计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (2): 458-467.DOI: 10.3778/j.issn.1673-9418.2111036
那峙雄1,+(), 樊涛2, 孙涛1, 谢祥颖3,1, 来广志1
收稿日期:
2021-11-05
修回日期:
2022-01-12
出版日期:
2022-02-01
发布日期:
2022-01-17
通讯作者:
+ E-mail: hellonazx@163.com作者简介:
那峙雄(1974—),男,北京人,硕士,工程师,主要研究方向为可再生能源与新型电力系统研究应用等。基金资助:
NA Zhixiong1,+(), FAN Tao2, SUN Tao1, XIE Xiangying3,1, LAI Guangzhi1
Received:
2021-11-05
Revised:
2022-01-12
Online:
2022-02-01
Published:
2022-01-17
About author:
NA Zhixiong, born in 1974, M.S., engineer. His research interests include renewable energy, new power system, etc.Supported by:
摘要:
针对工业生产线光伏组件隐性纹检测问题,为了降低人力成本,提高检测效率,并快速适应新型产品的隐裂检测,提出了一种多损失融合的小样本光伏组件隐裂检测算法。首先,为丰富卷积神经网络提取的语义信息,引入了Transformer的多头注意力机制,缓解各批次产品的分布差异对隐裂检测的影响,促使模型从多样化产品中关注于隐裂信息;其次,利用多损失结合约束模型训练的策略优化特征提取,在直接分类损失的基础上,利用三元组损失拉近含隐裂样本间特征距离;此外,设计了隐式分类损失以适应有无隐裂两类电池片内部也存在类型差异的特点,充分学习历史组件数据的多样性。该算法能够快速提取新型组件特征,利用少量的样本特征对新产品隐裂缺陷进行准确检测。在实际工业生产数据集上的实验结果表明,该算法对新型组件的隐裂检测的召回率相较于其他基线模型可提高10个百分点,能够有效缓解含隐裂样本数量不足的问题,极大地降低了频繁对每批新产品进行数据标记和训练的开销。
中图分类号:
那峙雄, 樊涛, 孙涛, 谢祥颖, 来广志. 多损失融合的小样本光伏组件隐裂检测算法[J]. 计算机科学与探索, 2022, 16(2): 458-467.
NA Zhixiong, FAN Tao, SUN Tao, XIE Xiangying, LAI Guangzhi. Micro-cracks Detection of Solar Cells Based on Few Shot Samples with Multi-loss[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(2): 458-467.
K-shot | 召回率/% | 精确率/% | F1-score |
---|---|---|---|
5 | 85.07 | 75.09 | 0.797 69 |
10 | 85.13 | 78.44 | 0.816 48 |
15 | 85.00 | 81.25 | 0.830 83 |
表1 K-shot下检测结果
Table 1 Detection results under K-shot
K-shot | 召回率/% | 精确率/% | F1-score |
---|---|---|---|
5 | 85.07 | 75.09 | 0.797 69 |
10 | 85.13 | 78.44 | 0.816 48 |
15 | 85.00 | 81.25 | 0.830 83 |
网络结构 | 损失策略 | 召回率/% | 精确率/% | |
---|---|---|---|---|
VGG19 | 交叉熵损失 | 65.87 | 68.46 | 0.671 40 |
+三元组损失 | 71.47 | 69.55 | 0.704 97 | |
+三元组&隐式 | 71.93 | 71.52 | 0.717 24 | |
ResNet50 | 交叉熵损失 | 65.53 | 64.38 | 0.649 50 |
+三元组损失 | 75.07 | 70.20 | 0.725 53 | |
+三元组&隐式 | 78.33 | 75.55 | 0.769 15 | |
ResNet50+ SE Block | 交叉熵损失 | 68.00 | 65.56 | 0.667 58 |
+三元组损失 | 79.13 | 71.49 | 0.751 16 | |
+三元组&隐式 | 83.27 | 72.26 | 0.773 75 | |
ResNet50+ Non-local Block | 交叉熵损失 | 66.60 | 65.76 | 0.661 77 |
+三元组损失 | 80.07 | 72.29 | 0.759 81 | |
+三元组&隐式 | 84.47 | 73.62 | 0.786 73 | |
ResNet50+ Transformer Block | 交叉熵损失 | 69.67 | 63.22 | 0.662 88 |
