The existing activation functions ReLU, Tanh, and Mish have problems such as "neuronal death", offset, and poor robustness. Aiming at these problems, the XExp activation function is proposed by combining the advantages of ReLU, Swish, and Mish functions, and the problem of negative half-axis neuronal death is optimized by using the nonlinearity of non-RELU family functions and the non-zero characteristics of negative half-axis functions, and the soft saturation of negative semi-axis is retained. By designing the position of the origin of the function, the problem of positive half-axis offset in the Swish and Mish functions are solved. In terms of convergence speed, the MNIST dataset achieved 93.87% training accuracy during the first batch training on the newly proposed activation function XExp function, which was more than 85% higher in convergence speed compared with the Relu function; In terms of model convergence stability, compared with the accuracy of the Relu function, the XExp function can still achieve 98.05% accuracy when the number of convolutional layers is increased to 25 layers. The two data sets of CIFAR-10 and CIFAR-100 verify their versatility and practicality in the field of object detection.

Wenxiao Wei, Jieyu Liu, Qiang Shen, Yajing Wang. Design of nonlinear segmentation activation functions for object detection. Academic Journal of Computing & Information Science (2022), Vol. 5, Issue 13: 69-76. https://doi.org/10.25236/AJCIS.2022.051311.

[1] Li HW, Wu QX. Implementation scheme of neural network activation function in smart sensors[J]. Sensors and Microsystems,2014,33(1):46-48.

[2] Glorot X, Bengio Y. Understanding the di-fficultyof training deep feedforward neural networks[J]. Journal of Machine Learning Research,2010,9:249-256.

[3] Nair V, Hinton G E. Rectified linear units improve restricted boltzmann machines[C]. International Conference on Machine Learning, Omnipress, 2010:807-814.

[4] Dubey A K, Jain V. A comparative study of relu and leaky-relu activation functions for convolutional neural networks [M]. Applications of computing, automation and wireless systems in electrical engineering.Springer, Singapore, 2019: 873-880.

[5] He K, Zhang X, Ren S, et al. Delving deep into rectifiers: beyond human-level performance on imagenet classification [C]. Proceedings of the IEEE International Conference on Computer Vision. 2015: 1026-1034.

[6] Shi Q. Research and validation of image classification optimization algorithm based on convolutional neural network[D]. Beijing Jiaotong University,2017.

[7] WANG Hongxia, ZHOU Jiaqi, KU Chenghao, LIN Hong. Design of activation functions in convolutional neural networks for image classification[J]. Journal of Zhejiang University (Engineering Edition),2019,53(07):1363-1373.

[8] Clevert, Djork Arné, Unterthiner T, et al. Fast andaccurate deep network learning by e-xponential linear units (ELU) [J]. Computer S-cience, 2015.

[9] Ramachandran P, Zoph B, Le Q V. Searching for an Activation Functions[C]. ICLR 2018 Conference. 2017-10-27.

[10] BALDI P, SADOWSKI P, LU Zhiqin. Learning in the Machine:Random Backpropagation and the Deep Learning Channel[J]. Artificial Intelligence, 2018, 260: 1-35.

[11] Bu F. Research on small target detection and segmentation algorithm based on convolutional neural network[D]. Xi'an University of Electronic Science and Technology, 2019.

[12] NAIR V, HINTON G E. Rectified Linear Units Improve Restircted Boltzmann Machines[C] /Proceedings of the27th International Conference on Machine Learning.New York:ACM, 2010:807-814.

[13] HOCHREITER S. The Vanishing Gradient Problem during Learning Recurrent Neural Nets and Problem Solutions[J]. International Journal of Uncertainty Fuzziness and Knowledge-Based Systems, 1998, 6 (2) :107-116.

[14] LECUN Y, BENGIO Y, HINTON G. Deep learning [J]. Nature, 2015, 521:436-444.

[15] LIU YC, WANG TH, et al. A novel adaptive activation function for deep learning neural networks[J]. Journal of Jilin University (Science Edition),2019,57(4):857-859.

[16] Lin L. Small target detection based on deep learning [D]. University of Electronic Science and Technology,2020.

[17] ZHOU P, NI B B, GENG C, et al. Scale-transferrable object detection[C]/Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018 : 528-537.

[18] LI Zechao, TANG Jinhui. Weakly Supervised Deep Metric Learning for Community-Contributed Image Retrieval[J]. IEEE Transactions on Multimedia, 2015, 17 (11) : 1989-1999.

[19] He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Developing Deep into Rectifiers:Surpassing Human-Level Performance on ImageNet Classification[EB/OL].2015-02-06. https:// arxiv.org/abs/1502.01852.

[20] CLEVERT D A, UNTERTHINER T, HOCHREITER S. Fast and Accurate Deep Network Learning by Exponential Linear Units (ELU) [C/OL]//ICLR 2016.2016-02-22. https://arxiv.org/abs/1511.07289.

[21] Bei Su. Real-time target detection based on multi-scale neural network and self-attentive mechanism [D]. Xi'an University of Electronic Science and Technology,2020.

[22] Li Huihui, Zhou Kangpeng, Han Taichu. Improved SSD ship target detection based on CReLU and FPN[J]. Journal of Instrumentation, 2020,41(04):183-190.

[23] Xu Yankai, Liu Zengmei, Xue Yaru, Cao Siyuan. A seismic random noise suppression method applying two-channel convolutional neural network[J]. Petroleum Geophysical Exploration, 2022,57(04):747-756+735.DOI:10.13810/j.cnki.issn.1000-7210.2022.04.001.

[24] Su Zengzhi, Wang Meiling, Yang Chengzhi, Wu Hongchao. Radar signal modulation mode identification based on improved EfficientNet [J/OL]. Telecommunications Technology: 1-9 [2022-08-28].

[25] Pang C, Jiang Y, Wu T, Liao C-W, Ma W-G. Effect of neural network parameters on earthquake type recognition[J]. Science Technology and Engineering,2022,22(18):7765-7772.

[26] Liang Ruobing, Liu Bo, Sun Yuehong. Advances in deep learning empirical loss function geomorphic analysis [J/OL]. Systems Engineering Theory and Practice:1-14 [2022-08-28].