HIDDEN OBJECT DETECTION WITH MICROWAVE AND MILLIMETER WAVE IMAGING SYSTEM

Abstract

Nowadays hidden object detection (HOD) system such as through-wall imaging (TWI) and millimeter wave (MMW) imaging radar is one of the most promising techniques which may be quite useful for the homeland security as well as defense applications because of its strong penetrating capability through opaque obstacles, high range resolution, and good resolving ability. It is one of the most rapidly emerging technologies where the aim is to ‘see’ through visually opaque obstacles like different types of walls and cloths and detect as well identify various targets present behind the wall or any opaque object such as cloths and book. The detection of objects inside a closed room and concealed targets under various cloths and book from standoff distance attracts more attention of researchers due to protection as well as the safety of humans and their assets. Hidden objects detection and identification involve inter-disciplinary research mainly electromagnetic propagation, antenna and waveform design, imaging, signal and image processing and detection techniques. The electromagnetic spectrum plays a major role in the field of active hidden object detection (HOD). Several researchers have demonstrated the capability of different imaging systems like gradiometer metal detectors, microwave imaging, THz imaging, infrared imaging, X- rays imaging and MMW imaging for hidden object detection with identification and carried out the comparison on the basis of the image quality, spatial resolution, atmospheric loss, and penetrating capability through obstacle materials. These techniques have their own pros and cons. The microwave frequency range especially 1-4 GHz has the better capability to penetrate the thick brick wall, but detection of small size concealed targets like blade, paper-cutter, and matchbox inside cloths and book create a problem because of constrained of spatial resolution in the microwave range. The THz and infrared imaging techniques attract more attention of researchers due to high spatial resolution and less side effect to the human body at the cost less standoff distance. The X-ray scanner used to detect concealed weapon but cannot scan human body because it is harmful to human body. The MMW imaging system can penetrate through cloths and book with very less attenuation and able to acquire the image of small size concealed targets such as, blade, knife, paper-cutter, and matchbox but not able to penetrate the thick obstacles like different types of walls. The detection and identification capability of hidden object mainly depends on two factors, one is penetration power and other is the resolution of the imaging system. These two factors are inversely related to each other. There is a tradeoff between penetration capability and resolution, the fine resolution is obtained at a higher frequency on the cost of penetration capability and vice - versa. Therefore, it can be inferred that the single frequency band cannot vi be utilized for all types of hidden object detection. So, in this thesis, we have used frequency ranges 1.5 - 3.5 GHz, check for through the wall detection, and frequency 53 to 65 GHz for Millimeter wave (MMW) imaging for targets shape recognition when targets are placed inside the human apparel or book because of its better resolution capability. HOD system can be based on either an active or passive technique. In active mode, the imaging system consists of transmitter and receiver, the transmitter illuminates the target to be screened and the receiver receives the scattered signal for further analysis. Therefore, this thesis focuses on two major tasks (i) design and test a compact TWI system for frequency range of 1.5 GHz to 3.5 GHz and (ii) explore the possibility to use MMW for detection and identification of the small targets like blade, matchbox, cigarette, and bookmark beneath the cloth and inside the book. In a typical active HOD system, an antenna is used to transmit the signal, which penetrates through the cloths, book, and wall. For HOD imaging application, it requires an antenna with compact size, high gain, and broad bandwidth. A good down range resolution can be obtained with the help of broad bandwidth for which a broadband antenna is required and at the same time lower frequency is indeed required to minimize through wall propagation losses. The size of the antenna depends on the operating frequency. The effective compromise in the choice of the broad bandwidth must be taken for designing the compact size antenna at a lower frequency. Therefore, at a lower frequency, particularly for TWI application, the compact size with wideband and high gain antenna design is a very challenging task. The antenna array may be one of the alternate solutions for the image the target. Several types of antenna array have been used for microwave radar imaging application but a major effort is needed to be paid in this direction for a simple, compact and cost-effective TWI antenna array design with good isolation among each element of the array. Several detection enhancement techniques exist in literature but still, efforts are required to develop an adaptive detection method with high accuracy. Furthermore, target’s shape identification is one of the most challenging tasks in case of using coordination invariant of hidden targets due to resolution limitation of the imaging system. To simplify reliable recognition of the tilted target shape, researchers have reported several invariant feature extraction techniques such as wavelet transforms, Hough transforms, Harris corner detector and Scale Invariant Feature Transform (SIFT). However, these techniques are not fully flexible and still extensive study is required to develop an adaptive method which can be applied effectively to reform the images of tilted hidden objects. Based on the observations, following objectives were framed; (i) Development of low-cost semi-prototype through-wall imaging system at frequency range of 1.5 GHz-3.5 GHz and vii evaluate its performance (ii) Millimeter-wave active imaging with neural network algorithm for concealed targets