传统高通量dPCR荧光图像分析结果易因假阳性点与非特异性扩增而导致阳性点识别率较低，因而本文提出一种多特征融合高通量dPCR荧光图像识别方法（HDFINet），以提高阳性点识别准确性。首先，在特征融合部分引入自上而下结构，使得下层特征在顶层被更有效地利用。在自上而下结构中，使用通道注意力来分配荧光图像通道权重，并使用空间注意力来分配特征图中荧光图像像素相应权重，使得特征映射能够更好地响应荧光图像特征。然后，在RPN中使用自适应交并比IOU计算阳性点包围框置信度，减少阳性点信息丢失可能性。最后，ROI Align将荧光图像候选区域中阳性点特征重新固定尺寸后，输入至全连接层和全卷积层，进行类别与回归框回归，输出阳性点识别结果。本文提出的HDFINet网络具有较高识别率，可以有效地实现荧光图像阳性点识别，与YOLOv4、VF-Net、GROIE相比，本文方法综合指标F1最高，相比于经典的深度学习网络Mask R-CNN网络，本方法对阳性点识别真阳性率提高了1.13%，阳性预测值提高了0.36%，综合指标F1的值提高了0.75%。本文提出的HDFINet网络具有良好的识别性能，能够有效识别荧光图像阳性点，对其他荧光图像分析研究具有参考价值。

Fluorescence microscopy technology uses fluorescent dyes to provide highly specific visualization of cell components, which plays an important role in understanding the subcellular structure. However, fluorescence microscopy has some limitations such as the risk of non-specific cross labeling in multi-labeled fluorescent staining and limited number of fluorescence labels due to spectral overlap. This paper proposes a deep learning-based fluorescence to fluorescence (Fluo-Fluo) translation method, which uses a conditional generative adversarial network to predict a fluorescence image from another fluorescence image and further realizes the multi-label fluorescent staining. The cell types used include human motor neurons, human breast cancer cells, rat cortical neurons, and rat cardiomyocytes. The effectiveness of the method is verified by successfully generating virtual fluorescence images highly similar to the true fluorescence images. This study shows that a deep neural network can implement Fluo-Fluo translation and describe the localization relationship between subcellular structures labeled with different fluorescent markers. The proposed Fluo-Fluo method can avoid non-specific cross labeling in multi-label fluorescence staining and is free from spectral overlaps. In theory, an unlimited number of fluorescence images can be predicted from a single fluorescence image to characterize cells.