基于SVM的高通量dPCR基因芯片荧光图像分类研究

目的 为了实现高通量dPCR基因芯片荧光图像的亮点分类与计数，提出一种基于支持向量机（SVM）的荧光图像分类与计数方法。方法 首先对荧光图像进行去噪、对比度增强等图像预处理，对预处理后荧光图像进行亮点区域提取标注，去除背景与暗点的冗余信息，利用方向梯度直方图（Histogram of Oriented Gradient, HOG）提取鉴别特征，计算合并所有样本的亮点特征得到HOG特征向量，根据已得到的HOG特征向量创建一个线性SVM分类器，利用训练好的SVM分类器对荧光图像亮点进行分类与计数。结果 对比传统算法，文中算法具有较高的分类识别精度，平均准确率高达98%以上，可以很好地实现荧光图像亮点分类与计数。结论 在有限的小样本标注数据下，文中算法具有良好的分类性能，能够有效识别荧光图像中的亮点，对其他荧光图像分类研究也具有一定参考价值。

Classification of Fluorescent Images in High-throughput dPCR Gene Chips Based on SVM

The work aims to propose a fluorescence image classification and counting method based on support vector machine (SVM) to achieve the classification and counting of bright spots in high-throughput dPCR gene chip fluorescence image. Firstly, image preprocessing such as denoising and contrast enhancement was performed on the fluorescent image, and bright spot region was extracted annotated on the preprocessed fluorescent image to remove redundant information of background and dark points. A histogram of orientation gradient (HOG) was used to extract discriminative features, and the bright spot features of all samples were combined to obtain the HOG feature vector. A linear SVM classifier was created based on the obtained HOG feature vectors. The trained SVM classifier was used to classify and count the bright spots of the fluorescent image. Compared with traditional algorithms, the proposed algorithm had higher classification and recognition accuracy, with an average accuracy rate of more than 98%, which could well achieve the classification and counting of bright spots in fluorescent images. With limited small sample annotation data, the algorithm in this paper has good classification performance, can effectively identify bright spots in fluorescent images, and has certain reference value for other fluorescent image classification studies.