Search

Objective.Sparse-view computed tomography (SVCT), which can reduce the radiation doses administered to patients and hasten data acquisition, has become an area of particular interest to researchers. Most existing deep learning-based image reconstruction methods are based on convolutional neural networks (CNNs). Due to the locality of convolution and continuous sampling operations, existing approaches cannot fully model global context feature dependencies, which makes the CNN-based approaches less efficient in modeling the computed tomography (CT) images with various structural information. Approach. To overcome the above challenges, this paper develops a novel multi-domain optimization network based on convolution and swin transformer (MDST). MDST uses swin transformer block as the main building block in both projection (residual) domain and image (residual) domain sub-networks, which models global and local features of the projections and reconstructed images. MDST consists of two modules for initial reconstruction and residual-assisted reconstruction, respectively. The sparse sinogram is first expanded in the initial reconstruction module with a projection domain sub-network. Then, the sparse-view artifacts are effectively suppressed by an image domain sub-network. Finally, the residual assisted reconstruction module to correct the inconsistency of the initial reconstruction, further preserving image details. Main results. Extensive experiments on CT lymph node datasets and real walnut datasets show that MDST can effectively alleviate the loss of fine details caused by information attenuation and improve the reconstruction quality of medical images. Significance. MDST network is robust and can effectively reconstruct images with different noise level projections. Different from the current prevalent CNN-based networks, MDST uses transformer as the main backbone, which proves the potential of transformer in SVCT reconstruction.

Pavement crack assessment is an important indicator for evaluating road health. However, due to the dark color of the asphalt pavement and the texture characteristics of the pavement, current asphalt pavement crack detection technology cannot meet the requirements of accuracy and efficiency. In this paper, we propose an end-to-end multi-scale full convolutional neural network to achieve the semantic segmentation of cracks in road images by learning the crack characteristics in the complex fine grain background of asphalt pavement. The method uses DenseNet and deconvolution network framework to achieve pixel-level detection and fuses features learned from different scales of convolutional kernels through a full convolutional network to obtain richer information on multi-scale features, allowing more detailed representation of crack features in high-resolution images. And the back end joins the SVM classifier to achieve crack classification after crack segmentation. Then we create a road test standard data set containing 12 cracks and evaluate it on the data. The experimental results show that the method achieves good segmentation effect for 12 types of cracks, and the crack segmentation for asphalt pavement is better than the most advanced methods.

Convolutional neural networks have achieved remarkable results in the detection of X-ray luggage contraband. However, with an increase in contraband classes and substantial artificial transformation, the offline network training method has been unable to accurately detect the rapidly growing new classes of contraband. The current model cannot incrementally learn the newly appearing classes in real time without retraining the model. When the quantity of different types of contraband is not evenly distributed in the real-time detection process, the convolution neural network that is optimized by the gradient descent method will produce catastrophic forgetting, which means learning new knowledge and forgetting old knowledge, and the detection effect on the old classes will suddenly decline. To overcome this problem, this paper proposes an incremental learning method for online continuous learning of models and incrementally learns and detects new classes in the absence of old classes in the new classes. First, we perform parameter compression on the original network by distillation to ensure stable identification of the old classes. Second, the area proposal subnetwork and object detection subnetwork are incrementally learned to obtain the recognition ability of the new classes. In addition, this paper designs a new loss function, which causes the network to avoid catastrophic forgetting and stably detect the object of the new contraband classes. To reliably verify the model, this paper produces a multi-angle dataset for security perspective images. A total of 10 classes of contraband are tested, and the interference between two object detections is analyzed by model parameters. The experimental results show that the model can stably perform new contraband object learning even when there is an uneven distribution of data types.

The sparse view problem of image reconstruction encountered in computed tomography (CT) is an important research issue due to its considerable potential in lowering radiation dose. Among the researches, the total variation (TV) method is especially effective in sparse view CT reconstruction for its good ability to preserve sharp edges and suppress noise. However, TV-based methods often produce undesired staircase artifacts in smooth regions of the reconstructed images since the reconstructed problem is usually ill-posed and TV regularization favors piecewise constant functions. Moreover, the image can be accurately approximated by sparse coefficients under a proper wavelet tight frame, which has good capability of sparsely estimating the piecewise smooth functions and the quality of reconstructed image can be improved by the sparse prior information. To deal with sparse view CT reconstruction problem, a minimization hybrid reconstruction model that incorporates TV with the wavelet frame has been proposed, which is to use the TV-norm of the low-frequency wavelet frame coefficients and the ℓ0-norm of the high-frequency wavelet frame coefficients to eliminate staircase effect while maintaining sharp edges, simultaneously provide enough regularization in smooth regions. In addition, considering that the two regularization terms produce more parameters, an alternating direction method of multipliers (ADMM) algorithm has been applied to solve the minimization problem by iteratively minimization separately. Finally, compared with several iterative reconstruction methods, the experimental results demonstrate the competitiveness of the proposed method in terms of preserving edges, suppressing staircase artifacts and denoising.

