基于深度卷积神经网络的γ光子图像分类识别研究

Research on Classification and Recognition ofγ-photon Images Based on Deep Convolutional Neural Network

通过对正电子探测成像技术获得的γ光子图像进行分类识别,有利于后续有针对性地快速获取图像所包含的有效信息。在MatConvNet上利用迁移学习的方法搭建深度卷积神经网络,通过对其参数的调整进一步提高网络分类识别的性能。为了验证网络性能,设计了10组不同形状的管材模型,利用仿真得到的扫描时间为1s的γ光子图像样本集对网络进行训练后,将其应用于扫描时间为0.1 s的γ光子图像样本进行分类识别,发现分类准确率在图像质量偏差、扫描时间为0.1 s的γ光子图像样本上仍然达到了94.72%。可见所搭建的深度卷积神经网络对γ光子图像具有很好的分类识别性能。

The classification and recognition ofγ-photon images obtained by positron detection imaging technology facilitates the subsequent processing of acquiring the valid information contained in the image quickly.The transfer learning method is used to build the deep convolutional neural network on MatConvNet.And the performance of the classification and recognition of the network is improved by parameter adjustment.To verify the performance of the network,10 different shapes of pipe models are designed.γ-photon images of scanning time 1 s is used to train the network,and then the trained network is applied to the classification and recognition ofγ-photon images of scanning time 0.1 s.The classification accuracy rate can still reach 94.72%on theγ-photon images with the worse quality.It can be seen that the deep convolutional neural network has good classification and recognition performance forγ-photon images.