随着计算机视觉领域的迅猛发展，深度学习技术已经在图像处理、目标识别等任务中取得了成就。然而，研究正逐渐转向视频，因为生活和工作环境中随处可见监控摄像头，如果仅靠人工监控，每时每刻产生的大量监控视频将耗费大量人力、财力和物力。人体动作识别任务也成为了监控视频下的一个重要任务，人体动作识别任务的场景通常是在良好的视觉条件下进行的，而在夜间视觉场景下研究较少。因此，本文针对现有的夜间人体动作识别算法准确率低的问题，对相关技术进行了研究与应用。本课题完成的主要工作与创新如下：

       （1）针对传统图像增强算法中参数固定而导致增强后的图像各个区域无法得到有效提升，且基于深度学习的图像增强算法太过于依赖于配对训练的数据集等问题，本文提出了一种基于MDIFE-Net曲线估计的夜间图像增强算法。首先，基于灰度变换方法设计了一种光照估计曲线，通过光照估计曲线对图像进行像素级的调整，将夜间低光图像域映射到增强图像域，有效消除光照不足所带来的影响；其次，提出了基于Mish函数的深度光照特征提取网络(Mish Deep Illumination Feature Extraction Network, MDIFE-Net)提取图像特征，去掉了无参考深度曲线估计网络模型所有的下采样层和批处理归一化层，防止其破坏相邻像素之间的关系，用更加平滑的Mish激活函数代替了Relu激活函数，从而可以使参数更好地进行更新；最后，设计了一种联合多项损失的光照估计损失函数来驱动夜间图像增强算法，解决了成对数据集难以构建的问题。实验结果表明，本文算法在夜间ARID数据集上的NIQE和STD指标结果分别达到了12.283和67.472，相较于新颖的Zero-DCE算法，分别降低和提升了1.866和13.605，能够有效提升夜间图像的清晰度和对比度，为后续人体动作识别提供了良好的基础。

       （2）针对深度学习领域中，人体动作识别算法对时间信息、空间信息以及背景信息总是进行同等处理，而造成人体动作识别算法精度不高的问题，本文提出了一种基于时空信息交互的人体动作识别算法。首先，提出了一个双路径网络以不同的刷新率分别学习空间和时间信息，包括一个在低帧率下运行以捕获空间语义信息的稀疏路径，以及一个并行的在高帧率下运行以捕获时序运动信息的密集路径；其次，为了从视频中提取更具有区分性的特征，提出了交叉双注意力交互模型将注意力集中在视频片段的重点区域，并在两条路径之间明确的交换时空信息。实验结果表明，本文算法在UCF101数据集和HMDB51数据集上的准确率分别达到了97.6%和78.4%，相较于新颖的Slowfast算法分别提升了1.8%和1.4%，取得了更高的准确率。结合基于MDIFE-Net曲线估计的夜间图像增强算法在夜间ARID数据集上的准确率达到了83.2%，比图像增强前的动作识别准确率提升了22.9%，能够有效的识别夜间人体动作，具有良好的实战意义。

       （3）本文将所提出的夜间图像增强模型与人体动作识别模型进行实际应用。通过系统的需求分析，设计并实现了一套基于B/S架构的夜间人体动作识别系统，并对结果进行了可视化的展示，最后对该系统进行了功能测试，得到了能够满足用户需求的夜间人体动作识别系统。

       综上所述，本文的工作主要从夜间图像增强和人体动作识别两个方向展开研究，针对夜间人体动作识别算法准确率低的问题，在夜间图像增强算法和人体动作识别算法上进行了改进和优化，搭建了相应的网络结构，通过实验进行了验证，达到了预期的研究目标，并将所提出的算法落地实用，搭建了一套基于B/S架构的夜间人体动作识别系统。

