Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/14611
Title: INVESTIGATIONS ON MM-WAVE IMAGING RADAR FOR SHAPE AND FAULT DETECTION OF TARGETS
Authors: Agarwal, Smriti
Keywords: Millimeter Wave;Researchers Pertaining;Finer Resolution;Personal Screening
Issue Date: Jun-2015
Publisher: Dept. of Nanotechnology IIT Roorkee
Abstract: In recent years, millimeter wave (MMW) frequency (30GHz-300GHz) has received tremendous interest among researchers pertaining to its unique favorable features in contrast to the over flooded microwave frequency band, such as, finer resolution, higher data rate and size reduction. This makes it an ideal spectrum for applications, like, personal screening, medical imaging, quality monitoring, wireless communication etc. [183]. MMW frequency offers interesting applications in stand-off imaging because of its non-ionizing radiation in contrast to ionizing X-rays; its ability to penetrate through packaged materials in contrast to visible and IR systems; its higher resolution capability in contrast to microwave imaging. Thereby, MMW imaging has emerged as a preferable modality for finding fine details of the targets which may be useful to identify different target’s characteristics, like, crack, void, delamination, corrosion, moisture content & porosity etc. For all these reasons, MMW technique is becoming increasingly important in different industrial, scientific and military applications. Antenna is an indispensable component of any RF based system. At MMW frequency, design of a simple, compact and cost effective antenna is quite challenging because of metallic losses, dielectric losses, higher order/ surface wave modes loss, resonances as well as radiations in transmission line structures. Planar antenna technologies are worthwhile owing to its low profile, light weight, low cost, conformable and rigid characteristics [86]. Recently, the increasing demand of miniaturized and multi-tasking systems has excel the need for dual resonant MMW antennas that provide concurrency as well as redundancy [18]. However, dual band antenna at MMW frequency is still the least explored with few reported works using split ring resonator (SRR) [128], flip-chip assembly [285] and movable plate [184] techniques etc. Most of these reported antennas require sophisticated and high precision fabrication techniques, which make them complex and cost inefficient. Hence, major effort is needed to be paid in this direction for a simple, compact and cost-effective MMW dual band antenna design. Abstract ii MMW imaging is one of the fascinating and rapidly expanding area for object’s identification for security applications and for quality monitoring for industrial applications. Henceforth, critical investigation of digital image processing techniques in context to MMW imaging is currently becoming a new focused area of research. Various research works are carried out and going on for stand-off target’s detection and its shape identification using different imaging techniques [52, 337, 352]. However, more focus is still required in this direction to develop an efficient MMW imaging methodology for accurate, non-invasive target’s detection, shape identification and material classification. In addition, target’s recognition process can be made more robust by employing size and rotation invariant capability for correct target’s shape identification in a real scenario. To enable reliable recognition of target’s shape, the essential information in the form of unique, invariant features is extracted, like, scale invariant feature transform (SIFT), wavelet transforms, Harris corner detector, discrete Fourier descriptors etc. [215]. However, these algorithms are not fully adaptive and still significant research is required to develop a robust methodology which can be applied successfully to reconstruct target’s images towards varying orientation, scale and translation errors due to non-uniform illumination and target’s deformations. Non-invasive inspection of different goods towards fatigue, wear and tear without hampering its utility and efficacy is also constantly desired. Since, MMW has the capability to penetrate through opaque material and can extract target’s characteristic information, it provides a good solution towards non-invasive, easy and accurate quality inspection [144]. Quality monitoring of packaged goods (packaged ceramic tiles) for industrial applications is one of the fascinating domain, which needs special attention for a competent and cost efficient manufacturing. Thereby, a critical investigation of different image post-processing techniques towards their accurate crack detection capability requires special emphasis. In addition, a robust methodology is needed to be developed for an automatic, adaptive quality monitoring (i.e., fault detection) of packaged goods for industrial applications. As per aforementioned research gaps, following tasks have been carried out in this thesis work: (i) Design of a simple, cost-effective and compact MMW dual frequency planar antenna, (ii) Stand-off target’s shape identification and its material classification using MMW imaging system, (iii) Development of size and rotation invariant target’s shape identification algorithm for MMW imaging system, (iv) Development of an adaptive quality monitoring algorithm to detect fault (crack) in