With the fast development of photon counting detection techniques, spectral computed tomography (CT) with a photon counting detector has attracted considerable attention by increasing energy resolution to identify and discriminate materials. The conventional analytic reconstruction algorithms can be directly applied to reconstruct spectral CT images for each spectrum or energy bin. However, a comprehensive evaluation of analytic reconstruction algorithms for spectral CT has not been reported yet. This motivates us to evaluate the analytic spiral cone-beam CT algorithms and to provide a fair comparison platform for the state-of-the-art iterative spectral CT reconstruction algorithms. Considering the fact that a narrow energy bin has high noise which degrades the imaging quality of spectral CT, an adaptive maximum a posterior projection restoration algorithm is first used to reduce the noise, and then, a 2-D/3-D weighted spiral Feldkamp–Davis–Kress algorithm is implemented to reconstruct the spectral CT images at different energy bins. Finally, the principle component analysis is employed to render the spectral reconstruction results into a color space. Our numerical results show that the analytic reconstruction approach is fast and it can provide high spatial resolution, high contrast resolution, and high signal-to-noise-ratio under higher helical pitches. This makes it possible to serve as a platform to evaluate the state-of-the-art iterative spectral CT algorithms.

In deflection tomography, Algebraic reconstruction technique (ART) is a common reconstruction technique. In this paper, we study that projection access order has an important effect on the reconstruction of the measured field in ART. We show that a multilevel scheme (MLS) projection access ordering in ART for deflection tomography. The algorithm is carried out by a three-peak Gaussian simulative temperature field. The simulation results are showed in this paper, and compared with the traditional sequential access projection method. Comparisons show that this algorithm improved reconstruction quality and the speed of convergence, especially in cases with noisy data added.

The variational level set approaches are widely used in image segmentation. To extract the objects from the inhomogeneity image, an improved local and global binary fitting active contours model is presented in this paper. The energy functional has been defined by a local intensity fitting term and a global intensity fitting term. The generalized Gaussian distribution has been used as the kernel function of the local binary fitting information. The experiment results of the homogeneity images, artificial and real inhomogeneous images have shown the advantages of LGBF model in terms of accuracy and robustness.

In view of the shortages of the reconstruction algorithm based on Total Variation (TV) minimum under the framework of measured field compressed sensing, we study the measured field sparse representation method and solving method of optimization equation, and propose the measured field reconstruction algorithms based on Diagonal Total Variation (DTV). When there is no obvious change in the reconstruction iteration of TV, gradient transformation of diagonal direction is introduced, the multi-directional information is used to obtain a more sparse representation of the measured field in the reconstruction. Under the condition of sparse projections, experimental results of this algorithm are demonstrated and compared with the results from the TV method. Comparisons show that this method can reconstruct high-quality measured field.

Energy auto-modulation is an important tool in X-ray imaging, as it can improve the quality and longevity of an x-ray imaging system. Because of the complex nature of imaged objects, X-ray energy auto-modulation may be difficult. If there is a physical model about imaging mechanism, one can forecast the best imaging parameters using a pre-scan that can be fed into this model. This paper offers a physical model, which is called the Voltage-Thickness-Gray (VTG) model. Based on equivalent single-energy, this paper uses the empirical formula of X-ray attenuation and X-ray photon intensity to build this VTG model. Then use linear regression to estimate the model's parameters, by multi-voltage imaging about the steel wedge block. At last, by the experiment of the steel step block, verify this model and forecast the imaging tube voltage. The result shows this model can better reflect X-ray attenuation imaging properties, and can be used to forecast the imaging voltage. Also the forecast precision can achieve 90% or so.

The sparse-view problem of image reconstruction encountered in computed tomography (CT) is an important research issue due to its considerable potential in decreasing radiation dose and improving detection efficiency. Among the current sparse-view CT reconstruction algorithms, iterative reconstruction algorithms that consider total variation (TV) regularization exhibit good performance. However, the gradient difference direction is singular or fixed in conventional TV algorithms, leading to undesired artefacts when minimizing TV in sparse-view CT. To effectively address this issue, based on TV minimization, the gradient difference directional information is introduced as additional prior information in the regularization term, and a new gradient-based directional total variation minimization algorithm is proposed, which adaptively chooses mutative gradient directions to calculate the directional difference operators, and calculates the sum of the direction difference operators. In addition, to solve the fault tolerance and computational load, considering redundant blocks of reconstructed image, we can estimate the gradient directional information of each subblock via the gradient approximation method. For simplicity, the proposed algorithm is termed the BDTV algorithm. To demonstrate the superiority of the proposed algorithm, the simulation data and actual CT data from different algorithms are compared, and the results indicate that the proposed algorithm is effective at preserving details that are lost in the TV minimization, artefact reduction and noise suppression in sparse-view CT.

