随着深度学习的发展,利用深度学习相关知识实现水面物体检测的信息化以及智能化具有重要的意义。受到水平面复杂环境的影响,水面物体的实例分割依然是如今的难点之一。以水面物体实例分割为研究对象,提出了一种改进的Mask R-CNN网络模型。在数据集的构建阶段,使用labelme标注工具对图像进行标注并且通过添加高斯噪声、随机亮度等方法实现图像的增强。对于Mask R-CNN网络模型的改进部分,采用ResNeXt50作为骨干网络,并且在FPN网络中引入注意力机制SENet模块。实验结果表明,在IOU=0.5的判别方式中,改进后的网络模型相较于原始的Mask R-CNN网络模型,检测以及分割的平均精度分别提高到89.3%、88.5%。因此,改进后的Mask R-CNN网络模型相较于原始的Mask R-CNN网络模型更适合于水面物体的实例分割。
Abstract
With the development of deep learning,it has significance to use the knowledge that is related to deep learning to realize the informatization and intellectualization for water surface object detection.Affected by the complex environment,the instance segmentation for water surface objects is still one of the difficulties today.An improved Mask R-CNN model is proposed based on the instance segmentation for water surface objects.In the construction phase of the dataset,the images are labeled using labelme.Image enhancement is achieved by adding Gaussian noise,random brightness,etc.In the improvement of Mask R-CNN model,ResNeXt50 is used as the backbone network,and adding attention mechanism in FPN Network.In the IOU=0.5 discrimination mode,compared with the original Mask R-CNN model,the average precision of detection and segmentation is improved to 89.3%,88.5% respectively.Therefore,the improved Mask R-CNN network model is more suitable than the original Mask R-CNN network model for instance segmentation of water objects.
关键词
深度学习 /
实例分割 /
Mask R-CNN /
SENet
{{custom_keyword}} /
Key words
deep learning /
instance segmentation /
Mask R-CNN /
SENet
{{custom_keyword}} /
中图分类号：
TP389.1
{{custom_clc.code}}
({{custom_clc.text}})
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] REN S,HE K,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.
[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,Las Vegas:USA,2016.
[3] LIU W,ANGUELOV D,ERHAN D,et al.Single shot multibox detector [C]//IEEE.Conference on Computer Vision,Amsterdam:Netherlands,2016.
[4] LONG J,SHELHAMER E,DARRELL T.Fully convolutional networks for semantic segmentation[C]//IEEE.Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,Boston:USA,2015.
[5] 袁明新,张丽民,朱友帅,等.基于深度学习方法的海上舰船目标检测[J].舰船科学技术,2019,41(1):111-115.
[6] 王贵槐,谢朔,初秀民,等.基于深度学习的水面无人船前方船只图像识别方法[J].船舶工程,2018,40(4):19-22.
[7] 罗逸豪,孙创,邵成,等.基于深度学习的水面无人艇目标检测算法综述[J].数字海洋与水下攻防,2022,5(6):524-538.
[8] 许德刚,王露,李凡.深度学习的典型目标检测算法研究综述[J].计算机工程与应用,2021,57(8):10-25.
[9] XIE S,GIRSHICK R,DOLLáR P,et al.Aggregated residual transformations for deep neural networks[C]//IEEE.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Honolulu:USA,2017.
[10] 谭鹤群,李玉祥,朱明,等.通过图像增强与改进Faster-RCNN网络的重叠鱼群尾数检测[J].农业工程学报,2022,38(13):167-176.
[11] LIN T Y,DOLLáR P,GIRSHICK R,et al.Feature pyramid networks for object detection[C]//IEEE.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Honolulu:USA,2017.
[12] HU J,SHEN L,ALBANIE S,et al.Squeeze-and-excitation networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2019,42(8):2011-2023.
[13] 罗海保.基于改进Faster R-CNN的多尺度小目标检测算法研究[D].北京:北京交通大学,2019.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(41706101)
{{custom_fund}}
PDF(13752 KB)