identification (iii) Development of efficient approach for MMW imaging system to identify targets inside the book for security applications, (iv) Development of adaptive approach for identification of targets (matchbox, pocket diary and cigarette box) under the cloth with MMW imaging system. The thesis consists of seven chapters. Chapter 1 presents the introduction which consists of motivation, scope, and objectives of the thesis. Chapter 2 provides a brief literature review for the considered objectives, their limitations, and research gaps. Chapter 3 presents the design of compact TWI system for targets detection and identification behind the brick wall at frequency 1.5 GHz to 3.5 GHz. An attempt has been made to assemble a TWI system which has several components such as antenna array, connecting switch, vector network analyzer (VNA) and a display device. The performance of the device was evaluated while detecting and identifying various targets behind the wall. The components of developed TWI system consist of five indigenously fabricated Vivaldi antenna, in which one is used to transmit the signal and remaining four is used to receive the reflected signal. The spacing between the antennas, which is 5 cm apart of each other, has been optimized in computer simulation technology (CST) simulation tool for the accurate performance of the TWI system. The interference reduction between transmitting and receiving antenna is one of the necessary steps which provides the focused image of the target. Therefore, the isolation between transmitting antenna and receiving antenna has been preserved by coating absorber material on the aluminum sheet, which is placed on both sides of transmitting antenna. The complete system has been enclosed inside the box, which is made of acrylic material due to its high transmittance (87.9-92.6%). The delay-sum algorithm is used for imaging. Developed TWI is based on stepped frequency continuous wave (SFCW) radar principle and worked in the frequency range 1.5 GHz to 3.5 GHz, due to use of stepped frequency ultra-wideband signal in radar, we can get a high range resolution profile of target by which it may possible to separate targets in downrange at 7.5 cm. With the help of the developed TWI system, the living and non-living targets behind brick wall have been detected. The shape of targets such as circular, rectangular, triangular metal sheet as well as tiles has been satisfactorily imaged behind the wall. It is found that for all tested algorithm, the developed TWI system works quite satisfactorily. To detect the small concealed targets like a gun, knife, blade, Pocket diary, cigarette box and matchbox is another challenging and interesting task, which may be quite useful for viii the various application. As the target size is very small, therefore, use of microwave frequency band because of its resolution limitation, may not be suitable to detect and identify these types of targets. Therefore, in the rest chapters 4, 5 and 6, the MMW imaging technique has been used because of its good resolution capability as well as moderate penetrating power for detection of considered small targets. Chapter 4 aims to identify the shape of different small concealed targets (beneath woolen and cotton cloth) like a gun, knife, blade, and matchbox. For this purpose, a methodology is developed with the application of various image processing techniques and artificial neural network (ANN) where the special emphasis has been given to minimize the orientation effect of the target for shape identification. To achieve this, a stepped frequency continuous wave (SFCW) active imaging radar system was ingeniously designed at our Microwave Imaging and Space Technology Applications Laboratory (MISTAL), IIT Roorkee, India using vector network analyzer (VNA) operating in 59 GHz to 61 GHz frequency band, which has range resolution 7.5 cm. For complete target data acquisition, C-scan (horizontal and vertical scanning) methodology was used, which gives us target information in terms of length and width. It is observed that the raw C-scan image is not having good capability to identify these targets because of clutter, noise, missed out pixel points during scanning. Therefore, a critical analysis of commonly used image processing techniques i.e., clutter removal, image enhancement, thresholding and pattern recognition have been carried out for selecting the suitable image processing techniques for said objective and their performance has been evaluated on the basis of peak signal to noise ratio (PSNR). After critical analysis it is found that Spatial Gaussian filter is well suited for clutter removal for the present case, in comparison to other techniques like; singular value decomposition (SVD), independent component analysis (ICA), factor analysis (FA), principle component analysis (PCA), average trace subtraction (ATS). After removing the clutter, various thresholding has been applied but this technique is not providing the good results due to wide contrast variation among the targets. Therefore, image segmentation technique has been proposed for application of thresholding techniques on separated region. Till now, we have not considered rotational effect for targets detection under various cloths. Accurate shape identification is one of the challenging problems for tiltted targets under cloths. Therefore, Hough transform and artificial neural network (ANN) have been used for scale and coordination invariant image reconstruction. The Hough transform may be quite useful for regular shape targets but it is observed that for irregular shape targets, this transform is not providing the results as expected because it finds the angle of rotation by searching the maximum value of collinear points in ix the straight line of the edge detected in the image. Hence, pattern recognition based ANN algorithm has been proposed to identify all types of considered targets, which has the capability to minimize any rotational variations effect. Our prime task is to identify the concealed targets placed at any angle. To resolve such problem, three targets such as a gun, knife, and matchbox of two different sizes have been taken and rotated