摘 要：在图像分类的实际应用场景中，受制于客观条件所以很难获取大规模的带标签数据集，针对缺少数据的场景，少样本学习得以广泛应用。然而现有少样本学习方法在图像处理时忽略了具有类别特点的局部细节对于分类的帮助，针对这一缺陷，对基于自适应局部细节增强（ALDR）注意力的少样本学习模型进行研究。实验证明，在 ALDR 注意力中通过对已学习数据提取的知识进行划分，利用不同种类的已学知识指导提取并增强新样例中具有类别特点的局部细节信息，在提升分类准确度上效果显著。
关键词：（国网江苏省电力有限公司无锡供电分公司，江苏 无锡 214000）
DOI:10.19850/j.cnki.2096-4706.2022.22.020
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2022）22-0081-05
A Few Sample Learning Model Based on ALDR Attention
YAN Minghao, QIANG Mengye, LU Qinxin
(Wuxi Power Supply Branch of State Grid Jiangsu Electric Power Co., Ltd., Wuxi 214000, China)
Abstract: In the actual application scenarios of image classification, it is difficult to obtain large-scale labeled datasets due to the objective conditions. For the scenes lacking data, small sample learning has been widely used. However, the existing few sample learning methods ignore the help of local details with category characteristics for classification in image processing. Aiming at this defect, the small sample learning model based on adaptive local detail enhancement (ALDR) attention is studied. The experiment proves that in ALDR attention, by dividing the knowledge extracted from the learned data and using different kinds of learned knowledge to guide the extraction and enhancement of local details with category characteristics in the new samples, the effect is significant in improving the classification accuracy.
Keywords: image classification; deep learning; ALDR attention; few sample learning
参考文献：
[1] SNELL J，SWERSKY K，ZEMEL R. Prototypical networks for few-shot learning [J]. Ad-vances in Neural Information Processing Systems，2017：4077-4087.
[2] VINYALS O，BLUNDELL C，LILLICRAP T. Matching networks for one shot learning [J].Proceedings of Conference and Workshop on Neural Information Processing Systems，2016：3630-3638.
[3] SUNG F，YANG Y，ZHANG L. Learning to compare：Relation network for few-shot learn-ing [J].Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni-tion，2018：1199-1208.
[4] FINN C，ABBEEL P，LEVINE S. Model-agnostic metalearning for fast adaptation of deep networks [J].Proceedings of the International Conference on Machine Learning，2017：1126-1135.
[5] LEE K，MAJI S，RAVICHANDRAN A，et al. Metalearning with differentiable convex op-timization [J].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2019：10657-10665.
[6] SIMON C，KONIUSZ K，NOCK R，et al. Adaptive subspaces for few-shot learning [J].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2020：4136–4145.
[7] HOU R，CHANG H，MA B，et al. Cross attention network for few-shot classification [J].Advances in Neural Information Processing Systems，2019.
[8] VASWANI A，SHAZEER N，PARMAR N. Attention is all you need [J]. Advances in Neural Information Processing Systems， 2017：5998-6008.
[9] HE K，ZHANG X，REN S，et al. Deep residual learning for image recognition [J].Pro-ceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2016：770-778.
[10] FEI-FEI L，FERGUS R，PERONA P. One-shot learning of object categories [J].IEEE transactions on pattern analysis and machine intelligence，2006，28（4）：594-611.
[11] 廖建平 . 基于中心聚类的深度学习遥感图像场景分类方法 [J]. 现代信息科技，2021，5（10）：27-29+33.
[12] 叶昭晖，王薇薇，张影 . 基于深度学习的图像分类方法研究 [J]. 信息网络安全，2021（S1）：143-146.
[13] 刘振 . 基于稀疏表示的图像分类若干新方法研究 [D]. 无锡：江南大学，2021.
[14] 常东良 . 基于深度学习的小样本图像分类方法研究 [D].兰州理工大学，2019.
[15] 代磊超，冯林，尚兴林，等 . 基于深度网络的快速少样本学习算法 [J]. 模式识别与人工智能，2021，34（10）：941-956.
[16] 王学良 . 基于自适应学习的少样本图像分类问题研究[D]. 合肥：中国科学技术大学，2021.
作者简介：晏明昊（1994.12—），男，汉族，江苏常熟人，助理工程师，硕士研究生，研究方向：电力信息；强梦烨（1995 －），女，汉族，江苏无锡人，助理级工程师，硕士研究生，研究方向：电力通信、信号处理。