+三元组损失 | 76.47 | 74.25 | 0.753 44 | |
+三元组&隐式 | 85.07 | 75.09 | 0.797 69 |
表2 5-shot下消融实验结果
Table 2 Results of ablation experiment under 5-shot
网络结构 | 损失策略 | 召回率/% | 精确率/% | |
---|---|---|---|---|
VGG19 | 交叉熵损失 | 65.87 | 68.46 | 0.671 40 |
+三元组损失 | 71.47 | 69.55 | 0.704 97 | |
+三元组&隐式 | 71.93 | 71.52 | 0.717 24 | |
ResNet50 | 交叉熵损失 | 65.53 | 64.38 | 0.649 50 |
+三元组损失 | 75.07 | 70.20 | 0.725 53 | |
+三元组&隐式 | 78.33 | 75.55 | 0.769 15 | |
ResNet50+ SE Block | 交叉熵损失 | 68.00 | 65.56 | 0.667 58 |
+三元组损失 | 79.13 | 71.49 | 0.751 16 | |
+三元组&隐式 | 83.27 | 72.26 | 0.773 75 | |
ResNet50+ Non-local Block | 交叉熵损失 | 66.60 | 65.76 | 0.661 77 |
+三元组损失 | 80.07 | 72.29 | 0.759 81 | |
+三元组&隐式 | 84.47 | 73.62 | 0.786 73 | |
ResNet50+ Transformer Block | 交叉熵损失 | 69.67 | 63.22 | 0.662 88 |
+三元组损失 | 76.47 | 74.25 | 0.753 44 | |
+三元组&隐式 | 85.07 | 75.09 | 0.797 69 |
网络结构 | 参数量/106 | 计算量/ GFLOPs | 检测速度/(张/s) |
---|---|---|---|
VGG19 | 20.0 | 19.5 | 93 |
ResNet50 | 23.5 | 4.1 | 268 |
ResNet50+SE | 25.1 | 4.1 | 266 |
ResNet50+Non-local | 48.7 | 5.3 | 226 |
ResNet50+Transformer | 28.3 | 4.9 | 233 |
表3 模型属性对比结果
Table 3 Comparison results of model attributes
网络结构 | 参数量/106 | 计算量/ GFLOPs | 检测速度/(张/s) |
---|---|---|---|
VGG19 | 20.0 | 19.5 | 93 |
ResNet50 | 23.5 | 4.1 | 268 |
ResNet50+SE | 25.1 | 4.1 | 266 |
ResNet50+Non-local | 48.7 | 5.3 | 226 |
ResNet50+Transformer | 28.3 | 4.9 | 233 |
对比方法 | 召回率/% | 精确率/% | F1-score |
---|---|---|---|
ProtoNet[ | 68.73 | 70.02 | 0.693 69 |
RelationNet[ | 71.26 | 53.62 | 0.611 94 |
MAML[ | 67.07 | 69.01 | 0.680 26 |
OptNet[ | 64.86 | 60.24 | 0.624 65 |
多损失融合 | 85.07 | 75.09 | 0.797 69 |
表4 5-shot 下对比实验结果
Table 4 Comparison results under 5-shot
对比方法 | 召回率/% | 精确率/% | F1-score |
---|---|---|---|
ProtoNet[ | 68.73 | 70.02 | 0.693 69 |
RelationNet[ | 71.26 | 53.62 | 0.611 94 |
MAML[ | 67.07 | 69.01 | 0.680 26 |
OptNet[ | 64.86 | 60.24 | 0.624 65 |
多损失融合 | 85.07 | 75.09 | 0.797 69 |
[1] |
TAFTI H D, MASWOOD A I, KONSTANTINOU G, et al. A general constant power generation algorithm for photovoltaic systems[J]. IEEE Transactions on Power Electronics, 2018, 33(5):4088-4101.
DOI URL |
[2] |
FUYUKI T, KITIYANAN A. Photographic diagnosis of crystalline silicon solar cells utilizing electroluminescence[J]. Applied Physics A, 2009, 96(1):189-196.
