【摘要】 自动目标识别是红外成像精确制导武器系统的关键技术,针对传统红外目标识别算法在复杂环境作战中存在目标特征建模复杂、识别率低等问题,提出一种基于改进的卷积神经网络(Convolutional Neural Network,CNN)方法。结合红外目标特性,调整ZFNet的卷积层和池化层数量,加入空间变换网络以提高对数据变换的鲁棒性;对Dropout层的丢弃率变化进行可视化分析并确定选取原则,以提高红外目标的识别率。通过试验结果与传统方法相比,该方法具有较高的识别率,能够为红外成像导引头目标识别算法设计提供参考。更多还原

【Abstract】 Automatic target recognition is a key technology of infrared imaging precision guided weapon system,aiming at the problems of complex target feature modeling and low recognition rate in the traditional infrared target recognition algorithm applied complex,the convolution neural network method based on improved the Dropout layer is proposed. Firstly,the number of ZFNet convolution layers and pooled layers are adjusted and the spatial transformation networks are in to improve the robustness of data transformation in combination with infrared target characteristics in ZFNet.Secondly,the discard rate of the Dropout layer is analyzed by visualization analysis during the process of training and the selection principle is determined to improve the recognition rate of infrared targets.The test results show that it has a higher recognition rate than the traditional method,it can provide reference for the design of infrared imaging seeker target recognition algorithm.更多还原