当前位置>主页 > 期刊在线 > 信息化应用 >

信息化应用22年11期

基于原型网络的泥石流沟谷图像预测
王旭¹,王保云¹,³,韩俊 ² ,徐繁树 ²
(1. 云南师范大学 数学学院,云南 昆明 650500;2. 云南师范大学 信息学院,云南 昆明 650500;3. 云南省高校复杂系统建模及 应用重点实验室,云南 昆明 650500)

摘  要:泥石流灾害发生迅速、破坏力极大,给人类生命财产安全带来了严重的威胁,云南省西北部地区极易发生泥石流灾害。针对泥石流灾害预测问题,文中以云南怒江流域为研究区域,以历史泥石流灾害数据为基础,提取该流域沟谷数字高程模型图,发生泥石流的沟谷图像记为正样本,未发生过泥石流的沟谷图像记为负样本。采用原型网络作为小样本学习框架,Conv4和 ResNet12 分别作为特征提取网络对沟谷图像进行训练、测试,实现了六分类预测。经实验结果对比,2-way 5-shot 条件下、ResNet12 作为特征提取网络时表现最佳,预测准确率达到 75.36%。


关键词:小样本学习;原型网络;泥石流;数字高程模型



DOI:10.19850/j.cnki.2096-4706.2022.011.033


基金项目:国家自然科学基金(61966040)


中图分类号:TP391.4;P407.8                          文献标识码:A                                 文章编号:2096-4706(2022)11-0130-03


Prediction of Debris Flow Valley Images Based on Prototypical Network

WANG Xu1, WANG Baoyun1,3, HAN Jun2, XU Fanshu2

(1.School of Mathematics, Yunnan Normal University, Kunming 650500, China; 2.School of Information, Yunnan Normal University, Kunming 650500, China; 3.Key Laboratory of Complex System Modeling and Application for Universities in Yunnan, Kunming 650500, China)

Abstract: Debris flow disasters occur rapidly and are extremely destructive. It poses a serious threat to the safety of human life and property. The northwestern region of Yunnan Province is highly prone to debris flow disasters. Aiming at the problem of debris flow disaster prediction, this paper takes the Nujiang River Basin in Yunnan as the research area, based on the historical debris flow disaster data, extracts the digital elevation model images of the valleys in the basin. The valley images which occurs debris flow are recorded as positive samples, and valley images which no debris flow are recorded as negative samples. The prototype network is used as a small sample learning framework, and Conv4 and ResNet12 are used as feature extraction networks to train and test valley images respectively, and achieve six-class prediction. Compared with the experimental results, under the condition of 2-way 5-shot, ResNet12 performs the best as the feature extraction network, and the prediction accuracy rate reaches 75.36%.

Keywords: small sample learning; prototypical network; debris flow; digital elevation model


参考文献:

[1] 孙显辰,王保云,刘坤香,等 . 云南省泥石流灾害影响因子分析 [J]. 人民长江,2020,51(11):121-127.

[2] 张蓉 . 基于支持向量机的泥石流危险性预测 [J]. 信息与电脑(理论版),2012(24):29-31.

[3] 徐黎明,王清,陈剑平,等 . 基于 BP 神经网络的泥石流平均流速预测 [J]. 吉林大学学报(地球科学版),2013,43(1):186-191.

[4] 孔艳 . 滇西高山峡谷区泥石流危险性预测及模拟——以怒江傈僳族自治州为例 [D]. 昆明:云南师范大学,2019.

[5] HE K M,ZHANG X Y,REN S Q,et al. Deep residual learning for image recognition [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Las Vegas:IEEE,2016:770-778.

[6] KRIZHEVSKY A,SUTSKEVER I,HINTON G E. ImageNet classification with deep convolutional neural networks [J].Communications of the ACM,2017,60(6):84-90.

[7] 孔艳,王保云,王乃强,等 . 滇西高山峡谷区泥石流危险性评价——以怒江傈僳族自治州为例 [J]. 云南师范大学学报(自然科学版),2019,39(3):63-70.

[8] SNELL J,SWERSKY K,ZEMEL R S. Prototypical Networks for Few-shot Learning [J/OL].arXiv:1703.05175 [cs.LG].[2022-03-12].https://arxiv.org/abs/1703.05175v1.

[9] 肖伟,冯全,张建华,等 . 基于小样本学习的植物病害识别研究 [J]. 中国农机化学报,2021,42(11):138-143.

[10] 黄彦乾,迟冬祥,徐玲玲 . 面向小样本学习的嵌入学习方法研究综述 [J]. 计算机工程与应用,2022,58(3):34-49.

[11] 李新叶,龙慎鹏,朱婧 . 基于深度神经网络的少样本学习综述 [J]. 计算机应用研究,2020,37(8):2241-2247.


作者简介:王旭(1997—),女,汉族,河北沙河人,硕士研究生在读,研究方向:图像处理和深度学习;韩俊(1990—),男,汉族,云南寻甸县人,硕士研究生在读,研究方向:图像处理和深度学习;徐繁树(1997—),男,汉族,安徽宿州人,硕士研究生在读,研究方向:图像处理和深度学习;通讯作者:王保云(1977—),男,汉族,云南华宁县人,副教授,博士,研究方向:机器学习及图像处理。