       With the rapid development of computer vision, deep learning technology has made achievements in image processing, target recognition and other tasks. However, research is gradually turning to video because surveillance cameras can be seen everywhere in your life and work environment. If you only rely on manual surveillance, the large amount of surveillance video generated at any time will consume a lot of manpower, money and material resources. Human motion recognition task has also become an important task under surveillance video. Scenes of human motion recognition task are usually performed under good visual conditions, but less studied under night vision. Therefore, in order to solve the problem of low accuracy of the existing night motion recognition algorithms, the related technologies are studied and applied. The main work and innovations completed in this project are as follows:

       （1）A night image enhancement algorithm based on MDIFE-Net curve estimation is proposed to solve the problem that the parameters of traditional image enhancement algorithms are fixed, resulting in the enhancement of each area of the image cannot be effectively improved, and the deep learning-based image enhancement algorithm is too dependent on paired training datasets. First, an illumination estimation curve is designed based on the gray transformation method. By adjusting the pixel level of the image with the illumination estimation curve, the night low-light image domain is mapped to the enhanced image domain, which effectively eliminates the impact of the insufficient illumination. Secondly, a Mish Deep Illumination Feature Extraction Network (MDIFE-Net) based on Mish function is proposed to extract image features, eliminating all down-sampling layers and batch normalization layers of the network model without reference depth curve estimation, so as to prevent them from destroying the relationship between adjacent pixels, and replacing the Relu activation function with a smoother Mish activation function, so that the parameters can be updated better. Finally, a light loss estimation function combined with multiple losses is designed to drive the night image enhancement algorithm, which solves the problem that paired datasets are difficult to build. The experimental results show that the NIQE and STD indices obtained by this algorithm on night ARID dataset are 12.283 and 67.472, respectively. Compared with the novel Zero-DCE algorithm, the NIQE and STD indices are reduced and improved by 1.866 and 13.605, respectively, which can effectively improve the sharpness and contrast of night images, and provide a good basis for subsequent human motion recognition.

       （2）In the field of deep learning, human motion recognition algorithms always process time information, spatial information and background information equally, which results in low accuracy of human motion recognition algorithms. This paper presents a human motion recognition algorithm based on space-time information interaction. First, a two-path network is proposed to learn spatial and temporal information at different refresh rates, including a sparse path running at a low frame rate to capture spatial semantic information and a dense parallel path running at a high frame rate to capture temporal motion information. Secondly, in order to extract more distinctive features from the video, a cross-bi-attention interaction model is proposed, which focuses attention on the key areas of the video clips and explicitly exchanges space-time information between the two paths. The experimental results show that the accuracy of this algorithm on UCF101 and HMDB51 datasets is 97.6% and 78.4%, respectively, which is 1.8% and 1.4% higher than that of the novel Slowfast algorithm. The night image enhancement algorithm combined with MDIFE-Net curve estimation achieves 83.2% accuracy on night ARID dataset and 22.9% higher accuracy than motion recognition before image enhancement. It can effectively recognize night human movements and has good practical significance.

       （3）The night image enhancement model and human motion recognition model proposed in this paper are applied in practice. Through the system requirements analysis, a night human motion recognition system based on B/S architecture is designed and implemented, and the results are visualized. Finally, the function of the system is tested, and a night human motion recognition system that can meet the needs of users is obtained.

       In summary, this paper mainly studies night image enhancement and human motion recognition. To solve the problem of low accuracy of night human motion recognition algorithm, the night image enhancement algorithm and human motion recognition algorithm are improved and optimized, and the corresponding network structure is built. The experimental results verify that the expected research goals are achieved, and the proposed algorithm is practical. A night motion recognition system based on B/S architecture is built.

[1]Zhao S, Blaabjerg F, Wang H. An overview of artificial intelligence applications for power electronics[J]. IEEE Transactions on Power Electronics, 2020, 36(4): 4633-4658.

[2]闫航. 康复训练场景下的动作与行为识别方法研究[D]. 郑州: 郑州大学, 2020.

[3]Mabrouk A B, Zagrouba E. Abnormal behavior recognition for intelligent video surveillance systems: A review[J]. Expert Systems with Applications, 2018, 91: 480-491.

