ABSTRACT. Various diseases are relevant to the abnormal blood flow in tissue. Diffuse correlation spectroscopy (DCS) is an emerging technology to extract the blood flow index (BFI) from light electric field temporal autocorrelation data. To account for tissue heterogeneity and irregular geometry, we proposed an innovative DCS algorithm (NL algorithm) previously, in which the DCS signals are fully utilized through iterative linear regression. Under the framework of NL algorithm, the BFI to be extracted is significantly influenced by the linear regression approach adopted. In this study, a total of three approaches to implement the iterative linear regression, i.e., least-squared minimization (L2 norm), least-one minimization (L1 norm) and support vector regression (SVR), were compared. Among these, L2 norm is a conventional approach to perform linear regression. L1 norm and SVR are the approaches newly introduced by us to process the DCS data. Computer simulations and the autocorrelation data collected from liquid phantom in rectangular volume were utilized to evaluate the three approaches. The results exhibit the best performance of SVR approach in extracting the BFI values, with error of 2.23% at 3.0 cm source-detector separation. The L1 norm generated a medium error of 2.81%. By contrast, L2 norm leads to the largest error (3.93%) in extracting the BFI values. The outcomes derived from this study will assist in improvement of the tissue blood flow measurements, which is critical for translation of the DCS technology to the clinic.

ABSTRACT. Already in the beginning of tomography functions were constructed that are invisible from two different directions. Later the functions invisible from a finite number of direction, building the null space of the Radon transform, were called ghosts, and the original statement of Smith, namely, A finite set of radiographs tells nothing at all, was the starting point of fruitful research. Here radiographs mean complete projections. Based on the characterization of the null space questions on the uncertainty and the possible resolution were analyzed. We give a brief historical sketch and present new results.

ABSTRACT. X-ray Computed Tomography (CT) imaging has been widely used in clinical diagnosis, non-destructive examination and public safety inspection. Restraining artefacts from various physical factors and pushing dose as low as reasonably achievable have been pursued consistently in decades in this area. With the great success of deep learning in computer vision field in this years, deep learning has also gained a lot of attention in X-ray imaging. It opens a bright and challenging future for CT data processing. We explored the possibility and capability of improving image quality of X-ray CT images with the deep learning technology by different strategies. Some of our initiative network designs especially designed for X-ray CT artefact removing and CT reconstruction will be presented in this talk, as well as very encouraging experimental results from simulation and practical systems. Finally, we give a discussion on the technics and difficulties in incorporating deep learning seamlessly into X-ray CT imaging.

ABSTRACT. Microwave nondestructive testing and the sensors required to affect the testing are less well known that other electromagnetic methods such as eddy currents techniques. Nevertheless, they occupy a critical role in testing as well as imaging of conditions and flaws in primarily nonmetallic structures including plastics, polymers and ceramics as well as wood structures, various composites and concrete. The methods and sensors range from microscopy to industrial scale testing and from simple open-ended waveguides to complex radars. The frequency range covers the whole microwave range and down to mm waves. The present talk discusses the use of microwave testing, its place in a nondestructive testing regime and a sampling of sensors that are representative of the art and its present state.

ABSTRACT. Classification of benign and malignant lung nodules using chest CT plays a key role in early detection of lung cancer. Although deep learning has achieved good results in natural image classification, it requires a large amount of training data, which is difficult to obtain a large amount of training data for routine medical image applications. In this paper, we proposes an adaptive feature selection algorithm based on transferred deep convolution neural network (DCNN) to distinguish benign and malignant lung nodules using limited chest CT data. Firstly, DCNN model pre-trained on ImageNet database was used to extract multiple features of multi-channel lung nodules images. Then, feature pooling and adaptive feature selection algorithm were used to further select sparse activation features. Finally, the 120-dimensional compact features were selected for benign and malignant lung nodule classification. We evaluate the proposed method on CT images from Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI). The experimental results show that the proposed algorithm does improve the classification performance, which can achieve the classification accuracy of 90.70% and the AUC of 0.94. In addition, we also tentatively explored the correlation between high level neurons in DCNN and lung nodule image features.

