基于卷积神经网络的细胞识别

Cell recognition based on convolutional neural network

结合深度学习理论,将卷积神经网络算法运用到细胞识别上。相比传统的细胞识别算法,基于卷积神经网络的细胞识别使流程变得简单,同时也使得细胞的识别率更高。与多层神经网络、支持向量机及决策树等机器学习算法相比,卷积神经网络算法由于本身网络的复杂度以及训练集的大样本量,其深度远大于传统的机器学习算法,能较这些手工提取特征的方法更好地表达特征和区分细胞,最终取得的分类效果也要优于前者。研究结果表明卷积神经网络算法能较好地用于细胞识别。

We introduce a strategy for cell recognition by combining the convolutional neural network（CNN） algorithm with the theory of deep learning. Compared with the conventional cell recognition algorithms, the CNN- based cell recognition simplified the cell recognition processes and achieved a higher recognition rate. Due to the complexity of network and large sample size of the training set, the depth of the CNN algorithm was much greater than that of such conventional machine learning algorithms as the multi-layer perceptron, support vector machine, and decision tree. Compared with the conventional methods that required manual extraction of the features, CNN algorithm allowed better expression of the features and differential recognition of the cells to achieve a better classification result. The results of the experiment demonstrated that the CNN algorithm can be effectively applied in cell recognition.