[4]乔阳阳. 面向智能家居的动作识别关键技术研究[D]. 辽宁: 辽宁科技大学, 2021.

[5]Chen J, Samuel R D J, Poovendran P. LSTM with bio inspired algorithm for action recognition in sports videos[J]. Image and Vision Computing, 2021, 112: 104-214.

[6]黄勇康, 梁美玉, 王笑笑, 陈徵, 曹晓雯. 基于深度时空残差卷积神经网络的课堂教学视频中多人课堂行为识别[J]. 计算机应用, 2022, 42(3): 736-742.

[7]贺斌. 基于深度学习的考场作弊行为分析与研究[D], 成都: 电子科技大学, 2021.

[8]陆晴. 基于深度学习的异常行为识别算法研究[D], 哈尔滨: 哈尔滨工业大学, 2018.

[9]中华人民共和国统计局. 中国统计年鉴[M]. 北京: 中国统计出版社, 2018: 22.

[10]Dhal K G, Das A, Ray S, Gálvez J, Das S. Histogram equalization variants as optimization problems: a review[J]. Archives of Computational Methods in Engineering, 2021, 28(3): 1471-1496.

[11]Cheng H, Long W, Li Y, Liu H. Two low illuminance image enhancement algorithms based on grey level mapping[J]. Multimedia Tools and Applications, 2021, 80: 7205-7228.

[12]Gandhamal A, Talbar S, Gajre S, Hani A F M, Kumar D. Local gray level S-curve transformation–a generalized contrast enhancement technique for medical images[J]. Computers in biology and medicine, 2017, 83: 120-133.

[13]Xu Y, Li D, Tang J. Single frame shadow segmentation based on image enhancement for video SAR[C]//Sixth International Workshop on Pattern Recognition. 2021, 11913: 28-35.

[14]Huang Z, Wang Z, Zhang J, Li Q, Shi Y. Image enhancement with the preservation of brightness and structures by employing contrast limited dynamic quadri-histogram equalization[J]. Optik, 2021, 226: 1-9.

[15]Tan S F, Isa N A M. Exposure based multi-histogram equalization contrast enhancement for non-uniform illumination images[J]. IEEE Access, 2019, 7: 70842-70861.

[16]Zhao L, Abdelhamed A, Brown M S. Learning Tone Curves for Local Image Enhancement[J]. IEEE Access, 2022, 10: 60099-60113.

[17]Wu X, Kawanishi T, Kashino K. Reflectance-guided histogram equalization and comparametric approximation[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 31(3): 863-876.

[18]He L, Long W, Liu S, Li Y, Ding W. A night low‐illumination image enhancement model based on small probability area filtering and lossless mapping enhancement[J]. IET Image Processing, 2021, 15(13): 3221-3238.

[19]Guo X, Li Y, Ling H. LIME: Low-light image enhancement via illumination map estimation[J]. IEEE Transactions on image processing, 2016, 26(2): 982-993.

[20]Wang R, Zhang Q, Fu C W, Shen X, Zheng W S, Jia J. Underexposed photo enhancement using deep illumination estimation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 6849-6857.

[21]Wang P, Wang Z, Lv D, Zhang C, Wang Y. Low illumination color image enhancement based on Gabor filtering and Retinex theory[J]. Multimedia Tools and Applications, 2021, 80(12): 17705-17719.

[22]Li M, Liu J, Yang W, Sun X, Guo Z. Structure-revealing low-light image enhancement via robust retinex model[J]. IEEE Transactions on Image Processing, 2018, 27(6): 2828-2841.

[23]Ma L, Liu R, Wang Y, Fan X, Luo Z. Low-light image enhancement via self-reinforced retinex projection model[J]. IEEE Transactions on Multimedia, 2022: 1-14.

[24]Lv F, Lu F, Wu J, Lim C S. Mbllen: low-light image/video enhancement using cnns[C]//BMVC. 2018, 220(1): 1-14.

