In this paper, we propose a transfer learning approach with Convolutional Neural Networks (CNN) for radar Automatic Target Recognition (ATR). Radar echo signals of moving targets introduce micro-Doppler signatures that can be analyzed using spectrograms. Spectrograms can show distinctive micro-Doppler signatures of different targets, and they are used in our approach as inputs. We are utilizing a pre-trained CNN as a feature extractor in which feature maps can be extracted from any of the layers to train a classifier. AlexNet was used as the pre-trained CNN and softmax was used as the classifier. Our approach was tested on RadEch database of 8 ground moving target classes. Our approach outperformed the state of the art methods, using the same database, and reached an accuracy of 99.9%.
In this paper, a deep learning approach for Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) is proposed. The novelty of the proposed framework stems from the fact that it is based on a transfer learning scheme, where a pre-trained Convolutional Neural Network (CNN) is employed to extract learned features in combination with a classical Support Vector Machine (SVM) for classification. The efficiency of the presented approach is validated on the MSTAR dataset, where ten target classes are used. A classification accuracy of 99.27% is achieved
This paper presents a system that uses a thermal image as an input in order to detect and distinguish man-made objects, humans, animals, military targets or any other objects of prominent heat profile. The object's classification should be implemented based on the shape analysis of the object, and how this shape is changing over the time. Many techniques were provided in the literature for detection, description and identification of shapes. Therefore, it is very important to verify those techniques and select the applicable one for thermal shape processing. Moreover, there is a need to study pre-processing of thermal images to identify the most useful algorithm for shape detection. After completion of this project, the future work can be done on some specific applications, where the image that includes, for example, animals will be captured and the remaining images will be ignored (such as in visual surveillance of wild life at night and recognition of different types of animals).
This paper presents the design of a wideband 2-6 GHz phased array transmitter using Rotman lens. This transmitter can be used as a Radar system transmitter. The design is implemented and analysed using SystemVue simulation tool, where different components of the transmitter are designed according to required specifications.
Phase-locked loops (PLLs) are widely used in the synchronization of grid connected single-phase power electronic-based equipments due to their ease of digital implementation and satisfactory response. Generally in PLLs, the estimated frequency undergoes a large transient when a phase angle jump occurs. To resolve this problem, a frequency-locked loop (FLL) is proposed. In the proposed approach, frequency and phase estimation algorithms are developed such that the phase estimation loop is external to the frequency estimation loop, therefore, contrary to PLLs, the proposed FLL minimizes the coupling that exists between frequency and phase. This reduces the large transient experienced in the estimated frequency when a phase angle jump occurs. Simulation result is presented to demonstrate the effectiveness of the proposed approach.
Recent researches in cognitive visual analysis of crowd have indicated that crowd modeling is conventionally based on density analysis. However, socio-cognitive behavior studies have demonstrated that crowds often display a wide variety of behaviors that arise spontaneously from the collective motions of unconnected individuals. Therefore, behavior analysis employing physics-based approaches only, thereby neglecting the socio-psychological aspects, may present diverse challenges to accurate inference. This means that by identifying and modeling some of the interacting agents that underpin the evolution of such behaviors, we can deliver contexts that can help in the autonomous analysis of social and antisocial behaviors in crowded environments. This paper discusses these issues from the machine vision perspective. In particular, socio-cognitive models of crowds are linked to low-level mechanisms of crowd modeling and features extraction. A survey of recent works on crowd behavior analysis is conducted under a proposed behavioral categorization based on the level of the performed analysis and identified behaviors.
Business Process (BP) is a set of activities performed to achieve specific organizational goals. The discipline that governs BPs is called Business Process Management (BPM), it includes different method for the modeling, executing and monitoring BPs. Typically, Business Rules (BR) are used to enforce regulations and policies on the BP. In this paper, we discuss a simple language for the generation of BR directly from BPMN models based on a fragment of First-Order Logic (FOL). The rules are based on control-flow aspects of the BPMN that is divided into a set of components. To automate the language, A JAVA-based tool to generate the FOL rules is implemented.
Fast search engines are required for real-time decision making in various fields including computer vision, machine learning and object recognition. In the case of Internet of things (IoT) devices that need to implement fast search engines, it is of paramount importance to keep both area and energy costs at minimal. Conventional CMOS-based search engines suffer from density and power limitations. In this paper, we propose a reconfigurable memristor-based XNOR CAM cell (VR-CAM) that can support Binary-CAM (BCAM), Ternary-CAM (TCAM) and Approximate search. The proposed cell accepts the input search data as voltage and the database as memristor value (resistor). Whereas the output result in the form of voltage for TCAM and BCAM and resistance for approximate search. Based on this cell, a memristor-based CAM architecture is proposed. The architecture is composed of multiple banks of specially connected memristors for each bit. Simulations of the proposed architecture for search functionalities were carried out using LTSpice circuit simulator. The proposed CAM architectures achieves a 1-ns search cycle time. It utilizes 2 memristor devices per cell with option for stateful output.
This paper addresses the connectivity issues in urban Vehicular Ad-Hoc Network (VANET) using the cluster-based QoS-OLSR protocol. VANET in urban suffers from frequent topology changes and disconnections due to several urban characteristics such as intersections, obstacles and traffic lights. Therefore, network connectivity is an important metric for routing in urban. The proposed protocol is an extension of the existing cluster-based QoS-OLSR which connects only 2-hop away heads through an MPR. The enhanced protocol connects 2-hop away heads as well as 3-hop away heads which increase the network connectivity. The proposed protocol outperformed the existing protocol in terms of packet delivery ratio and end-to-end delay.
In this paper, a technique that explores a new scheme that is suitable for the available hardware requirements when modeling and linearizing the power amplifier (PA) with wide-band signals is proposed. The aim of the proposed technique is to relax the sampling rate requirements of the analog-to-digital convertors (ADC) in the digital pre-distortion (DPD) systems. The proposed resampling approach is based on under-sampling the output of the PA that results in a low-rate ADC, and efficiently restoring the under-sampled signal for DPD model extraction. The validity of the method is evaluated using a 20-MHz LTE input signal, and it shows excellent potential in restoring the full-rate signal from the under-sampled.