packaged ceramic tiles for industrial applications with MMW imaging Abstract iii system. This 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 objective(s), their limitations and research gaps. Chapter 3 explores the design for MMW concurrent dual frequency planar antenna. Designing of an antenna at MMW is a quite challenging task because of several limitations, which are needed to be undertaken, for an optimal antenna performance, like, choice of suitable substrate: since at MMW frequency, surface waves are more likely to be excited which reduces antenna radiation efficiency; fabrication constraint: commonly used techniques at lower microwave frequencies, viz., slot, notch, spur, multiple stacked patches, metamaterials etc. needs stringent fabrication and alignment accuracies (due to correspondingly low wavelength λ/2≈1.5 mm to 2.5 mm), so a relatively simpler technique is needed to be used without complicating the fabrication procedure or raising the cost. Therefore, in this chapter, a single layer, simple, compact and conformable antenna structure has been proposed operating concurrently at the two MMW frequencies 60 GHz and 85 GHz, which are having future commercial, strategic and industrial applications. The technique investigated for concurrent MMW antenna operation is stub loaded rectangular patch because of its simple and feasible fabrication. The proposed MMW antenna design was firstly simulated and analyzed using 3D full wave EM solver HFSS on Rogers RT5880 substrate (ɛr = 2.2, h = 5 mil (0.127 mm), tanδ = 0.002). Further, fabrication and measurement of MMW antenna prototype was performed and measured results were found in close approximation with that of simulated ones, giving fractional bandwidth 1% /6.4 % and measured gain 8.95 dBi/5.37 dBi, at 60/85 GHz, respectively. At MMW, in order to maintain measurement compatibility with GSG probe, a 50Ω coplanar waveguide (CPW) feedline and a wideband CPW to microstrip transition structure were separately designed and later on integrated into the antenna structure in order to couple maximum power from CPW feed to the radiating patch. The designed antenna structure is simple, cost-effective and compact having 3.7 mm2 cross-sectional area. Chapter 4 explores the capability of MMW frequency towards standoff target’s shape identification and its material classification. For this, a stepped frequency continuous wave (SFCW) active imaging radar system was ingeniously designed at our MISTA lab, IIT Roorkee, India using vector network analyzer (VNA) operating in V band (60 GHz) [6]. For complete target’s data acquisition, C-scan (horizontal and vertical scanning) methodology was used, which gives us target’s information in terms of length and width. The data acquired undergoes several Abstract iv signal pre-processing steps, such as, frequency to time domain (IFFT) conversion, time to spatial domain conversion, relative calibration and windowing for generating C-scan images. This C-scan image is known as raw C-scan image. The main aim of this chapter is to identify the shape and classify the material of the targets. So, raw C-scan image undergoes several post-processing steps, viz, clutter removal, thresholding, edge detection and classification. One of the major problem for this type of radar images is the clutter. To remove clutter, we have applied singular value decomposition (SVD) technique, which has the advantage of improved image quality, compressibility and is better in preserving edge details as compared to other techniques, like, Principle Component Analysis (PCA), Factor analysis (FA) and Independent Component Analysis (ICA) [46]. After removing the clutter, thresholding based image segmentation was carried out because of its simplicity of implementation. The image statistics based mean and standard deviation global thresholding has been used. Edges are useful features for segmentation and object identification in images, which characterize object boundaries. Criterion relevant to edge detector’s optimum performance is good detection, good localization and single response. The common edge detection algorithms were critically analyzed and it was found that canny based edge detector is better than the other techniques, like, sobel, prewitt, roberts, laplace [210]. Therefore, canny based edge detector has been used after thresholding to identify the shape of the respective targets, i.e., triangle, square, circle, rectangle. Various models, like, estimation of dielectric is one of the way to identify and classify the target’s material. But in this work an image analysis based technique, i.e., probability density function (pdf) based approach has been proposed to classify the target’s materials (we have considered wood and metal only). It has been observed that this pdf based approach has a good potential to classify the target’s materials. Chapter 5, proposes a novel artificial neural network (ANN) based scale and rotation invariant image reconstruction model. Neural network has been used as an effective signal processing technique for classification/recognition of targets for various applications, like, speech recognition, character recognition, pattern recognition etc. Correct shape identification is one of the challenging problems because it suffers from orientation and size deformations in any practical environment. In