ABSTRACT. In this study, we developed a logical and complete mathematical analysis of the relationship between the discrete Fourier transform (DFT) magnitude and periodic feature of a signal. The physical meaning of the DFT magnitude index corresponds to the periodic signal feature; further analysis makes clear the relationship among DFT magnitude, magnitude index, and number of samples. The proposed analysis method also elucidates the relationship between alternating current (AC) magnitude index and frequency. The proposed analysis method is suitable for the machine vision and image processing in general.

ABSTRACT. Ground-based synthetic aperture radar (GBSAR) is a very effective means to perform high-repetition observation in local areas by using synthetic aperture radar (SAR) imaging principle. Conventional GBSAR imaging projects the scene image to the slant-range plane. In order to be able to invert the three-dimensional image of the observation area, the linear scan of the GBSAR system can be changed to a rotary scan. This paper proposes a three-dimensional imaging method for the rotary scanning system without making any assumptions about the target plane height. Firstly, the GBSAR signal is mapped to the three-dimensional wavenumber domain. Then the three-dimensional imaging is performed by the polar coordinate format algorithm.

ABSTRACT. In this paper, the characteristics of clustering model is first analyzed, then the support vector machine (SVM) is used to classify the data after clustering, and the classified results of the K means clustering model and space pyramid model using radial basis kernel and histogram intersection kernel are compare, the results show that the recognition rate of the space pyramid model with the histogram intersec-tion kernel can reach as high as 0.98.

ABSTRACT. The measurement and analysis of the spatial morphology and dynamic characteristics of the electric arc plasma inside low voltage circuit breaker is an effective way to improve its breaking performance. In this paper, a measurement system of switching electric arc inside low voltage circuit breaker based on is designed. A switching arc magnetic field measurement board based on Hall-effect was designed in this work to obtain the 2D distribution of the magnetic field around the arc during the arcing process, meanwhile the optical images of arc plasma corresponding to the moment is captured by a light sensor array and the waveforms of arc current and voltage are detected by hall tramsforms. A Multi-channel data acquisition system is developed to achieve high-speed synchronous measurement of arc current and arc voltage, arc shape and the magnetic field distribution around the arc plasma. The result show that the magnetic field distribution is consistent with the morphology of arc, which can be used to reveal the spatial morphology of arc plasma.

ABSTRACT. Composite material brings many challenges in structural health monitoring (SHM), especially in internal damage detecting. CFRP-OFBG, using Optical Fiber Bragg Grating (OFBG) sensors embedded in Carbon Fiber Reinforced Polymer (CFRP) composite structures, has been widely used in the field of structural reinforcement with smart sensing features. This work developed a real-time monitor system to detect internal damage by using dense arrayed fi-ber-optic sensor embedded in CFRP-OFBG. A classical triangulation proce-dure is selected and improved in damage location detection algorithm. Experimental results showed this design is an efficient and lightweight system in detecting internal damage for CFRP-OFBG materials.

ABSTRACT. In order to realize the detection of magnetic catalyst in fluidized bed, the research on using TMR (Tunneling Magneto Resistance) sensor instead of detection coil is conducted in this paper. The TMR sensor array is designed after the contrast test between TMR sensor and detection coil, the direction of TMR sensitive axis is determined by simulation experiment, then a new sensitivity matrix based on TMR is obtained by using perturbation method. Finally, the feasibility of detecting the distribution of magnetic catalyst in the new EMT system based on TMR sensor is verified by experiment.

ABSTRACT. This paper judged the goodness-of-fit of five commonly used distribution models with synthetic aperture radar (SAR) dataset collected by C-band Sentinel-1 radar off the Strait of Malacca. The purpose was to find out the optimal model that suits the data, then construct constant false alarm rate (CFAR) detector for vessel detection. The Kullback-Leibler (K-L) Distance was adopted to judge the fitting degree. The figure of merit (FOM), the probability of detection (PoD) and the false alarm rate (FAR) were calculated to evaluate detector’s performance.