In conventional X-ray imaging, the ray image of a complicated structural component (complex in shape or structure or composed of multiple materials) often appears to be overexposed or underexposed. This appearance is because of the variations in the effective thickness of the component along the orientation of the X-ray penetration exceed the limit of the dynamic range of the X-ray imaging system. Complete structural information cannot be obtained, and this lack will impact the quality of the X-ray image. To solve this problem, the technology of multi-voltage image stack reconstruction has been advanced. In this new method, the tube voltage is adjusted several times to accommodate the full necessary range of effective thickness, and then, the extracted images are fused to reconstruct the full projection information. The new technology can adequately enlarge the hardware potential to completely represent the internal structure information of a complicated structural component.

A novel deterministic SEIS model for the transmission of email viruses in growing communication networks is formulated. Interestingly, the model is different from classical SEIS models not only in the form, but also in the dynamical features. We study the equilibria and their stability and analyse the bifurcation dynamics of the model. In particular, we find that the virus-free equilibrium is locally asymptotically stable for any parameter values, which may attribute to the absence of the basic reproduction number. It is shown that the model undergoes a saddle-node bifurcation and admits the bistable phenomenon. Moreover, on the basis of the Lyapunov function, the domains of attraction of equilibria are estimated by solving an LMI optimization problem. Based on the above theoretical results, some effective strategies are also provided to control the propagation of the email viruses. Additionally, our results are confirmed by numerical simulations.

This paper proposed a radial basis function neural network optimization algorithm with a hybrid adaptive mutation particle swarm. During the optimization of RBF neural networks, the HAMPSO method is adopted to train the network structure and applied to solve the problems of the target localization. The HAMPSO algorithm is a dynamically adaptive optimization approach using uniform distribution mutation and Gaussian distribution mutation to escape local optima. We propose a HAMPSO method that can expedite convergence toward the global optimum during the iterations. In order to verify that the proposed HAMPSO-RBF approach has effect, comparisons with the RBF, genetic algorithm based RBF and PSO-based RBF approach are made. The computational results proved that the proposed HAMPSO-RBF approach exhibits much better and faster convergence performance in the training process as well as better prediction ability in the validation process than the results of other three approaches.

A computer code was developed to simulate the operation of radioscopic or tomographic devices. The simulation is based on ray-tracing techniques and on the X-ray attenuation law. However, in the process of projection simulation, we can’t use an expression to describe the distribution of the consecutive X-ray spectrum which is simulated by the Monte Carlo method. We only use numerical integration to realize the projection simulation. So in order to enhance the fidelity of the projection simulation, we research X-ray spectrum sampling technology, and propose projection simulation algorithm based on self-adaptive Simpson integration. The algorithm mainly uses self-adaptive Simpson integration to sampling, based on the distribution of the X-ray spectrum. Then realize the integration of the projection process. With simulation experiment and practical experiment, we have demonstrated that the new sampling method is not only more accurate but also very efficient. This new sampling method is expected to be very useful in X-ray projection simulation, as well as in computing similar sampling or integrals in other area.

In the industrial CT detection systems, because of the restrictions of noises such as scattering noise, the traditional methods for CT reconstruction have high noise and low resolution, and cannot detect smaller flaws. So this paper has proposed using the algorithm of characteristic reconstruction to reconstruct the characteristic of object. About this algorithm, relying on the inner structure character of detected object, according to the character of wavelet function before and after RADON transform, the characteristic part of the image is reconstructed by prefiltering the projected data. Then some experiment show that the method can protrude the inner small flaw of detected object with the less sacrifice of the resolution of image background.

For complicated structural components characterized by wide X-ray attenuation ranges, the conventional computed tomography (CT) imaging using a single tube-voltage at each rotation angle cannot obtain all structural information. This limitation results in a shortage of CT information, because the effective thickness of the components along the direction of X-ray penetration exceeds the limitation of the dynamic range of the X-ray imaging system. To address this problem, high-dynamic-range CT (HDR-CT) reconstruction is proposed. For this new method, the tube’s voltage is adjusted several times to match the corresponding effective thickness about the local information from an object. Then, HDR fusion and HDR-CT are applied to obtain the full reconstruction information. An accompanying experiment demonstrates that this new technology can extend the dynamic range of X-ray imaging systems and provide the complete internal structures of complicated structural components.