chapter 4, while target’s shape identification; we have not considered orientation and size effect. Whereas, in this chapter we have proposed a pattern recognition based neural network algorithm to identify target’s shape, which has the capability to minimize any orientation and size variations. For algorithm development, we have used four different regular target’s shapes (i.e., rectangle, square, triangle, and circle). A single shape has been considered Abstract v in different sizes and is put into different orientation angles (randomly ranging from 0° to 90°), thereby, experimental data set of total 33 samples of C-scan data is captured using the MMW imaging system having varying combinations of orientations and sizes of different considered regular shaped targets. As we want to recognize shape, the output training data was generated in accordance to the input target’s size and was in binary (0, 1) matrix form. For image recognition problem formulation, a multilayer feed-forward neural network has been used. Any independent target’s data when fed as the input to the ANN model, the output generated will be the correct target’s shape and size irrespective of its orientation. The proposed methodology, thus, has the capability to identify the target’s shape by minimizing size and orientation error. Chapter 6 deals in non-invasive undercover fault detection methodology for industrial quality control applications using the ingeniously designed MMW radar system. Accurate monitoring and classification of packaged ceramic tiles (cracked/non-cracked) is quite important, which could otherwise result in financial/ repute loss to the company, besides it is particularly challenging because of non-uniform illumination, alignment errors and insufficient contrast in practical industrial scenario. Our main purpose is the fault detection, and here, we have considered crack as a fault. Therefore, large number of packaged ceramic tile targets with different crack and non-crack configurations were used for algorithm development and its validation. For this purpose, two approaches have been proposed. First: feature extraction based classification approach. Second: Spatial statistical based approach. In the first approach, critical investigation of different commonly used feature extraction techniques has been done for accurate model development with minimum false alarm. For this, an automatic feature based neural network classifier model has been proposed by investigating five different commonly used feature extraction techniques, i.e., discrete fourier transform (DFT), Daubechies wavelet transform (WT), texture features, principal component analysis (PCA) features and histogram of oriented gradient (HOG) based features. Out of these, the optimal classification accuracy has been achieved by HOG-NN classifier model because HOG feature has an advantage of developing the training sample on a cell/ window basis, which has more information than the whole tile based feature techniques. This has been validated using an independent set of packaged ceramic tiles. It is a quantitative approach, which tells whether the packaged tile is cracked or non-cracked. Further, to visualize the crack location, we have also developed a spatial statistics based model, which is based on image reconstruction approach and provides information about location of the crack also. The statistical parameters (maxima, minima, median, and standard Abstract vi deviation) of target’s image play an important role in determining particular target’s characteristics, and also vary largely even for similar type of targets owing to non-uniform illumination, clutter, different multipath reflections, etc. Hence, it is quite essential to develop a statistics based generalized, adaptive crack classification and localization algorithm to make it more robust and adjustable as per frequent changes in image parameters in real scenario. Since, crack could be anywhere on the ceramic tile and of any random type, a window based statistical pattern search model has been proposed to scan the complete tile sequentially and to detect cracked locations, apart from just the crack/ non-crack classification. A mathematical formulation has also been developed for achieving user defined performance goals of accuracy and false alarm and an optimum threshold value is searched through GA optimization for crack/ non-crack location estimation. Here, while algorithm development main emphasis was to achieve near to zero false alarm for non-cracked full packaged ceramic tiles so as to avoid any unwanted loss and to further improve industry quality production. The developed proposed algorithm was also validated through independent packaged ceramic tile targets and was found to perform quite well. In chapter 7, the contributions made in the thesis are summarized and scope of future work is outlined. The proposed MMW robust target’s shape identification and adaptive quality monitoring model for ceramic tiles will be certainly helpful to make a complete automated system for industry.
URI: http://hdl.handle.net/123456789/14611
Research Supervisor/ Guide: Singh, Dharmendra
metadata.dc.type: Thesis
Appears in Collections:DOCTORAL THESES (Nano tech)

Files in This Item:
File Description SizeFormat 
G24523_Smriti_-T.pdf5.21 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.