基于多尺度信息融合的图像识别改进算法

doi:10.13229/j.cnki.jdxbgxb20191182

摘要/Abstract

摘要：

针对医学图像样本中存在噪声元素多、正负样本区分度不明显的问题，设计了一种递进式的多尺度信息融合机制，并结合自然图像分类中的注意力机制，对现有卷积神经网络结构进行改进，建立了端到端的图像自动识别模型。该模型对卷积神经网络不同层的输出选择相对应融合策略，对图像的不同层次的信息进行综合利用。采用改进后的模型对肺炎疾病正负样本X光片进行识别，准确率达到95.11%，查准率达到90.75%，查全率达到90.28%。设计了多组交叉对比实验，实验结果表明:改进后的模型在时间和空间复杂度上的优越性以及本文设计的多尺度信息融合机制对图像识别任务准确率的提升性更高。

关键词: 计算机应用技术, 卷积神经网络, 肺炎疾病识别, 注意力机制, 多尺度信息融合机制

Abstract:

For the problem that there are many noise elements in the medical image samples and the distinction between positive and negative samples is not obvious, this paper designed a progressive type of multi-scale information fusion system, an end-to-end image automatic recognition model is established based on the existing convolution neural network structure with the attention mechanism in the classification of natural images. The model selected the corresponding fusion strategy for the output of different layers of the convolutional neural network and comprehensively utilized the information of different layers of the image. The improved model was used to identify the X-ray images of pneumonia patients and normal human beings, with an accuracy rate of 95.11%, an precision rate of 90.75%, and a recall rate of 90.28%, then a multi-group cross-contrast experiment was designed. Experimental results show that the improved model is superior in time and space complexity and the multi-scale information fusion mechanism designed in this paper improves the accuracy of image recognition tasks.

Key words: computer application technology, convolutional neural network, neumonia disease identification, attentional mechanism, multi-scale information fusion mechanism

中图分类号:

TP391.4

车翔玖,董有政. 基于多尺度信息融合的图像识别改进算法[J]. 吉林大学学报(工学版), 2020, 50(5): 1747-1754.

Xiang-jiu CHE,You-zheng DONG. Improved image recognition algorithm based on multi⁃scale information fusion[J]. Journal of Jilin University(Engineering and Technology Edition), 2020, 50(5): 1747-1754.

图/表 12

图1

图2

图3

图4

图5

表1

表2

表3

图6

图7

图8

图9

参考文献 14

1	Krizhevsky A, Sutskever I,Hinton G. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6)： 84-90.
2	Kawahara J, Hamarneh G. Multi-resolution-tract cnn with hybrid pretrained and skin-lesion trained layers[C]∥Medical Image Computing and Computer-Assisted Intervention Workshop on Machine Learning in Medical Imaging. Springer International Publishing, 2016: 164-171.
3	Shen W, Zhou M, Yang F, et al. Multi-scale convolutional neural networks for lung nodule classification[J]. Inf Process Med Imaging, 2015, 24: 588-599.
4	Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,USA, 2017: 2117-2125.
5	Liu W, Anguelov D, Erhan D, et al. Ssd: single shot multibox detector[C]∥European Conference on Computer Vision， Springer, Cham, 2016: 21-37.
6	Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
7	Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]∥Proceedings of the IEEE Cconference on Computer Vision and Pattern Recognition, USA, 2015: 1-9.
8	He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]∥Proceedings of the IEEE Cconference on Computer Vision and Pattern Recognition, USA, 2016: 770-778.
9	He K, Zhang X, Ren S, et al. Identity mappings in deep residual networks[C]∥European Conference on Computer Vision， Springer, Cham, 2016: 630-645.
10	Bell S, Lawrence Z C, Bala K, et al. Inside-outside net: detecting objects in context with skip pooling and recurrent neural networks[C]∥Proceedings of the IEEE Cconference on Computer Vision and Pattern Recognition, USA, 2016: 2874-2883.
11	Jaderberg M, Simonyan K, Zisserman A. Spatial transformer networks[C]∥Advances in Neural Information Processing Systems, Cambridge, MA, 2015: 2017-2025.
12	Hu J,Shen L, Sun G. Squeeze-and-excitation networks[C]∥Proceedings of the IEEE Cconference on Computer Vision and Pattern Recognition, USA, 2018: 7132-7141.
13	Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]∥European Conference on Computer Vision， Springer, Cham, 2014: 818-833.
14	Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]∥Proceedings of the IEEE Cconference on Computer Vision and Pattern Recognition, USA, 2015: 3431-3440.

