摘要
激活函数在超分辨率重建算法中可以增加算法的非线性、提高算法的复杂程度。利用ReLU激活函数在算法训练时间短的优点,针对负值通过ReLU激活函数置零导致部分神经元失活的问题,将失活的部分通过rReLU函数重新加入到模型中,该方法称为FSRCNN补充模块算法。实验过程中分别测试了FSRCNN模型在激活函数为ReLU、PReLU以及使用ReLU激活函数加入补充模块后的算法性能,结果表明:在放大倍数为4的条件下,补充模块算法的峰值信噪比结果高于原FSRCNN算法0.1dB。因此,补充模块能够提高模型的性能,增强模型对信息的提取。
The activation function can increase the nonlinearity of the algorithm and improve the complexity of the algorithm in the super-resolution reconstruction algorithm.Using the ReLU activation function,the algorithm has the advantage of short training time,and in view of the problem that the negative value is zeroed through the ReLU activation function,resulting in the inactivation of some neurons,the inactivated part is re-added to the model through the rReLU function,and the method is named FSRCNN supplementary module.The results show that the peak signal-to-noise ratio of the supplementary module algorithm is 0.1dB higher than that of the original FSRCNN algorithm under the condition of magnification of 4,so the supplementary module can improve the performance of the model and enhance the extraction of information by the model.
作者
陈蔚瑞
侯培国
CHEN Wei-rui;HOU Pei-guo(Institute of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China)
出处
《计量学报》
CSCD
北大核心
2023年第11期1667-1672,共6页
Acta Metrologica Sinica
基金
国家自然科学基金(61379065)。