基于卷积神经网络的小型建筑物检测算法

doi:10.12305/j.issn.1001-506X.2021.11.08

摘要/Abstract

摘要：

针对基于传统卷积神经网络的建筑物目标检测算法对于小型建筑物检测准确率低的问题, 提出一种基于Mask-区域卷积神经网络(Mask-region convoluional neural networks, Mask-RCNN)模型的小目标检测算法模型。该模型对Mask-RCNN模型中的特征提取网络进行了改进, 设计了一种带有注意力机制的多尺度组卷积神经网络, 有效解决了小目标有用特征较少且易被背景特征和噪声干扰的问题。航拍图像实验结果表明, 改进的检测模型使小型建筑物目标检测准确率较原始Mask-RCNN模型提升了28.9%, 达到了0.663。并且整体检测准确率达到了0.843, 有效提升了航拍建筑物检测准确性。

关键词: 建筑物检测, 小目标检测, 卷积神经网络, 特征提取, 注意力机制

Abstract:

Aiming at the low accuracy of building target detection algorithm based on traditional convolutional neural network for small buildings, a small target detection algorithm model based on Mask-region convolutional neural networks (Mask-RCNN) model is proposed. The model improves the feature extraction network in Mask-RCNN model, and designs a multi-scale group convolution neural network with attention mechanism, which effectively solves the problem that small targets have few useful features and are easy to be disturbed by background features and noise. The experimental results of aerial images show that the improved detection model improves the target detection accuracy of small buildings by 28.9% compared with the original Mask-RCNN model, reaching 0.663. And the overall detection accuracy reaches 0.843, which effectively improves the accuracy of aerial building detection.

Key words: building detection, small target detection, convolutional neural network(CNN), feature extraction, attention mechanism

中图分类号:

TP391

赵若辰, 王敬东, 林思玉, 顾东泽. 基于卷积神经网络的小型建筑物检测算法[J]. 系统工程与电子技术, 2021, 43(11): 3098-3106.

Ruochen ZHAO, Jingdong WANG, Siyu LIN, Dongze GU. Small building detection algorithm based on convolutional neural network[J]. Systems Engineering and Electronics, 2021, 43(11): 3098-3106.