ABSTRACT. As an efficient non-destructive checking technique, scanning acoustic mi-croscopy has been rapidly developed and widely used in nondestructive de-tection for composites, chip manufacturing, bio–pharmaceuticals and so on. Scanning acoustic microscopy can obtain the two-dimensional image and re-construct three-dimensional model of the internal structure inside the sample, compared with the optical microscope. Due to the geometrical features of high frequency transducer, the precision of focus has already become the key fac-tor for the digital imaging and defect detecting. Especially for the current 3D packaging chips and multi-layer composite materials, thickness of layer has reached the level of nm. Different focus plane will provide a completely dif-ferent digital image. However, the process of focusing has been relying on manual operation without great development until now. To improve the speed and accuracy of the focusing, we proposed a quick auto-focus method on scanning acoustic microscopy. This method can be divided into three steps. The first one is fast auto-focusing on upper surface, the second other one is auto-focusing on the interlayer base on curve, and the last one is to im-prove the A-Scan signal of points in the curve with spars signal repre-sentation. This method is integrated into Scanning Acoustic Microscopy imag-ing software and has been validated in the experiment with Olympus trans-ducers at 30\50MHz. It can be seen that this method can effectively improve the accuracy of focus and shorten the focus time of Scanning Acoustic Mi-croscopy through experiments.

ABSTRACT. It is commonly believed that more coils in electromagnetic tomography (EMT) sensor would obtain better performance of reconstructed images. In order to study the impact of number of coils in EMT sensors on quality of reconstructed images, five kinds of sensors with different number of coils including 4, 8, 12, 16 and 20, are involved to conduct simulations. The forward problem can be solved through implementing finite element method (FEM), then measurements and sensitivity matrix are obtained, which can be used to solve EMT inverse problem with proper image reconstruction algorithms. Five typical conductivity distributions are used to verify the performance of EMT sensors with different number of coils. The sensitivity matrices of different EMT sensors are analysed to further understand the essential reason of these simulation results through using singular value decomposition (SVD). EMT sensor with 16 coils gives the best image reconstruction results for most of the conductivity distributions. Limited improvement can be obtained in the quality of reconstructed images when the number of coil is more than 16.

ABSTRACT. With the rapid development of UAV technology and the continuous advancement of computer vision technology, target tracking has been widely used in various fields related to UAV. A quick and accurate evaluation of target tracking effect on UAV platform can provide effective support for the UAV mission planning and tracking strategy adjustment. However, nowadays the main target tracking effect evaluation methods cover too many aspects, and the reference indicators are complicated. Meanwhile, the real-time performances are poor. Therefore, this paper proposes a real-time evaluation method based on adaptive KCF to evaluate the target tracking effect of the UAV platform. Based on the tracking algorithm, the target tracking effect on UAV platform is analyzed in real time, and an evaluation system is built to evaluate the target tracking effect of the UAV platform, which can get accurate and stable evaluation results.

ABSTRACT. Video Quality Assessment (VQA) is an important and indispensable performance assessment index in many applications. A non-reference VQA metric for H.264 bit-stream is proposed in this paper. Firstly, the parameters, including I-frame and P-frame coding modes, ICT (Integer Cosine Transform) coefficients and Motion Vectors (MV), are extracted from the bit-stream. Then two indexes are defined to characterize the motion complexity and texture richness of the video content based on the statistical analysis of the parameters. The two indexes are combined together with Quantization Parameter (QP) to form the feature parameter set. Finally, Support Vector Regression (SVR) is exploited to establish a mapping model between the feature parameter set and the subjective Mean Opinion Scores MOS (MOS). The mapping model can be used to predict the corresponding MOS of H.264 bit-stream. Experimental results show that, compared with the state-of-art metrics, the proposed metric can obtain more accurate assessment results only using H.264 bit-stream parameters and have good accordance with human visual perception.