In the process of X-ray digital imaging detection, previous test usually adopts the projection data of 360° to reconstruct. But in practical test, it's often restricted to the installation of the equipment and the cost, so it cannot meet the condition of reconstruction. This paper studies a method to solve the coordinate of circle center about a particular component by some experience information, which is based on the projection values of three views in order to determine the distribution of holes. Experiment shows that the method not only reduces the collection angles and the cost, which satisfies the problem of detection in practical application, but also guarantees the accuracy of the positional information of the internal holes, which is compared with the 900 views under reconstruction.

Fixed-voltage Digital Radiography(DR) image has over-exposed or under-exposed regions since dynamic range of the imaging device is not enough when testing a composite component with a complicated internal structure. Some structure information of the composite component is lost. This problem can be solved by fusing the multi-voltage DR images. A multi-voltage DR images fusion model is proposed based on the gray consistency of the images. It adaptively determines the weights of all images by solving the model. What's more, this method is without prior knowledge. The experiment shows the new method can reconstruct the whole information of the complicated structural component. Also the new method has higher fusion efficiency because of no prior knowledge.

Computer users’ reactions to the outbreak of Internet worm directly determine the defense capability of the computer and play an important role in the spread of worm. In this paper, in order to characterize the impacts of adaptive user protection behaviors, an improved SIS model is proposed to describe the Internet worm propagation. The results of theoretical analysis indicate that the protective campaigns of users can indeed reduce the worm’s reproduction number to values less than one. But it may not be sufficient to eradicate the worm. In certain condition, a backward bifurcation leading to bistability can occur. These are new findings in the worm propagation model that bring new challenges to control the spread of the worm and further demonstrate the importance of user behaviors in controlling the worm propagation. Corresponding to the analysis results, defense and control strategies are provided.

It is quite important to obtain the sensor nodes location information in the underwater acoustic sensor networks localization. A method of acoustic sensor network node self-localization based on adaptive particle swarm optimization is proposed aiming at the stringent difficulties of the underwater acoustic sensor node localization and the shortage of standard particle swarm optimization (PSO) algorithm which is easily trapped into the local optimum. In the method, the global search ability and the local performance of the PSO algorithm are effectively improved by balancing the stochastic inertia weight. At the same time, the proposed method finds easy and elegant solutions to get rid of the local optimization by adopting the adaptive mutation strategy. The experimental results indicated that the new method can effectively solve the current problem in the underwater acoustic sensor node localization, and the pointing accuracy achieves 0.605m.

The convergence speed of algebraic reconstruction technique (ART) depends heavily on the order in which the projections are considered. In this study, a projection access scheme based on prime number increment is proposed, which is applicable to uniform projection sampling in any angle range. We compared the results reconstructed from the proposed method with the results reconstructed from the conventional sequential method, the prime number decomposition method and random ordering method, for cone-beam X-ray computed tomography reconstruction and for the case of circular acquisition. The results indicate that using the proposed method can accelerate the convergence of ART and produces more accurate images with fewer artifacts.

A method that can fast, real-time get the projection matrix in ART reconstruction is proposed based on helical cone-beam CT trajectory in this paper. For a given ray this method can rapidly acquire the indices and corresponding lengths of the voxels passed through by this ray through sorting the intersections between the ray and mesh planes. Meanwhile we only store the projection matrix elements required for a given ray and release them after using them. By doing so the computation time and the storage space are significantly reduced. The experimental results show that this method can heighten the speed and better suppress noise.

Target localization based on time difference of arrival (TDOA) measurements has important applications in sonar, radar and sensor networks. This paper simply introduced the target localization principle of moving emitter and the position location algorithm. Further more presented an improved particle swarm optimization with stochastic inertia weight and adaptive mutation, and adopts it to solve the target localization problem according to the batch of continuous TDOA measurements. The experimental results show that the new algorithm has higher localization accuracy, better algorithm stability and faster convergence rate.

Ordered-subset simultaneous algebraic reconstruction technique (OS-SART) was studied by Ge Wang and Ming Jiang in 2004. It accelerate the convergence of SART, but it has some disadvantages, such as increasing the number of subsets accelerates iterative convergence, but there is a point beyond which image quality degrades due to a lack of statistical information within subset. In this paper, a new method of subset partition based on statistical test is proposed as an improved OS- SART (IOS-SART). IOS-SART can automatically adjust the number of the subsets for each iteration according to the statistical information content within subset demanded by user. Numerical simulation and application to practical data demonstrate that this algorithm converge faster and can provide high quality reconstructed images after a small number of iterations.