图/表 16

图1

图2

图3

表1

图4

图5

图6

图7

图8

图9

表2

表3

图10

图11

表4

图12

参考文献 30

1	顾炼. 基于深度学习的遥感图像建筑物检测及其变化检测研究[D]. 杭州: 浙江工商大学, 2018.
	GU L. Detection for buildings and their changes in remote sensing images based on deep learning[D]. Hangzhou: Zhejiang Gongshang University, 2018.
2	刘文涛. 基于卷积神经网络的城市区域建筑物自动提取研究[D]. 成都: 电子科技大学, 2018.
	LIU W T. Study on automatic extraction of urban area buildings based on convolution neural network[D]. Chengdu: University of Electronic Science and Technology of China, 2018.
3	冯丽英. 基于深度学习技术的高分辨率遥感影像建设用地信息提取研究[D]. 浙江: 浙江大学, 2017.
	FENG L Y. Research on construction land information extraction from high resolution images with deep learning technology[D ]. Hangzhou: Zhejiang University, 2017.
4	LIN J B , JING W P , SONG J P . ESFNet: efficient network for building extraction from high-resolution aerial images[J]. IEEE Access, 2019, 7, 54285- 54294. doi: 10.1109/ACCESS.2019.2912822
5	CHEN J , ZHANG D , NANEHKARAN Y A , et al. Detection of rice plant diseases based on deep transfer learning[J]. Journal of the Science of Food and Agriculture, 2020, 100 (7): 3246- 3256. doi: 10.1002/jsfa.10365
6	LIU Y B , ZHANG Z X , ZHONG R F , et al. Multilevel building detection framework in remote sensing images based on convolutional neural networks[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2018, 11 (10): 3688- 3700. doi: 10.1109/JSTARS.2018.2866284
7	李大军, 何维龙, 郭丙轩, 等. 基于Mask-RCNN的建筑物目标检测算法[J]. 测绘科学, 2019, 44 (10): 172- 180.
	LI D J , HE W L , GUO B X , et al. Building target detection algorithm based on Mask-RCNN[J]. Science of Surveying and Mapping, 2019, 44 (10): 172- 180.
8	HE K M, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//Proc. of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
9	AUDEBERT N, SAUX B L, LEFÈVRE, S. Semantic segmentation of earth observation data using multimodal and multi-scale deep networks[C]//Proc. of the Asian Conference of Computer Vision, 2016: 180-196.
10	LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015: 3431-3440.
11	伍广明, 陈奇, RYOSUKES, 等. 基于U型卷积神经网络的航空影像建筑物检测[J]. 测绘学报, 2018, 47 (6): 864- 872.
	WU G M , CHEN Q , RYOSUKE S , et al. High precision building detection from aerial imagery using a U-Net like convolutional architecture[J]. Journal of Geodesy and Geoinformation Science, 2018, 47 (6): 864- 872.
12	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proc. of the Conference on Medical Image Computing and Computer-Assisted Intervention, 2015: 234-241.
13	ZUO T C, FENG J, CHEN X. HF-FCN: hierarchically fused fully convolutional network for robust building extraction[C]//Proc. of the Asian Conference of Computer Vision, 2016: 291- 302.
14	于洋, 施国武, 刘斌, 等. 基于全卷积神经网络的无人机影像建筑物提取[J]. 水利水电技术, 2020, 51 (7): 31- 38.
	YU Y , SHI G W , LIU B , et al. Fully convolutional network-based building extraction of image from unmanned aerial vehicle[J]. Water Resources and Hydropower Engineering, 2020, 51 (7): 31- 38.
15	李强. 基于全卷积神经网络的建筑物提取方法研究[D]. 成都: 西南交通大学, 2016.
	LI Q. Research on building extraction method based on full convolution neural network[D]. Chengdu: Southwest Jiaotong University, 2016.
16	BADRINARAYANAN V , KENDALL A , CIPOLLA R . SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Trans.on Pattern Analysis and Machine Intelligence, 2017, 39 (12): 2481- 2495.
17	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 636-644.
18	BISCHKE B, HELBER P, FOLZ J, et al. Multi-task learning for segmentation of building footprints with deep neural networks[EB/OL]. [2020-09-18]. https://arxiv.org/abs/1709.05932.
19	HE K M, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
20	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proc. of the IEEE Confe-rence on Computer Vision and Pattern Recognition, 2017: 936-944.
21	GAO H, LIU Z, MAATEN L V, et al. Densely connected convolutional networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2261-2269.
22	LI J N , WEI Y C , LIANG X D , et al. Attentive contexts for object detection[J]. IEEE Trans.on Multimedia, 2016, 19 (5): 944- 954.
23	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 2818-2826.
24	KRIZHEVSKY A , SUTSKEVER I , HINTON G . Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60 (6): 84- 90. doi: 10.1145/3065386
25	BAHDANAU D, CHO K, BENGIO Y. Neural machine translatio n by jointly learning to align and translate[EB/OL]. [2020-09-01]. https://arxiv.org/abs/1409.0473.
26	LI X, WANG W H, HU X L, et al. Selective kernel networks[C]// Proc. of the Conference on Computer Vision and Pattern Recognition, 2019: 5 10-519.
27	LIN T, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[EB/OL]. [2020-09-16]. https://arxiv.org/abs/1405.0312.
28	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2020-09-04]. https://arxiv.org/abs/1409.1556.
29	SZEGEDY C, LOFFE S, VANHOUCKE V, et al. Inception-v4 : inception-resNet and the impact of residual connections on learning[C]//Proc. of the 31st AAAI Conference on Artificial Intelligence, 2017: 4278-4284.
30	SAINING X, ROSS G, PIOTR D, et al. Aggregated residual transformations for deep neural networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 5987 -5995.

网络层	Resnet101	Densenet169
C1	7×7, 2k 3×3, max pool	7×7, 2k 3×3, MaxPooling
C2	$\left[ {\begin{array}{*{20}{c}}{1 \times 1, 2k}\\{3 \times 3, 2k}\\{1 \times 1, 8k}\end{array}} \right] \times 3 $	$\begin{array}{c}\left[ {\begin{array}{*{20}{c}}{1 \times 1, k}\\{3 \times 3, k}\\{1 \times 1, k}\end{array}} \right] \times 6\\1 \times 1, 0.5k\\2 \times 2\;\;{\rm{AveragePooling}}\end{array}$
C3	$\left[ {\begin{array}{*{20}{c}}{1 \times 1, 4k}\\{3 \times 3, 4k}\\{1 \times 1, 16k}\end{array}} \right] \times 4 $	$\begin{array}{c}\left[ {\begin{array}{*{20}{c}}{1 \times 1, k}\\{3 \times 3, k}\\{1 \times 1, k}\end{array}} \right] \times 12\\1 \times 1, 0.5k\\2 \times 2\;\;{\rm{AveragePooling}}\end{array} $
C4	$\left[ {\begin{array}{*{20}{c}}{1 \times 1, 8k}\\{3 \times 3, 8k}\\{1 \times 1, 32k}\end{array}} \right] \times 23 $	$ \begin{array}{c}\left[ {\begin{array}{*{20}{c}}{1 \times 1, k}\\{3 \times 3, k}\\{1 \times 1, k}\end{array}} \right] \times 32\\1 \times 1, 0.5k\\2 \times 2\;\;{\rm{AveragePooling}}\end{array}$
C5	$\left[ {\begin{array}{*{20}{c}}{1 \times 1, 16k}\\{3 \times 3, 16k}\\{1 \times 1, 64k}\end{array}} \right] \times 3 $	$ \begin{array}{c}\left[ {\begin{array}{*{20}{c}}{1 \times 1, k}\\{3 \times 3, k}\\{1 \times 1, k}\end{array}} \right] \times 32\\1 \times 1, 0.5k\\2 \times 2\;\;{\rm{AveragePooling}}\end{array}$

