We introduce a novel technique for knowledge transfer, where knowledge from a pretrained deep neural network (DNN) is distilled and transferred to another DNN. As the DNN performs a mapping from the input space to the output space through many layers sequentially, we define the distilled knowledge to be transferred in terms of flow between layers, which is calculated by computing the inner product between features from two layers. When we compare the student DNN and the original network with the same size as the student DNN but trained without a teacher network, the proposed method of transferring the distilled knowledge as the flow between two layers exhibits three important phenomena: (1) the student DNN that learns the distilled knowledge is optimized much faster than the original model, (2) the student DNN outperforms the original DNN, and (3) the student DNN can learn the distilled knowledge from a teacher DNN that is trained at a different task, and the student DNN outperforms the original DNN that is trained from scratch.

Inverse synthetic aperture radar (ISAR) is a coherent imaging system formed by conducting signal processing on the received data consisting of radar cross section reflected from a maneuvering target. Autofocus is an essential step of ISAR imaging, whose performance has a great influence on the quality of the radar image. Minimum Shannon entropy phase adjustment (MSEPA) and minimum Renyi entropy-based algorithm (MREA) have been widely used to compensate for the phase error in ISAR autofocus. However, MSEPA and MREA have some drawbacks in terms of the noise sensitivity and computational efficiency. Tsallis entropy (nonextensive entropy) is a useful measure to describe the thermostatistical properties of physical systems. This paper concentrates on the performance of minimum Tsallis entropy phase adjustment (MTEPA) instead of the Shannon entropy. By minimizing the Tsallis entropy of an ISAR image, the MTEPA can significantly improve the computational efficiency, while retaining the image focal quality of the restored ISAR images, as compared to MSEPA and MREA. The order q of Tsallis entropy can be experimentally found by analyzing both an image quality metric and the computation time. In experimental results, the effectiveness of the MTEPA is illustrated and analyzed with simulated targets consisting of point scatterers and measured Boeing-747 data.

In this paper, we propose a novel cross-range scaling technique to estimate the rotational velocity (RV) of a maneuvering target. The proposed method includes three steps. First, a feature from accelerated segment test (FAST) is applied to two sequential inverse synthetic aperture radar (ISAR) images to find the locations of their robust feature points. Second, the rotation angle (RA) is estimated using two major axes, which are obtained using a principal component analysis (PCA) of the two feature data sets scaled by a candidate RV. Third, an RV search operation based on the measured RA is carried out via the bisection algorithm, which optimizes a newly devised cost function. Compared with the conventional method, the proposed method has two main advantages: 1) it requires no information about the rotation center of a target, and 2) it can efficiently generate a well-scaled ISAR image within a very short time. Finally, the results of experiments using point scatterers and real flying aircraft are provided to demonstrate the validity of the proposed method.

The bistatic configuration used for inverse synthetic aperture radar (ISAR) imaging has attracted much attention because of its potential for overcoming certain monostatic ISAR imaging limitations. However, translational motion (TM) compensation (TMC) and rotational motion (RM) compensation (RMC) are the most fundamental requirements for both bistatic ISAR (Bi-ISAR) and monostatic ISAR imaging to obtain the high levels of image resolution desired. Moreover, range and crossrange scaling (RCRS) and bistatic distortion correction (BDC) are essential functionalities for the efficient use of Bi-ISAR images in applications. In this paper, a novel approach to sparse aperture (SA) Bi-ISAR imaging is presented for simultaneously performing TMC, RMC, RCRS, and BDC on maneuvering targets based on compressive sensing. Instead of solving conventional optimization problems constrained by the sparsity of the signals, the proposed method utilizes sensing matrix estimation technique for Bi-ISAR image reconstruction using parametric signal model reconstruction by looking for the basis functions that best represent the behavior of sensing dictionary matrix comprised of the observed SA data. The reconstruction of the sensing matrix is based on a modified orthogonal matching pursuit (MOMP)-type basis function-searching scheme. Finally, we generate a focused and scaled Bi-ISAR image from the complete Bi-ISAR signal recovered using the proposed method. Several simulation results reveals that the proposed method is very efficient in forming Bi- ISAR images of high-speed maneuvering targets in terms of the Bi-ISAR signal reconstruction accuracy

We propose a new teacherstudent framework (TSF)-based knowledge transfer method, in which knowledge in the form of dense flow across layers is distilled from a pre-trained “teacher” deep neural network (DNN) and transferred to another “student” DNN. In the case of distilled knowledge, multiple overlapped flow-based items of information from the pre-trained teacher DNN are densely extracted across layers. Transference of the densely extracted teacher information is then achieved in the TSF using repetitive sequential training from bottom to top between the teacher and student DNN models. In other words, to efficiently transmit extracted useful teacher information to the student DNN, we perform bottom-up step-by-step transfer of densely distilled knowledge. The performance of the proposed method in terms of image classification accuracy and fast optimization is compared with those of existing TSF-based knowledge transfer methods for application to reliable image datasets, including CIFAR-10, CIFAR-100, MNIST, and SVHN. When the dense flow-based sequential knowledge transfer scheme is employed in the TSF, the trained student ResNet more accurately reflects the rich information of the pre-trained teacher ResNet and exhibits superior accuracy to the existing TSF-based knowledge transfer methods for all benchmark datasets considered in this study.

This letter presents the design for a low-profile planar inverted-F antenna (PIFA) that can be stuck to metallic objects to create a passive radio frequency identification (RFID) tag in the UHF band. The designed PIFA, which uses a dielectric substrate for the antenna, consists of a U-slot patch for size reduction, several shorting pins, and a coplanar waveguide feeding structure to easily integrate with an RFID chip. The impedance bandwidth and maximum gain of the tag antenna are about 0.3% at 914 MHz for a voltage standing wave ratio (VSWR) of less than 2 and 3.6 dBi, respectively. The maximum read range is about 4.5 m as long as the tag antenna is on a metallic object.