相关文章 12

[1]	赵宏伟,王鹏,范丽丽,胡黄水,刘萍萍. 相似性保持实例检索方法[J]. 吉林大学学报(工学版), 2019, 49(6): 2045-2050.
[2]	周柚,杨森,李大琳,吴春国,王岩,王康平. 基于现场可编程门电路的人脸检测识别加速平台[J]. 吉林大学学报(工学版), 2019, 49(6): 2051-2057.
[3]	车翔玖,刘华罗,邵庆彬. 基于Fast RCNN改进的布匹瑕疵识别算法[J]. 吉林大学学报(工学版), 2019, 49(6): 2038-2044.
[4]	马子骥,卢浩,董艳茹. 双通道单图像超分辨率卷积神经网络[J]. 吉林大学学报(工学版), 2019, 49(6): 2089-2097.
[5]	郭继昌,吴洁,郭春乐,朱明辉. 基于残差连接卷积神经网络的图像超分辨率重构[J]. 吉林大学学报(工学版), 2019, 49(5): 1726-1734.
[6]	欧阳丹彤,肖君,叶育鑫. 基于实体对弱约束的远监督关系抽取[J]. 吉林大学学报(工学版), 2019, 49(3): 912-919.
[7]	徐岩,孙美双. 基于卷积神经网络的水下图像增强方法[J]. 吉林大学学报(工学版), 2018, 48(6): 1895-1903.
[8]	王方石, 王坚, 李兵, 王博. 基于深度属性学习的交通标志检测[J]. 吉林大学学报(工学版), 2018, 48(1): 319-329.
[9]	马淼, 李贻斌. 基于多级图像序列和卷积神经网络的人体行为识别[J]. 吉林大学学报(工学版), 2017, 47(4): 1244-1252.
[10]	李琳辉, 伦智梅, 连静, 袁鲁山, 周雅夫, 麻笑艺. 基于卷积神经网络的道路车辆检测方法[J]. 吉林大学学报(工学版), 2017, 47(2): 384-391.
[11]	胡冠宇, 乔佩利. 基于云群的高维差分进化算法及其在网络安全态势预测上的应用[J]. 吉林大学学报(工学版), 2016, 46(2): 568-577.
[12]	佟金, 王亚辉, 樊雪梅, 张书军, 陈东辉. 生鲜农产品冷链物流状态监控信息系统[J]. 吉林大学学报(工学版), 2013, 43(06): 1707-1711.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

名称	Resnet-18结构	Resnet-34结构
Conv_1	Conv 7×7 Max Pool 3×3	Conv 7×7 Max Pool 3×3
Conv_2	Building block ×2	Building block ×3
Conv_3	Building block ×2	Building block ×4
Conv_4	Building block ×2	Building block ×6
Conv_5	Building block ×2	Building block ×3
Downsample	Max Pool 2×2 BN，Relu，Conv 1×1	Max Pool 2×2 BN，Relu, Conv 1×1
Aggregation	Add	Add

网络类型	准确率/%	查准率/%	查全率/%
VGG-16	89.65	82.43	83.36
Resnet-18	92.25	85.61	84.54
Resnet-18-A	94.44	89.26	89.13
Resnet-18-B	94.48	89.03	89.44
Resnet-18-C	94.59	90.35	88.76

网络类型	准确率/%	查准率/%	查全率/%
VGG-16	89.65	82.43	83.36
Resnet-34	93.21	86.73	86.86
Resnet-34-A	94.04	87.84	88.84
Resnet-34-B	95.03	90.08	90.55
Resnet-34-C	95.11	90.75	90.28