建筑物数量	训练集	测试集
小型建筑物/个	2 652	156
中型建筑物/个	3 736	186
大型建筑物/个	2 600	137
总数/个	8 988	479

方法	AP	AP_s	AP_m	AP_l
Resnet 101	0.739	0.514	0.849	0.847
Densenet 169	0.781	0.563	0.882	0.892
Densenet 169 +多尺度组卷积	0.815	0.652	0.891	0.899
Densenet 169+多尺度组卷积+SKnet	0.843	0.663	0.929	0.934

网络	参数
网络	FPS	FLOPs	AP
VGG16	1.93	15.5G	0.554
Resnet101	1.86	7.85G	0.739
InceptionV4	1.98	2.85G	0.755
Resnext101	1.67	8.03G	0.780
Densenet169	1.65	3.42G	0.781
本文网络	1.64	3.47G	0.843

[1]	宋爽, 张悦, 张琳娜, 岑翼刚, 李浥东. 基于深度学习的轻量化目标检测算法[J]. 系统工程与电子技术, 2022, 44(9): 2716-2725.
[2]	韩啸, 陈世文, 陈蒙, 杨锦程. 基于互易点学习的LPI信号开集识别[J]. 系统工程与电子技术, 2022, 44(9): 2752-2759.
[3]	王彩云, 吴钇达, 王佳宁, 马璐, 赵焕玥. 基于改进的CNN和数据增强的SAR目标识别[J]. 系统工程与电子技术, 2022, 44(8): 2483-2487.
[4]	马泽煊, 李进, 路艳丽, 陈晨. 融合WaveNet和BiGRU的网络入侵检测方法[J]. 系统工程与电子技术, 2022, 44(8): 2652-2660.
[5]	李郝亮, 陈思伟, 王雪松. 海面角反射器的极化旋转域特性研究[J]. 系统工程与电子技术, 2022, 44(7): 2065-2073.
[6]	韦娟, 杨皇卫, 宁方立. 基于NMF与CNN联合优化的声学场景分类[J]. 系统工程与电子技术, 2022, 44(5): 1433-1438.
[7]	徐平亮, 崔亚奇, 熊伟, 熊振宇, 顾祥岐. 生成式中断航迹接续关联方法[J]. 系统工程与电子技术, 2022, 44(5): 1543-1552.
[8]	魏文晓, 刘洁瑜, 沈强, 李成. 基于人眼视点图的特征融合小目标检测算法[J]. 系统工程与电子技术, 2022, 44(4): 1120-1127.
[9]	陈冬, 句彦伟. 基于语义分割实现的SAR图像舰船目标检测[J]. 系统工程与电子技术, 2022, 44(4): 1195-1201.
[10]	方伟, 王玉, 闫文君, 林冲. 基于神经网络的符号化飞行动作识别[J]. 系统工程与电子技术, 2022, 44(3): 737-745.
[11]	孙晶明, 虞盛康, 孙俊. 基于深度学习的HRRP识别姿态敏感性分析[J]. 系统工程与电子技术, 2022, 44(3): 802-807.
[12]	刘恒燕, 张立民, 闫文君, 钟兆根, 凌青, 梁晓军. 基于WBP-CNN算法的LDPC译码[J]. 系统工程与电子技术, 2022, 44(3): 1030-1035.
[13]	邵凯, 朱苗苗, 王光宇. 基于生成对抗与卷积神经网络的调制识别方法[J]. 系统工程与电子技术, 2022, 44(3): 1036-1043.
[14]	吴涛, 王伦文, 朱敬成. 基于迁移学习和注意力机制的伪装图像分割[J]. 系统工程与电子技术, 2022, 44(2): 376-384.
[15]	张玺, 金正猛, 姜亚琴. 融合深度图像先验的全变差图像着色算法[J]. 系统工程与电子技术, 2022, 44(2): 385-393.