With the image monitoring as background, the key techniques related in designing real-time micro displacement monitor and control system based on PC-104 BUS are studied. On the basis of a concrete practical system requirement and modular design method, extracting the targets’ characteristic points using the center of gravity method, the preliminary system has been created which can detect 500m away and has the resolution of 1cm. In addition, this system can monitor multi-targets, communicate remotely and adapt to all kinds of weather conditions.
基于卷积结构的信号调制识别神经网络的识别性能受信号调制类型种类限制。例如,在12 d B信噪比条件下,同时对24种信号调制类型进行识别,其识别准确率仅为80%。若需要进一步提高识别性能,则要求更复杂的网络模型,导致网络训练所需数据集规模和硬件资源成本增大。鉴于此,针对无线电信号特征,设计一种适用于无线电信号调制识别的紧致残差神经网络,将其作为信号调制类型特征学习和特征提取工具,实现从原始I、Q数据到信号调制类型的端到端识别。利用迁移学习降低网络重新训练所需样本数,增强在无线信道响应发生变化时的环境适应能力,降低训练阶段所需的硬件资源和训练数据集规模。研究表明,当信道脉冲响应改变时,所提的信号调制识别神经网络在信噪比为12 d B条件下的识别性能达到95%,多个对比实验验证本文所设计神经网络的识别性能具有优势。