DOI URL |
[3] | RODRÍGUEZ CASTAÑEDA C A, CHATTOPADHYAY S, OH J, et al. Field inspection of PV modules: quantitative determination of performance loss due to cell cracks using EL images[C]//Proceedings of the 2017 IEEE 44th Photovoltaic Specialist Conference, Washington, Jun 25-30, 2017. Piscat-away: IEEE, 2017: 1858-1862. |
[4] | LI Y, ZHANG Y H. Application research of computer vision technology in automation[C]//Proceedings of the 2020 Interna-tional Conference on Computer Information and Big Data Applications, Guiyang, Apr 17-19, 2020. Piscataway: IEEE, 2020: 374-377. |
[5] | 侯永宏, 吕晓冬, 陈艳芳, 等. 深度神经网络在森林步道视觉识别中的应用[J]. 计算机科学与探索, 2019, 13(2):263-274. |
HOU Y H, LV X D, CHEN Y F, et al. Application of deep neural networks in visual recognition of forest trails[J]. Journal of Frontiers of Computer Science and Technology, 2019, 13(2):263-274. | |
[6] | XIE X, LIU H, NA Z, et al. DPiT: detecting defects of photovoltaic solar cells with image transformers[J]. IEEE Access, 2021, 9:154292-154303. |
[7] |
DEITSCH S, CHRISTLEIN V, BERGER S, et al. Automatic classification of defective photovoltaic module cells in electroluminescence images[J]. Solar Energy, 2019, 185:455-468.
DOI URL |
[8] | CHEN S R, SHAN S, XIE L P, et al. A deep two-stage scheme for polycrystalline micro-crack detection[C]//Proceedings of the 2020 International Conference on Pattern Recognition and Intelligent Systems, Athens, Jul 2020. New York: ACM, 2020: 1-5. |
[9] | ZHANG N, SHAN S, WEI H, et al. Micro-cracks detection of polycrystalline solar cells with transfer learning[J]. Journal of Physics Conference Series, 2020, 1651:012118. |
[10] | JI R, SHAN S, ZHANG K, et al. The self-labeling of unsup-ervised polycrystalline solar cell micro-crack images[C]// Proceedings of the 5th International Conference on Mech-anical, Control and Computer Engineering, 2020: 2237-2240. |
[11] | HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Computer Society, 2016: 770-778. |
[12] | 程天艺, 王亚刚, 龙旭, 等. 多层次降维的头颈癌图像特征选择方法[J]. 计算机科学与探索, 2020, 14(4):669-679. |
CHENG T Y, WANG Y G, LONG X, et al. Multi-level dimensionality reduction of head and neck cancer image feature selection method[J]. Journal of Frontiers of Computer Science and Technology, 2020, 14(4):669-679. | |
[13] | WANG Y, YAO Q, KWOK J T, et al. Generalizing from a few examples: a survey on few-shot learning[J]. ACM Com- puting Surveys, 2020, 53(3):1-34. |
[14] | SNELL J, SWERSKY K, ZEMEL R S. Prototypical networks for few-shot learning[C]//Proceedings of the Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 4077-4087. |
[15] | SUNG F, YANG Y, LI Z, et al. Learning to compare: relation network for few-shot learning[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, Jun 18-23, 2018. Piscataway: IEEE, 2018: 1199-1208. |
[16] | YE H J, HU H, ZHAN D C, et al. Few-shot learning via embedding adaptation with set-to-set functions[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, Jun 13-19, 2020. Piscataway: IEEE, 2020: 8805-8814. |
[17] |
HOSPEDALES T M, ANTONIOU A, MICAELLI P, et al. Meta-learning in neural networks: a survey[J]. IEEE Transa-ctions on Pattern Analysis and Machine Intelligence, 2021. DOI: 10.1109/TPAMI.2021.3079209.
