排序方式: 共有15条查询结果,搜索用时 10 毫秒
1.
2.
3.
为了提高基于深度学习的图像降噪效率,提出了一种基于Res2-Unet-SE的多阶段监督深度残差(Multi-stage Supervised Deep Residual,MSDR)降噪神经网络。首先基于该神经网络,将图像降噪分为多阶段处理过程;然后在各处理阶段,将不同分辨率图像块输入到Res2-Unet子网络中获取不同尺度特征信息,并通过通道注意力机制将自适应学习的特征融合信息传递到下阶段;最后将不同尺度特征信息叠加,完成高质量的图像降噪。实验选择BSD400数据集用于训练,通过Set12数据集进行高斯噪声的降噪测试;通过SIDD数据集完成真实噪声的降噪测试。通过与常见的降噪神经网络对比表明,对图像添加σ=15,25,50的高斯噪声时,经本文算法降噪后的图像PSNR比对高斯噪声消除性能较好的DNCNN分别提高0.03 dB,0.05 dB,0.14 dB;在σ=25,50时,相较于MPRNET分别提高了0.02 dB, 0.06 dB。对含真实噪声的图像,经本文算法降噪后的图像PSNR比CBDNET算法提高0.48 dB。实验分析表明,本文算法在图像降噪上具有较高的鲁棒性,不仅能从噪声... 相似文献
4.
5.
The traditional eye sensitivity function based on photopic vision is not applicable in the mesopic vision state. The mesopic vision is studied by using the photopic and scotopic vision state sensitivity functions. And the model which links the mesopic sensitivity with the photopic and scotopic states is built. Based on the model, the luminous efficacy of mesopic vision is calculated and applied to the spectrum distribution of LED light sources. The results show that the luminous efficacy of a commercial YAG phosphor converted white LED is up to 172.7 lm/W at mesopic vision, which is 67.2 % higher than that of photopic vision state. We also conclude that the optimal spectral power distribution of the LED will greatly increase the mesopic luminous efficacy. 相似文献
6.
7.
8.
9.
10.
为产生频率间隔相同而又平坦的光频梳,基于单个多量子阱电吸收调制器强度调制特性,设计出一种新型超平坦双光梳产生方案。通过在电吸收调制器前置矩形滤波器,精密控制多量子阱电吸收调制器的反向偏置电压与射频驱动信号幅度,滤除频谱的中心谱线后,得到了平坦度为0.01dB的双光梳。利用Optisystem7.0软件进行仿真,其对不同线宽(100kHz、10 MHz和20 MHz)的激光光源均可产生位于中心谱线两侧对称的、带宽均为300 GHz、谱线均为15条、谱线间距均为20GHz以及平坦度可达0.01dB的双光梳。 相似文献