[25]Zhu M, Pan P, Chen W, Yang Y. Eemefn: low-light image enhancement via edge-enhanced multi-exposure fusion network[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(7): 13106-13113.

[26]Zhu G, Ma L, Liu R, Fan X, Luo Z. Collaborative reflectance-and-illumination learning for high-efficient low-light image enhancement[C]//2021 IEEE International Conference on Multimedia and Expo (ICME). 2021: 1-6.

[27]Zhao L, Lu S P, Chen T, Yang Z, Shamir A. Deep symmetric network for underexposed image enhancement with recurrent attentional learning[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2021: 12075-12084.

[28]Shi Y, Wu X, Zhu M. Low-light image enhancement algorithm based on retinex and generative adversarial network[J]. arXiv preprint arXiv:1906.06027, 2019: 1-9.

[29]Ignatov A, Kobyshev N, Timofte R, Vanhoey K, Van Gool L. Dslr-quality photos on mobile devices with deep convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. 2017: 3277-3285.

[30]Jiang Z, Qin L. Low-light image enhancement method based on U-net generative adversarial network[J]. Acta Electonice Sinice, 2020, 48(2): 258-264.

[31]Lore K G, Akintayo A, Sarkar S. Llnet: a deep autoencoder approach to natural low-light image enhancement[J]. Pattern Recognition, 2017, 61: 650-662.

[32]Zhang Y, Zhang J, Guo X. Kindling the darkness: A practical low-light image enhancer[C]//Proceedings of the 27th ACM international conference on multimedia. 2019: 1632-1640.

[33]Shen L, Yue Z, Feng F, Chen Q, Liu S, Ma J. Msr-net: low-light image enhancement using deep convolutional network[J]. arXiv preprint arXiv:1711.02488, 2017: 1-9.

[34]Gharbi M, Chen J, Barron J T, Hasinoff S W, Durand F. Deep bilateral learning for real-time image enhancement[J]. ACM Transactions on Graphics (TOG), 2017, 36(4): 1-12.

[35]Wei C, Wang W, Yang W, Liu J. Deep retinex decomposition for low-light enhancement[J]. arXiv preprint arXiv:1808.04560, 2018: 1-12.

[36]Chen C, Chen Q, Xu J, Koltun V. Learning to see in the dark[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 3291-3300.

[37]Alaspure P, Hambarde P, Dudhane A, et al. DarkGAN: night image enhancement using generative adversarial networks[C]//Computer Vision and Image Processing: 5th International Conference. 2021: 293-302.

[38]Liu Z, Wang K, Wang Z, Lu H, Yuan L. PatchNet: a tiny low-light image enhancement net[J]. Journal of Electronic Imaging, 2021, 30(3): 1-13.

[39]Zhu J Y, Park T, Isola P, Efros A A. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//Proceedings of the IEEE international conference on computer vision. 2017: 1-10.

[40]Chan K C K, Wang X, Xu X, Gu J, Loy C C. Glean: generative latent bank for large-factor image super-resolution[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 14245-14254.

[41]Jiang Y, Gong X, Liu D, Cheng Y, Fang C, Shen X, Yang J, Zhou P, Wang Z. EnlightenGAN: deep light enhancement without paired supervision[J]. IEEE transactions on image processing, 2021, 30: 2340-2349.

[42]Guo C, Li C, Guo J, Loy C C, Hou J, Kwong S, Cong R. Zero-reference deep curve estimation for low-light image enhancement[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 1780-1789.

[43]Abu‐Bakar S A R. Advances in human action recognition: an updated survey[J]. IET Image Processing, 2019, 13(13): 2381-2394.

[44]Fan Z, Zhao X, Lin T, Su H. Attention-based multiview re-observation fusion network for skeletal action recognition[J]. IEEE Transactions on Multimedia, 2018, 21(2): 363-374.

[45]Liao Z, Hu H, Liu Y. Action recognition with multiple relative descriptors of trajectories[J]. Neural processing letters, 2020, 51(1): 287-302.

