基于改进PIDNet的水位线检测算法

doi:10.19678/j.issn.1000-3428.0068301

摘要/Abstract

摘要：

PIDNet是三分支网络构成的语义分割模型, 在众多竞赛数据集中均保持优秀的分割精度。然而, 积分分支中进行多次下采样和金字塔池化模块中多分支特征融合冗余的缺点限制了算法精度的提高。在水位线检测任务中, 现有算法的缺点会导致局部细节信息丢失, 使得水体边缘精细化检测的能力有所下降。为了缓解这个问题, 提出一种基于改进PIDNet的水位线检测算法。首先设计一种结合通道注意力的轻量化像素增强模块, 在积分分支下采样过程中进行像素增强, 减少局部信息丢失。然后对金字塔池化模块进行重构, 在减小池化输出特征大小的基础上减少并行分支数, 同时在特征融合时结合通道注意力进一步加强关注重要特征的能力, 提高水位线边缘的分割精度。此外, 融合多场景的河流数据集, 避免复杂场景下检测出的水位线位置发生偏移和断线。实验结果表明, 所提方法(S和M)在水位线检测任务中相对原算法(S和M)在3个性能指标上都有所提高, 以M规模为例, 像素正确率提高了1.47个百分点, 平均交并比提高了1.04个百分点, 检测延迟降低了0.9 ms。

关键词: 语义分割, 水位线检测, 金字塔池化模块, 注意力, 多场景

Abstract:

PIDNet is a semantic segmentation model composed of three branch networks, which maintains excellent segmentation accuracy in many competitive datasets. However, the shortcomings of multiple downsampling in the integral branch and the redundancy of multi-branch feature fusion in the pyramid pooling module limit the improvement in the accuracy of the algorithm. Existing algorithms for water level line detection suffer from shortcomings that result in the loss of local detailed information, thereby reducing their ability to detect water edges. To alleviate this problem, a water level line detection algorithm based on improved PIDNet is proposed. First, a Lightweight Pixel Enhancement Module(LPEM) combined with channel attention is designed to perform pixel enhancement to reduce local information loss during integral branch downsampling. The pyramid pooling module is then reconfigured to reduce the number of parallel branches by reducing the pooling output feature size. Combining channel attention during feature fusion further enhances the ability to focus feature attention and improves the water level line edge segmentation accuracy. In addition, this study combines a multi-scene river dataset to overcome situations in which the detected water level line position will shift or even break when the scene is complicated. The experimental results show that the method(S and M) in this study improves three performance metrics relative to the original algorithm(S and M) in the water level line detection task. Considering method(M) in this study as an example, the Pixel Accuracy(PA) is improved by 1.47 percentage points, the Mean Intersection over Union(mIoU) is improved by 1.04 percentage points, and the detection delay is reduced by 0.9 ms.

Key words: semantic segmentation, water level line detection, pyramid pooling module, attention, multi-scene

李仲, 冒睿瑞, 王晓龙, 王根一, 安国成. 基于改进PIDNet的水位线检测算法[J]. 计算机工程, 2024, 50(8): 102-112.

Zhong LI, Ruirui MAO, Xiaolong WANG, Genyi WANG, Guocheng AN. Water Level Line Detection Algorithm Based on Improved PIDNet[J]. Computer Engineering, 2024, 50(8): 102-112.

https://www.ecice06.com/CN/Y2024/V50/I8/102

图/表 17

图1 PIDNet网络结构

Fig.1 Network structure of the PIDNet

图2 积分分支的LPEM改进

Fig.2 LPEM improvement of integral branch

图3 LCPPM结构图

Fig.3 Structure diagram of LCPPM

图4 简单场景数据集训练的本文方法(M)在不同场景下的检测效果对比

Fig.4 Comparison of detection performance of the proposed method(M) trained by simple scene datasets in different scenes

图5 新增河流数据集场景划分

Fig.5 Scene division in new river dataset

图6 下采样过程改进前后的PIDNet(M)损失曲线

Fig.6 PIDNet(M) loss curve before and after the improvement of the downsampling process

图7 结合LPEM和LCPPM的PIDNet(M)模型改进前后的损失曲线

Fig.7 The loss curve before and after the improvement of the PIDNet(M) model combined with LPEM and LCPPM

图8 PIDNet(M)算法2次改进前后的水位线检测效果对比

Fig.8 Comparison of water level detection effect of PIDNet(M) algorithm before and after two improvements

图9 本文方法(M)与常见SOTA算法训练精度对比

Fig.9 The training accuracy comparison of the proposed method(M) and common SOTA algorithms

