Road defect detection algorithm based on improved YOLOv7-tiny

doi:10.11996/JG.j.2095-302X.2024050987

Abstract

Abstract:

To address the current issues in road damage detection methods, such as large parameter sizes, poor performance in detecting small targets, and high rates of false positives and missed detections, an improved YOLOv7-tiny-based road defect detection algorithm was proposed. The ELAN-SimAM-D structure was designed by introducing depthwise separable convolution (DSC) and a parameter-free attention mechanism, which could reduce computational and parameter sizes to achieve a lightweight model while enhancing the model’s feature extraction and fusion capabilities. The SPPAda structure, which incorporated adaptive exponential pooling and adaptive fusion, was introduced as a spatial pyramid pooling structure to enhance the retention of road defect information and improve detection accuracy. A new P2 small target network layer was added to strengthen the detection capability for smaller target defects, improving detection accuracy. A new loss function, NWD-EIOU, was designed to replace the original CIOU loss function, improving the localization accuracy for small targets. Experimental results showed that compared to the original YOLOv7-tiny algorithm, the improved YOLOv7-tiny algorithm achieved an mAP@0.5 of 83.14% on a self-built experimental dataset, an increase of 3.50%, with a 4.96% improvement in recall rate, and a 33.84% reduction in the model’s parameter size, meeting the requirements for road defect detection.

Key words: YOLOv7-tiny, road defect detection, adaptive exponential pooling, SimAM attention mechanism, SPPAda structure, P2 small target network layer

CLC Number:

TP391

XIE Guobo, LIN Songze, LIN Zhiyi, WU Chenfeng, LIANG Lihui. Road defect detection algorithm based on improved YOLOv7-tiny[J]. Journal of Graphics, 2024, 45(5): 987-997.

Figures/Tables 14

Fig. 1 YOLOv7-tiny algorithm network structure

Fig. 2 The structure diagram of standard convolution

Fig. 3 The structure diagram of depthwise separable convolution

Fig. 4 SimAM attention mechanism

Fig. 5 ELAM-SimAM-D structure ((a) ELAN structure; (b) ELAN-SIMAM-D structure)

Fig. 6 Comparison of different pooling methods ((a) Average pooling; (b) Max pooling; (c) Exponential maximum pooling; (d) Exponential dice-s?rensen coefficient weighting pooling; (e) Adaptive exponential pooling)

Fig. 7 AdaFusion structure diagram

Fig. 8 Comparison of structural diagrams ((a) SPPCSPC structure; (b) SPPAda structure)

Fig. 9 Improved YOLOv7-tiny algorithm network structure

Fig. 10 Sliding window cropping algorithm

Table 1 Comparison of ablation experiment results

组别	模型	P/%	R%	mAP@0.5/%	Params/M	FPS
1	YOLOV7-tiny	73.04	78.74	79.64	11.70	72.23
2	YOLOV7-tiny+A	73.82	79.71	81.22	7.47	64.24
3	YOLOV7-tiny+A+B	74.12	81.52	81.79	7.60	63.21
4	YOLOV7-tiny+A+B+C	75.42	82.81	82.74	7.74	60.41
5	YOLOV7-tiny+A+B+C+D	75.62	83.70	83.14	7.74	60.41

Table 2 Comparison of experimental results

方法	R/%	Params/M	mAP@0.5/%
SSD	56.40	93.70	69.50
YOLOV5	72.26	27.20	78.79
YOLOV7	74.90	142.10	72.58
YOLOV8	79.72	42.60	80.14
文献[10]	75.62	244.12	74.68
文献[11]	79.24	46.40	77.64
YOLOV7-tiny	78.74	11.70	79.64
本文算法	83.70	7.74	83.14

Fig. 11 Comparison of loss functions before and after algorithm improvement

Fig. 12 The comparison chart of road anomaly detection ((a) Road surface pollution; (b) Transverse cracks; (c) Road closure construction & Crack repair; (d) Pothole repair; (e) Crack)