[46]Shi L, Zhang Y, Cheng J, Lu H. Two-stream adaptive graph convolutional networks for skeleton-based action recognition[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 12026-12035.

[47]Li T, Fan L, Zhao M, Liu Y, Katabi D. Making the invisible visible: action recognition through walls and occlusions[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 872-881.

[48]Simonyan K, Zisserman A. Two-stream convolutional networks for action recognition in videos[J]. Advances in neural information processing systems, 2014, 27: 1-9.

[49]Chen H, Li M, Jing L, Cheng Z. Lightweight long and short-range spatial-temporal graph convolutional network for skeleton-based action recognition[J]. IEEE Access, 2021, 9: 161374-161382.

[50]Lin J, Gan C, Han S. Tsm: temporal shift module for efficient video understanding[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019: 7083-7093.

[51]Feichtenhofer C, Fan H, Malik J, He K. Slowfast networks for video recognition[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019: 6202-6211.

[52]Pang C, Lu X, Lyu L. Skeleton-based Action Recognition through Contrasting Two-Stream Spatial-Temporal Networks[J]. IEEE Transactions on Multimedia, 2023: 1-4.

[53]Tran D, Bourdev L, Fergus R, Torresani L, Paluri M. Learning spatiotemporal features with 3d convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. 2015: 4489-4497.

[54]Tran D, Ray J, Shou Z, Chang S F, Paluri M. Convnet architecture search for spatiotemporal feature learning[J]. arXiv preprint arXiv:1708.05038, 2017: 1-12.

[55]Carreira J, Zisserman A. Quo vadis, action recognition? a new model and the kinetics dataset[C]//proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 6299-6308.

[56]Feichtenhofer C. X3d: Expanding architectures for efficient video recognition[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 203-213.

[57]Li J, Han Y, Zhang M, Li G, Zhang B. Multi-scale residual network model combined with Global Average Pooling for action recognition[J]. Multimedia Tools and Applications, 2022, 81(1): 1375-1393.

[58]Donahue J, Anne Hendricks L, Guadarrama S, Rohrbach M, Venugopalan S, Saenko K, Darrell T. Long-term recurrent convolutional networks for visual recognition and description[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 2625-2634.

[59]Si C, Jing Y, Wang W, Wang L, Tan T. Skeleton-based action recognition with spatial reasoning and temporal stack learning[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 103-118.

[60]Li Z, Gavrilyuk K, Gavves E, Jain M, Snoek C G. Videolstm convolves, attends and flows for action recognition[J]. Computer Vision and Image Understanding, 2018, 166: 41-50.

[61]Aljarrah A A, Ali A H. Human activity recognition using PCA and BiLSTM recurrent neural networks[C]//2019 2nd International Conference on Engineering Technology and its Applications (IICETA). 2019: 156-160.

[62]Chenhao W, Yongquan W E I, Dong G U O, Jun G. Human behavior recognition under occlusion based on two-stream network combined with BiLSTM[C]//2020 Chinese Control And Decision Conference (CCDC). 2020: 3311-3316.

[63]Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, Kaiser L, Polosukhin I. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30: 1-11.

[64]Wang Q, Wu B, Zhu P, Li P, Zuo W, Hu Q. ECA-Net: efficient channel attention for deep convolutional neural networks[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 11534-11542.

[65]Woo S, Park J, Lee J Y, Kweon I S. Cbam: convolutional block atention module[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.

[66]Fu J, Liu J, Tian H, Li Y, Bao Y, Fang Z, Lu H. Dual attention network for scene segmentation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 3146-3154.

[67]Pan B, Cao Z, Adeli E, Niebles J C. Adversarial cross-domain action recognition with co-attention[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(7): 11815-11822.

[68]Liu Z, Luo D, Wang Y, Wang L, Tai Y, Wang C, Li J. Huang F, Lu T, Teinet: towards an efficient architecture for video recognition[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(7): 11669-11676.

