基于迭代注意力归一化流的低光图像增强

针对网络层级间特征融合不足并缺乏高频特征的精准定位和获取,以及低光图像和多个正常曝光图像之间的不确定映射问题,提出一种迭代注意力归一化流(Iterative attention normalization flow,IANFlow)网络。迭代注意力模块使用空间和通道注意力对输入特征图的高频特征区域定位后进行特征获取,通过递进式层级定位和融合促使深层特征图包含更多的高频特征;可逆归一化流模块学习低光照图像和正常曝光图像之间复杂的条件分布以及将负对数似然(negative log likelihood,NLL)最小化建立低光图像和参考图像之间一对多的映射。在三个数据集上分别对比LLFlow网络,IANFlow网络的峰值信噪比(peak signal-to-noise ratio,PSNR)分别提高了1.1 dB、1.27 dB、2.14 dB。

基于归一化流概率模型的水电机组异常声音检测

为推进水电机组智能化运维的发展,提出了一种自注意多阶统计量池化(SAMOSP)归一化流条件概率模型(NFCPM)用于水电机组的无监督异常声音检测。文中首次提出了自注意多阶统计量池化模块。该模块首先用一维压缩卷积层和瓶颈压缩激励部分自注意到时间帧的权重向量。权重向量用来计算多阶统计池化向量。然后再分频段的自注意到多阶统计池化量的不同权重,并按其提取不同频段的重要统计量信息,从而得到时频图的自注意统计池化特征向量。接着用归一化流条件概率模型对从水轮机组正常音频信号中提取到的自注意统计池化特征向量进行正常数据的概率建模。不同时间段的测试样本在该正常数据概率模型中进行测试,并得到相应的分数。分数越低表示概率密度越低,则异常程度越大,从而实现水电机组音频信号的无监督异常检测。

RS-based research on the relationship between vegetation development and groundwater in upper-middle reaches region of the Yellow River basin

The relationship between the groundwater and the surface eco-environment in arid area is very close. In this paper, the authors extracted normalized difference vegetation index （NDVI） and vegetation conditional ratio （VCR） from MODIS images, and analyzed the relationships among NDVI, VCR and the measured data of groundwater of the same location in the research region. Based on this, the depth of groundwater suitable for vegetation growth in the upper-middle reaches of the Yellow River basin has been calculated. The results show that the depth of groundwater suitable for vegetation growth in the research region ranges from 0.8 to 4.5m, and the optimal groundwater depth is 1.2m. The method developed in this study is applicable to research the relationship between the groundwater and land surface vegetation environment on large-scale in arid area.