ABSTRACT. The group targets resolution of the VHF radar is limited by its bandwidth. By fusion the signals of multiple sub-bands can increase the resolution. To improve the coherent compensation, this paper makes two aspects based on the existing high-band fusion method. Firstly the Norm optimization is used to reconstruct the cost function to improve the registration accuracy. Secondly the sample data is improved to estimate the unknown frequency band more accurately in certain range. Simulations show that the band fusion method proposed by the VHF radar is better than the existing method in group targets resolution.

ABSTRACT. Location of the shallow subsurface vibration source involves many applications fileds. At present, most of the location methods for the shallow subsurface vibration source have large errors. It is not suitable for high accuracy positioning with shallow underground vibration. The SRP-PHAT is currently a method with high accuracy of sound source, but it is mainly used to locate sound sources in known media(e.g. air), such as microphone array. The SRP-PHAT can not be directly applied to source location in unknown shallow subsurface media. We propose a new SRP-PHAT method which is optimized by bat algorithm and combined with velocity scanning. The experimental results show that the proposed method can accurately locate the source and obtain the average velocity of wave propagation in shallow subsurface unknown media.

ABSTRACT. The SIFT key-point serves as an indispensable role in many computer vision applications. This paper presents an approximation of the SIFT scale space for key-point detection with high efficiency while preserving the accuracy. We build the scale space by repeated averaging filters to approximate the gaussian filters used in SIFT algorithm. The accuracy of the proposed method is guaranteed by that an image undergo repeated smoothing with an averaging filter is approximately equivalent to the smoothing with a specified gaussian filter, which can be proved by centre limit theorem. The efficiency is improved by using integral image to fast compute the averaging filtering. In addition, we also present a method to filter out unstable keypoints on the edge. Experimental results demonstrate the proposed method can generate high repeatable key-points quite close to the SIFT with only about one tenth of computational complexity of SIFT, and at concurrently the proposed method does outperform many other methods.

ABSTRACT. Background: Currently, the diagnosis of depression is largely based on clinical judgments due to the absence of objective biomarkers. There are increasing evidences that depression (DP) is associated with structural abnormalities. However, the previous analyses have a poor predictive power for individuals. Objective: To discriminate DP patients from normal controls (NCs) studied with structural magnetic resonance (MR) images using the method of support vector machine (SVM) combined with recursive feature elimination (RFE). Materials and Methods: In this study, 40 DP patients and 40 age- and sex-matched NCs were recruited from Guangzhou Brain Hospital and the local community, respectively. We employed ROI analyses based on the human Brainnetome Atlas, including 210 cortical and 36 subcortical subregions, to calculate gray matter volume (GMV) and white matter volume (WMV). The groups difference between DP patients and NCs were compared. The method of SVM combined with RFE was applied into the discriminative analysis of DP patients from NCs, in which discriminative features were drawn from GMV and WMV. Results: We found that the DP patients showed significant GMV reductions primarily involved in eight brain regions and showed significant WMV reductions primarily involved in ten brain regions. The classifier using GMV as input features achieved the best performance (an accuracy of 86.25%, a sensitivity of 85%, and a specificity of 87.5%) in the discriminative analyses between DP patients and NCs. Conclusions: These findings provided evidences that specific structural brain regions associated with DP patients might qualify as a potential biomarker for disease diagnosis, and the machine-learning method of SVM with RFE may reveal neurobiological mechanisms in distinguishing DP patients from NCs.

ABSTRACT. Compact polarimetric SAR is currently drawing more attention owing to its advantage to earth observations. In this paper, based on scattering vector in hybrid mode and X-Bragg scattering model, a new method is presented for evaluating ship detection performance. By using this method, three polarization features, including circular polarization ratio, relative phase and roundness, were analyzed selectively. Experiments performed using hybrid mode emulated from C-band RADARSAT-2 full polarimetric SAR data validate the feasibility of the method in analyzing the ship detection performance.