图10 数据增强后的水位线检测结果

Fig.10 The detection results of water level line after data enhancement

参考文献 36

1	BARREDO J I. Major flood disasters in Europe: 1950-2005. Natural Hazards, 2007, 42(1): 125- 148. doi: 10.1007/s11069-006-9065-2
2	赵夏阳. 多源信息实时融合的实景还原及其应用[D]. 南京: 河海大学, 2007.
	ZHAO X Y. Landscape simulation and applications based multi-sources information[D]. Nanjing: Hohai University, 2007. (in Chinese)
3	VITTUCCI C, GUERRIERO L, FERRAZZOLI P, et al. River water level prediction using passive microwave signatures—a case study: the Bermejo Basin. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7(9): 3903- 3914. doi: 10.1109/JSTARS.2014.2321457
4	BISWAS R K, JAYAWARDENA A W. Water level prediction by artificial neural network in a flashy transboundary river of Bangladesh. Global Nest Journal, 2014, 16(2): 432- 444. doi: 10.30955/gnj.001226
5	OUMA Y, TATEISHI R. Urban flood vulnerability and risk mapping using integrated multi-parametric AHP and GIS: methodological overview and case study assessment. Water, 2014, 6(6): 1515- 1545.
6	TSUBAKI R, FUJITA I, TSUTSUMI S. Measurement of the flood discharge of a small-sized river using an existing digital video recording system. Journal of Hydro-environment Research, 2011, 5(4): 313- 321. doi: 10.1016/j.jher.2010.12.004
7	CREUTIN J D, MUSTE M, BRADLEY A A, et al. River gauging using PIV techniques: a proof of concept experiment on the Iowa River. Journal of Hydrology, 2003, 277(3/4): 182- 194.
8	白俊卿, 韩柏迅, 张丰侠. 基于深度学习的无人机图像语义分割算法研究. 计算机工程, 2023, 49(4): 233- 239. URL
	BAI J Q, HAN B X, ZHANG F X. Deep learning-based UAV image semantic segmentation algorithm research. Computer Engineering, 2023, 49(4): 233- 239. URL
9	程淑红, 赵考鹏, 张仕军, 等. 基于U-net的水位线检测. 计量学报, 2019, 40(3): 361- 366. URL
	CHENG S H, ZHAO K P, ZHANG S J, et al. Water level detection based on U-net. Acta Metrologica Sinica, 2019, 40(3): 361- 366. URL
10	冷建伟, 沈芳婷. 基于HSV色彩模型与区域生长法的水文图像分割. 计算机工程, 2017, 43(7): 223- 228. URL
	LENG J W, SHEN F T. Hydrological image segmentation based on HSV color model and region growing algorithm. Computer Engineering, 2017, 43(7): 223- 228. URL
11	STECCANELLA L, BLOISI D D, CASTELLINI A, et al. Waterline and obstacle detection in images from low-cost autonomous boats for environmental monitoring. Robotics and Autonomous Systems, 2020, 124, 103346. doi: 10.1016/j.robot.2019.103346
12	MUHADI N A, ABDULLAH A F, BEJO S K, et al. Deep learning semantic segmentation for water level estimation using surveillance camera. Applied Sciences, 2021, 11(20): 9691. doi: 10.3390/app11209691
13	AHMAD T, BEBIS G, NICOLESCU M, et al. Horizon line detection using supervised learning and edge cues. Computer Vision and Image Understanding, 2020, 191, 102879. doi: 10.1016/j.cviu.2019.102879
14	傅启凡, 路茗, 张质懿, 等. 基于语义分割的水位监测方法研究. 激光与光电子学进展, 2022, 59(4): 0410004. URL
	FU Q F, LU M, ZHANG Z Y, et al. Water level monitoring method based on semantic segmentation. Laser & Optoelectronics Progress, 2022, 59(4): 0410004. URL
15	LO S W, WU J H, LIN F P, et al. Visual sensing for urban flood monitoring. Sensors, 2015, 15(8): 20006- 20029. doi: 10.3390/s150820006
16	XU J C, XIONG Z X, BHATTACHARYYA S P. PIDNet: a real-time semantic segmentation network inspired by PID controllers[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2206.02066.
17	XU J C, BHATTACHARYYA S P. A PID controller architecture inspired enhancement to the PSO algorithm[C]//Proceedings of the 2022 Future of Information and Communication Conference. Berlin, Germany: Springer, 2022: 587-603.
18	MA R J, LI S Y, ZHANG B, et al. Towards fast and robust real image denoising with attentive neural network and PID controller. IEEE Transactions on Multimedia, 2022, 24, 2366- 2377. doi: 10.1109/TMM.2021.3079697
19	FAN M Y, LAI S Q, HUANG J S, et al. Rethinking BiSeNet for real-time semantic segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Washington D. C., USA: IEEE Press, 2021: 9716-9725.
20	HONG Y D, PAN H H, SUN W C, et al. Deep dual-resolution networks for real-time and accurate semantic segmentation of road scenes[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2101.06085.
21	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(CVPR). Washington D. C., USA: IEEE Press, 2017: 2117-2125.
22	SHRIVASTAVA A, GUPTA A, GIRSHICK R. Training region-based object detectors with online hard example mining[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Washington D. C., USA: IEEE Press, 2016: 761-769.
23	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2005.12872.
24	CUI C, GAO T, WEI S, et al. PP-LCNet: a lightweight CPU convolutional neural network[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2109.15099.
25	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Washington D. C., USA: IEEE Press, 2016: 770-778.
26	ANDREW G H. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/1704.04861.
27	HAN Q, FAN Z J, DAI Q, et al. On the connection between local attention and dynamic depth-wise convolution[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2106.04263.
28	CUI Z T, LI K C, ZHU Y Y, et al. You only need 90K parameters to adapt light: a light weight transformer for image enhancement and exposure correction[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2205.14871.
29	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[C]//Proceedings of the 15th European Conference on Computer Vision. Berlin, Germany: Springer, 2018: 122-138.
30	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Washington D. C., USA: IEEE Press, 2018: 7132-7141.
31	ASHISH V, NOAM S, PARMAR N, et al. Attention is all you need[C]//Proceedings of the 2022 Future of Information and Communication Conference. Berlin, Germany: Springer, 2017: 5998-6008.
32	WOO S, PARK J, LEE J, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision. Berlin, Germany: Springer, 2018: 3-19.
33	LI X T, YOU A S, ZHU Z, et al. Semantic flow for fast and accurate scene parsing[C]//Proceedings of the European Conference on Computer Vision. Berlin, Germany: Springer, 2020: 775-793.
34	PENG J C, LIU Y, TANG S Y, et al. PP-LiteSeg: a superior real-time semantic segmentation model[EB/OL]. [2023-07-10]. https://arxiv.org/pdf/2204.02681.
35	YU C Q, GAO C X, WANG J B, et al. BiSeNet V2: bilateral network with guided aggregation for real-time semantic segmentation. International Journal of Computer Vision, 2021, 129(11): 3051- 3068. doi: 10.1007/s11263-021-01515-2
36	NIRKIN Y, WOLF L, HASSNER T. HyperSeg: patch-wise hypernetwork for real-time semantic segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Washington D. C., USA: IEEE Press, 2021: 4061-4070.

