基于卷积神经网络的蛋胚活性精准检测方法研究

孵化的蛋胚是生产禽流感疫苗的载体，蛋胚的活性检测是疫苗生产中的关键环节，通过光电容积脉搏法检测蛋胚活性是提高蛋胚活性检测准确率的关键。为了提高蛋胚活性检测效率和检测准确率，采用滑动功率谱方法（PSD）将蛋胚脉搏波可视化，基于卷积神经网络对蛋胚活性进行精准分类。实验结果显示，采用卷积神经网络对单个蛋胚信号的计算时间仅为12.6 ms，与人工检测方法相比，检测效率提高近200倍。可视化后的蛋胚脉搏波的卷积神经网络分类准确率可达94.14%，其中活胚、死胚和弱胚的真阳率分别为99.74%、93.73%、84.39%。基于卷积神经网络的蛋胚活性分类模型，可在大规模生产中精准地辨识蛋胚活性，对疫苗生产过程具有重要的应用价值。

Research on accurate detection method of egg embryo activity based on convolutional neural network

The hatched egg embryo is the carrier for the production of avian influenza vaccine. The activity detection of the egg embryo is a key link in the production of vaccine. The detection of the egg embryo activity by the photoelectric volume pulse method is the key to improve the accuracy of the egg embryo activity detection rate. In order to improve the detection efficiency and accuracy of egg embryo activity, the sliding power spectrum method was adopted to visualize egg embryo pulse wave, which accurately classified egg embryo activity based on convolutional neural network. The experimental results show that the calculation time of a single egg embryo signal using a convolutional neural network is only 12.6 ms, and the detection efficiency is increased by nearly 200 times in comparison with the manual detection method. The convolutional neural network classification accuracy of the visualized egg embryo pulse wave can reach 94.14%, among which the true positives of live embryos, dead embryos and weak embryos are 99.74%, 93.73%, and 84.39%, respectively. The egg embryo activity classification model based on the convolutional neural network can accurately identify the egg embryo activity in large-scale production, which has important application value for the vaccine production process.