between 00 to 3600 at every 300 under two dissimilar covering cloth such as woolen and cotton. Data set of 150 samples were made with twelve dissimilar rotation directions (00 to 3600) for the three targets under two dissimilar covering cloth (Cotton and Woolen) and six without targets. In consideration of target identification, 135 randomly data (90%) out of entire 150 samples data have been selected to train ANN model and for testing and validation point of view of the trained neural network, the remaining 15 data (10%) has been considered. A multilayer feed-forward neural network (MFFNN) is used which consists of pattern recognition network. For all considered targets, a pattern has been generated on the basis of the image matrix, which is defined as the signature of targets. The first step after taking the independent readings, the signature of the reference targets are matched with the signature of the obtained image matrix with the help of the correlation coefficient. The proposed algorithm has also been evaluated for different types of scenario i.e., blade and same size of paperboard placed beneath the cloth and the satisfactory result have been obtained, which shows that the proposed technique has a good capability to identify the considered concealed targets inside the cloth. The targets like a blade, bookmark, and paper cutter are generally placed inside the book, which detection is again quite challenging and it needs a good range resolution. The range resolution of imaging system should be lower than the thickness of book for perfect detection of concealed targets at any position inside the book. The range resolution of the active imaging system depends on the operating bandwidth. Therefore, Chapter 5 proposes statistics based efficient technique for identification targets like, blade, knife, and plastic bookmark, which is placed inside the book for MMW imaging system for the frequency range from 53 to 65 GHz. We have considered 12.0 GHz bandwidth, at this bandwidth, the range resolution becomes 1.25 cm. So it will be beneficial for detecting the targets inside the book, whereas book thickness is more than 1.25 cm. The targets have been concealed inside the book at a different position and orientation and observation were taken from a standoff distance of 100 cm the imaging system. The dimension of a book, blade, knife, and plastic bookmark are (18X16X1.5) cm, (4.5X2.5) cm, (7X3) cm and (10X2.5) cm respectively. The C- Scan image was generated for complete target's data acquisition. A time-gating technique is used for clutter reduction from the system, antenna- air interference and background. The reflection from a x book has a similar nature over each range bin after applying time gating process. A wide intensity variation has been observed of particular range bin where the target is placed inside a book. This observation gives an idea to find the variance of an intensity level of each range bin image. Entropy is another feature that can be used to decide the target inside a book. Entropy is a measurement of the degree of variation in a data set. Wide variation in data set provides a higher value of entropy and vice versa. For finding the Entropy values, firstly, the histogram of each range bin image is calculated after that probability density function can be assimilated by normalizing the histogram. For target identification, standard deviation (std) and Entropy of each range bin are calculated and find that the maximum value of std and entropy have larger if targets present inside the book. The less variation has been observed in std value for a book only (i.e., no target inside the book) which entropy is almost zero value. On the other hand, if targets present inside a book, large variation in std value has been observed and entropy has also large value. Hence on the basis of std and entropy, with the help of curve fitting approach, the decision has been taken to present of targets inside the book. The values of std and entropy are helpful to identify which type of targets present inside the book. It is observed that the proposed algorithm has good capability to detect and identify the considered targets inside the book. Matchbox, Pocket diary, and Cigarette box look quite a similar when it is in the pocket. Therefore, the detection and identification of these targets from a standoff distance may be quite useful for the various agencies. Matchbox, Pocket diary may have a very low reflection because these are non-metallic whereas Cigarette box provides the good reflection due to the presence of polythene cover and aluminum foil, outside and inside of cigarette box. So detection of these targets is again a difficult task. Therefore, in Chapter 6, firstly, we have attempted to detect these targets and secondly its identification. For this purpose an adaptive statistics based algorithm has been proposed which have the capability to detect the target and shape the considered targets. After that a probability density function (pdf) based decision tree algorithm has been proposed which is capable to distinguish these targets. Finally, the proposed algorithm has been effectively verified with the help of numerous independent data sets and a quite satisfactory results were obtained. Chapter 7 presents the summary of the thesis with concluding statement as well as the scope of future work. The indigenously developed compact, portable and semi-prototype through-wall imaging system has been used for living as well as non-living targets detection and identification and proposed adaptive MMW signal processing algorithms for robust target shape detection as well as target identification on the basis of statistics. The developed techniques can be used for hidden object detection and xi identification for search and rescue operations after a natural disaster like an earthquake or to detect & neutralize enemy targets inside a building as well as the security of the public and their assets, like the airport, shopping mall, and playground stadium etc. The developed techniques may have good potential for use in different sensors (active and passive, different frequency spectrum) and can be explored further.