DOI |
[18] | FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]// Proceedings of the 34th International Conference on Machine Learning, Sydney, Aug 6-11, 2017: 1126-1135. |
[19] | RAVI S, LAROCHELLE H. Optimization as a model for few-shot learning[C]//Proceedings of the 5th International Conference on Learning Representations, Toulon, Apr 24-26, 2017. |
[20] | REN M, TRIANTAFILLOU E, RAVI S, et al. Meta-learning for semi-supervised few-shot classification[C]//Proceedings of the 6th International Conference on Learning Representa-tions, Vancouver, Apr 30-May 3, 2018. |
[21] | BENGIO Y, LECUN Y. Scaling learning algorithms towards AI[J]. Large-scale Kernel Machines, 2007, 34(5):1-41. |
[22] | VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 5998-6008. |
[23] | MÜLLER R, KORNBLITH S, HINTON G. When does label smoothing help?[C]//Proceedings of the Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, Vancouver, Dec 8-14, 2019. Red Hook: Curran Associates, 2019: 4694-4703. |
[24] | SCHROFF F, KALENICHENKO D, PHILBIN J. FaceNet: a unified embedding for face recognition and clustering[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, Jun 7-12, 2015. Washington: IEEE Computer Society, 2015: 815-823. |
[25] | SIMON C, KONIUSZ P, NOCK R, et al. Adaptive subspaces for few-shot learning[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, Jun 13-19, 2020. Piscataway: IEEE, 2020: 4135-4144. |
[1] | 安凤平, 李晓薇, 曹翔. 权重初始化-滑动窗口CNN的医学图像分类[J]. 计算机科学与探索, 2022, 16(8): 1885-1897. |
[2] | 曾凡智, 许露倩, 周燕, 周月霞, 廖俊玮. 面向智慧教育的知识追踪模型研究综述[J]. 计算机科学与探索, 2022, 16(8): 1742-1763. |
[3] | 彭豪, 李晓明. 多尺度选择金字塔网络的小样本目标检测算法[J]. 计算机科学与探索, 2022, 16(7): 1649-1660. |
[4] | 刘艺, 李蒙蒙, 郑奇斌, 秦伟, 任小广. 视频目标跟踪算法综述[J]. 计算机科学与探索, 2022, 16(7): 1504-1515. |
[5] | 赵小明, 杨轶娇, 张石清. 面向深度学习的多模态情感识别研究进展[J]. 计算机科学与探索, 2022, 16(7): 1479-1503. |
[6] | 张麟宇, 涂志莹, 杭少石, 张柏林, 初佃辉. 面向智慧康养的数据集构建方法及其应用[J]. 计算机科学与探索, 2022, 16(7): 1543-1551. |
[7] | 夏鸿斌, 肖奕飞, 刘渊. 融合自注意力机制的长文本生成对抗网络模型[J]. 计算机科学与探索, 2022, 16(7): 1603-1610. |
[8] | 孙方伟, 李承阳, 谢永强, 李忠博, 杨才东, 齐锦. 深度学习应用于遮挡目标检测算法综述[J]. 计算机科学与探索, 2022, 16(6): 1243-1259. |
[9] | 刘雅芬, 郑艺峰, 江铃燚, 李国和, 张文杰. 深度半监督学习中伪标签方法综述[J]. 计算机科学与探索, 2022, 16(6): 1279-1290. |
[10] | 程卫月, 张雪琴, 林克正, 李骜. 融合全局与局部特征的深度卷积神经网络算法[J]. 计算机科学与探索, 2022, 16(5): 1146-1154. |
[11] | 钟梦圆, 姜麟. 超分辨率图像重建算法综述[J]. 计算机科学与探索, 2022, 16(5): 972-990. |
[12] | 赵鹏飞, 谢林柏, 彭力. 融合注意力机制的深层次小目标检测算法[J]. 计算机科学与探索, 2022, 16(4): 927-937. |
[13] | 王忠民, 赵玉鹏, 郑镕林, 贺炎, 张嘉雯, 刘洋. 脑电信号情绪识别研究综述[J]. 计算机科学与探索, 2022, 16(4): 760-774. |
[14] | 许嘉, 韦婷婷, 于戈, 黄欣悦, 吕品. 题目难度评估方法研究综述[J]. 计算机科学与探索, 2022, 16(4): 734-759. |
[15] | 裴利沈, 赵雪专. 群体行为识别深度学习方法研究综述[J]. 计算机科学与探索, 2022, 16(4): 775-790. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||