[69]Wang X, Xiong X, Neumann M, Piergiovanni A J, Ryoo M S, Angelova A, Kitani K M, Hua W. Attentionnas: spatiotemporal attention cell search for video classification[C]//Computer Vision–ECCV 2020: 16th European Conference, 2020: 449-465.

[70]Zhou J, Yao J, Zhang W, Zhang D. Multi-scale retinex-based adaptive gray-scale transformation method for underwater image enhancement[J]. Multimedia Tools and Applications, 2022: 1-21.

[71]Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference. 2015: 234-241.

[72]Hu Z Y, Lee E J. Human motion recognition based on improved 3-dimensional convolutional neural network[C]//2019 IEEE International Conference on Computation, Communication and Engineering (ICCCE). 2019: 154-156.

[73]He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[74]Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.

[75]Fukui H, Hirakawa T, Yamashita T, Fujiyoshi H. Attention branch network: learning of attention mechanism for visual explanation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 10705-10714.

[76]Agarap A F. Deep learning using rectified linear units (relu)[J]. arXiv preprint arXiv:1803.08375, 2018: 1-7.

[77]Misra D. Mish: A self regularized non-monotonic activation function[J]. arXiv preprint arXiv:1908.08681, 2019: 1-14.

[78]Ma L, Ma T, Liu R, Fan X, Luo Z, Toward fast, flexible, and robust low-light image enhancement[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022: 5637-5646.

[79]Hernandez-Juarez D, Parisot S, Busam B, Leonardis A, Slabaugh G, McDonagh S. A multi-hypothesis approach to color constancy[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 2270-2280.

[80]Cai J, Gu S, Zhang L. Learning a deep single image contrast enhancer from multi-exposure images[J]. IEEE Transactions on Image Processing, 2018, 27(4): 2049-2062.

[81]Xu Y, Yang J, Cao H, Mao K, Yin J, See S. ARID: a new dataset for recognizing action in the dark[C]//Deep Learning for Human Activity Recognition: Second International Workshop. 2021: 70-84.

[82]McGaughey D, Potvin G. Quality Metrics for Atmospherically Distorted Images[C]//Propagation Through and Characterization of Atmospheric and Oceanic Phenomena. Optica Publishing Group, 2021: 1-6.

[83]Jin X, Jiang Q, Yao S, Zhou D, Nie R, Hai J, He K. A survey of infrared and visual image fusion methods[J]. Infrared Physics & Technology, 2017, 85: 478-501.

[84]Xu J, Li Z, Du B, Zhang M, Liu J. Reluplex made more practical: leaky relu[C]//2020 IEEE Symposium on Computers and communications (ISCC). IEEE, 2020: 1-7.

[85]Crnjanski J, Krstić M, Totović A, Pleros N, Gvozdić D. Adaptive sigmoid-like and PRelu activation functions for all-optical perceptron[J]. Optics Letters, 2021, 46(9): 2003-2006.

[86]Rasamoelina A D, Adjailia F, Sinčák P. A review of activation function for artificial neural network[C]//2020 IEEE 18th World Symposium on Applied Machine Intelligence and Informatics (SAMI). 2020: 281-286.

[87]Niu Z, Zhong G, Yu H. A review on the attention mechanism of deep learning[J]. Neurocomputing, 2021, 452: 48-62.

[88]Picaud S, Dalkara D, Marazova K, Goureau O, Roska B, Sahel J A. The primate model for understanding and restoring vision[J]. Proceedings of the National Academy of Sciences, 2019, 116(52): 26280-26287.

[89]Dai W, Chen Y, Huang C, Gao M K, Zhang X, Two-stream convolution neural network with video-stream for action recognition[C]//2019 International Joint Conference on Neural Networks (IJCNN). IEEE, 2019: 1-8.

[90]Wu M C, Chiu C T, Wu K H. Multi-teacher knowledge distillation for compressed video action recognition on deep neural networks[C]//ICASSP 2019-2019 IEEE International Conference on Acoustics. 2019: 2202-2206.