelling and Underground Space Technology, 2023, 140: 105278. doi: 10.1016/j.tust.2023.105278.
[30]	胡榮明, 李鑫, 競霞, 等. YOLOv7在探地雷達(dá)B-Scan圖像解譯中的應(yīng)用[J]. 測繪通報(bào), 2023(8): 29–33. doi: 10.13474/j.cnki.11-2246.2023.0227. HU Rongming, LI Xin, JING Xia, et al. Application of YOLOv7 in GPR B-Scan image interpretation[J]. Bulletin of Surveying and Mapping, 2023(8): 29–33. doi: 10.13474/j.cnki.11-2246.2023.0227.
[31]	WANG Niannian, ZHANG Zexi, HU Haobang, et al. Underground defects detection based on GPR by fusing simple linear iterative clustering phash (SLIC-Phash) and convolutional block attention module (CBAM)-YOLOv8[J]. IEEE Access, 2024, 12: 25888–25905. doi: 10.1109/access.2024.3365959.
[32]	LI Shupeng, CUI Xihong, GUO Li, et al. Enhanced automatic root recognition and localization in GPR images through a YOLOv4-based deep learning approach[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5114314. doi: 10.1109/tgrs.2022.3181202.
[33]	方濤濤, 王池社, 王潔, 等. 基于YOLO v8n的探地雷達(dá)圖像管線定位方法[J]. 國外電子測量技術(shù), 2023, 42(11): 170–177. doi: 10.19652/j.cnki.femt.2305179. FANG Taotao, WANG Chishe, WANG Jie, et al. Ground penetrating radar image pipeline location based on YOLO v8n[J]. Foreign Electronic Measurement Technology, 2023, 42(11): 170–177. doi: 10.19652/j.cnki.femt.2305179.
[34]	蘭天, 趙毅, 陳宏暢, 等. 基于擬合誤差消除的探地雷達(dá)圖像魯棒雙曲線識(shí)別模型[J]. 信號(hào)處理, 2023, 39(9): 1699–1710. doi: 10.16798/j.issn.1003-0530.2023.09.014. LAN Tian, ZHAO Yi, CHEN Hongchang, et al. A robust hyperbola recognition model with fitting-errors-based eliminating in GPR B-scan image[J]. Journal of Signal Processing, 2023, 39(9): 1699–1710. doi: 10.16798/j.issn.1003-0530.2023.09.014.
[35]	WANG C Y, BOCHKOVSKIY A, and LIAO H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, Canada, 2023: 7464–7475. doi: 10.1109/CVPR52729.2023.00721.
[36]	ZHOU Chuande, LU Zhenyu, LV Zhongliang, et al. Metal surface defect detection based on improved YOLOv5[J]. Scientific Reports, 2023, 13(1): 20803. doi: 10.1038/s41598-023-47716-2.
[37]	竇智, 高浩然, 劉國奇, 等. 輕量化YOLOv8的小樣本鋼板缺陷檢測算法[J]. 計(jì)算機(jī)工程與應(yīng)用, 2024, 60(9): 90–100. doi: 10.3778/j.issn.1002-8331.2311-0070. DOU Zhi, GAO Haoran, LIU Guoqi, et al. Small sample steel plate defect detection algorithm of lightweight YOLOv8[J]. Computer Engineering and Applications, 2024, 60(9): 90–100. doi: 10.3778/j.issn.1002-8331.2311-0070.
[38]	XIONG Hongqiang, LI Jing, LI Zhilian, et al. GPR-GAN: A ground-penetrating radar data generative adversarial network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5200114. doi: 10.1109/TGRS.2023.3337172.
[39]	RUSSELL B C, TORRALBA A, MURPHY K P, et al. LabelMe: A database and web-based tool for image annotation[J]. International Journal of Computer Vision, 2008, 77(1-3): 157–173. doi: 10.1007/s11263-007-0090-8.
[40]	CAI Zhaowei and VASCONCELOS N. Cascade R-CNN: High quality object detection and instance segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(5): 1483–1498. doi: 10.1109/tpami.2019.2956516.
[41]	LYU Chengqi, ZHANG Wenwei, HUANG Haian, et al. RTMDet: An empirical study of designing real-time object detectors[J]. arXiv preprint arXiv: 2212.07784, 2022. doi: 10.48550/arXiv.2212.07784.
[42]	BOUSMAHA R, LAOUEDJ S, AGGOUNE L, et al. YOLOv7-face: A real-time face detector[C]. The 2023 International Conference on Networking and Advanced Systems (ICNAS), Algiers, Algeria, 2023: 1–6. doi: 10.1109/ICNAS59892.2023.10330470.