[1]	李俊俊, 董建刚, 李坤. 基于Kubernetes的集群节能策略研究[J]. 计算机工程, 2024, 50(9): 82-91.
[2]	杨明强, 卢健. 基于跨模态注意力的目标语音提取[J]. 计算机工程, 2024, 50(9): 121-129.
[3]	林畅, 郭伟, 任哲聪, 金海波. 基于Transformer的目标跟踪与分割统一算法[J]. 计算机工程, 2024, 50(9): 130-141.
[4]	李泽霖, 吕兆峰, 陈富强, 李克. 基于多跳信息融合的实体对齐模型[J]. 计算机工程, 2024, 50(9): 142-152.
[5]	王汝英, 马嘉骏, 董建强, 刘万龙, 张海涛, 尹凯, 赵博超. 基于MTS-BiGRU-DMHSA的工业负荷预测方法[J]. 计算机工程, 2024, 50(9): 169-178.
[6]	蔡俊民, 梁正友, 孙宇, 陈子奥. 基于可变形三维图卷积的轻量级点云分类研究[J]. 计算机工程, 2024, 50(9): 255-265.
[7]	朱凯, 李理, 张彤, 江晟, 别一鸣. 基于Transformer的多阶段运动模糊图像修复网络[J]. 计算机工程, 2024, 50(9): 276-285.
[8]	张天鹏, 韩晶, 吕学强. 基于多任务学习的超分辨率辅助小目标检测[J]. 计算机工程, 2024, 50(9): 304-312.
[9]	郭敏, 张熙涵, 李阳. 融合注意力的教师互一致性半监督医学图像分割[J]. 计算机工程, 2024, 50(9): 313-323.
[10]	曾钰琦, 刘博, 钟柏昌, 钟瑾. 智慧教育下基于改进YOLOv8的学生课堂行为检测算法[J]. 计算机工程, 2024, 50(9): 344-355.
[11]	饶日昕, 王怡文, 曾砺志, 童心恬, 赵海涛. 面向废旧电缆检测的轻量化网络模型[J]. 计算机工程, 2024, 50(8): 22-30.
[12]	李华昱, 张智康, 闫阳, 岳阳. 基于知识图谱增强的领域多模态实体识别[J]. 计算机工程, 2024, 50(8): 31-39.
[13]	王蕾, 党时鹏, 潘丰. 基于卷积神经网络的隐匿性旁路预测模型[J]. 计算机工程, 2024, 50(8): 40-49.
[14]	陈瀚, 赵春蕾, 蒋昊达, 王春东. 基于融合模型与语义网络的App用户意图识别研究[J]. 计算机工程, 2024, 50(8): 50-63.
[15]	赵婉秋, 张俊虎, 李海涛. 用于建筑物分割的平行结构特征融合网络[J]. 计算机工程, 2024, 50(8): 239-248.

选择文件类型/文献管理软件名称

选择包含的内容