References 22

[1]	DIESING G. How AI and machine vision intersect[J]. Quality, 2022, 61(2): 14.
[2]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 779-788.
[3]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]// The 14th European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[4]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// The IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2014: 580-587.
[5]	REN S Q, HE K M, 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 PMID
[6]	翟永杰, 赵晓瑜, 王璐瑶, 等. IDD-YOLOv7: 一种用于输电线路绝缘子多缺陷的轻量化检测方法[J]. 图学学报, 2024, 45(1): 90-101. DOI
	ZHAI Y J, ZHAO X Y, WANG L Y, et al. IDD-YOLOv7: a lightweight method for multiple defect detection of insulators in transmission lines[J]. Journal of Graphics, 2024, 45(1): 90-101 (in Chinese). DOI
[7]	苑朝, 赵亚冬, 张耀, 等. 基于YOLO轻量化的多模态行人检测算法[J]. 图学学报, 2024, 45(1): 35-46. DOI
	YUAN C, ZHAO Y D, ZHANG Y, et al. Lightweight multi-modal pedestrian detection algorithm based on YOLO[J]. Journal of Graphics, 2024, 45(1): 35-46 (in Chinese). DOI
[8]	孙朝云, 裴莉莉, 李伟, 等. 基于改进Faster R-CNN的路面灌封裂缝检测方法[J]. 华南理工大学学报: 自然科学版, 2020, 48(2): 84-93.
	SUN Z Y, PEI L L, LI W, et al. Pavement sealed crack detection method based on improved faster R-CNN[J]. Journal of South China University of Technology: Natural Science Edition, 2020, 48(2): 84-93 (in Chinese).
[9]	罗晖, 贾晨, 李健. 基于改进YOLOv4的公路路面病害检测算法[J]. 激光与光电子学进展, 2021, 58(14): 1410025.
	LUO H, JIA C, LI J. Road surface disease detection algorithm based on improved YOLOv4[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410025 (in Chinese).
[10]	杨振, 李林, 罗文婷, 等. 改进YOLOv5的沥青路面病害检测算法[J]. 计算机工程与设计, 2023, 44(11): 3360-3372.
	YANG Z, LI L, LUO W T, et al. Improved YOLOv5 asphalt pavement disease detection algorithm[J]. Computer Engineering and Design, 2023, 44(11): 3360-3372 (in Chinese).
[11]	朱振祥, 高国华, 段美栋, 等. 基于YOLOV7架构的轻量化路面病害检测模型[EB/OL]. (2024-04-25) [2024-05-26]. https://doi.org/10.16112/j.cnki.53-1223/n.2024.06.481.
	ZHU Z X, GAO G H, DUAN M D, et al. Lightweight model of pavement damage detection based on YOLOV7 architecture[EB/OL]. (2024-04-25) [2024-05-26]. https://doi.org/10.16112/j.cnki.53-1223/n.2024.06.481 (in Chinese).
[12]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 7464-7475.
[13]	CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]// The IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 1800-1807.
[14]	YANG L X, ZHANG R Y, LI L D, et al. SimAM: a simple, parameter-free attention module for convolutional neural networks[EB/OL]. [2024-03-20]. https://dblp.uni-trier.de/db/conf/icml/icml2021.html#YangZLX21.
[15]	OUYANG D L, HE S, ZHANG G Z, et al. Efficient multi-scale attention module with cross-spatial learning[C]// ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing. New York: IEEE Press, 2023: 1-5.
[16]	LIU Y C, SHAO Z R, HOFFMANN N. Global attention mechanism: retain information to enhance channel-spatial interactions[EB/OL]. [2024-03-20]. https://arxiv.org/abs/2112.05561.
[17]	STERGIOU A, POPPE R. AdaPool: exponential adaptive pooling for information-retaining downsampling[J]. IEEE Transactions on Image Processing, 2023, 32: 251-266.
[18]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. DOI PMID
[19]	REZATOFIGHI H, TSOI N, GWAK J Y, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]// The IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 658-666.
[20]	YU J H, JIANG Y N, WANG Z Y, et al. UnitBox: an advanced object detection network[C]// The 24th ACM International Conference on Multimedia. New York: ACM, 2016: 516-520.
[21]	ZHANG Y F, REN W Q, ZHANG Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157.
[22]	WANG J W, XU C, YANG W, et al. A normalized Gaussian Wasserstein distance for tiny object detection[EB/OL]. [2024-03-20]. https://arxiv.org/abs/2110.13389.