基于改进U-net和CNN的绝缘子自爆检测方法研究

针对绝缘子自爆故障人工检测效率低,成本高的问题,基于改进U-net和卷积神经网络（CNN）模型,提出一种可有效识别绝缘子自爆故障的双阶段目标检测算法。首先,在语义分割阶段使用改进U-net模型,通过翻倍提高图像分辨率的方法有效提高图像分割精度。其次,在图像分类阶段提出更适合所提问题且有效提高分类准确度的新型CNN模型。最后,使用无人机拍摄的绝缘子图片为实验数据进行实验。实验结果表明所提算法识别精度较高。     

X-parameter measurement on a GaN HEMT device: complexity reduction study of load-pull characterization test setup

Characterization of power transistors is an indispensable step in the design of radio frequency and microwave power amplifiers. A full harmonic load-pull measurement setup is normally required for the accurate and comprehensive characterization of RF power transistors. The setup is usually highly complex, leading to a relatively high hardware cost and low measurement throughput. This paper presents X-parameter measurement on a gallium nitride(GaN) high-electron-mobility transistor and studies the potential of utilizing an X-parameter-based modeling technique to highly reduce the complexity of the harmonic load-pull measurement setup for transistor characterization. During the X-parameter measurement and characterization, load impedance of the device is tuned and controlled only at the fundamental frequency and is left uncontrolled at other higher harmonics. However,it proves preliminarily that the extracted X-parameters can still predict the behavior of the device with moderate to high accuracy, when the load impedance is tuned up to the third-order harmonic frequency. It means that a fundamental-only load-pull test setup is already enough even though the device is to be characterized under load tuning up to the third-order harmonic frequency, by utilizing X-parameters.