Sparse-view projection can reduce the scan does and scan time of Cone-Beam Computed Tomography (CBCT) effectively but brings a lot of streak artifacts to the reconstructed images. Sinogram inpainting can generate projection data for missing angles and improve the quality of reconstructed images. Based on the above, a Residual Encoder-Decoder Generative Adversarial Network (RED-GAN) was proposed for sinogram inpainting to reconstruct sparse-view CBCT images. In this network, the U-Net generator in Pix2pixGAN (Pix2pix Generative Adversarial Network) was replaced with the Residual Encoder-Decoder (RED) module. In addition, the conditional discriminator based on PatchGAN (Patch Generative Adversarial Network) was used to distinguish between the repaired sinograms from the real sinograms, thereby further improving the network performance. After the network training using real CBCT projection data, the proposed network was tested under 1/2, 1/3 and 1/4 sparse-view sampling conditions, and compared with linear interpolation method, Residual Encoder-Decoder Convolutional Neural Network (RED-CNN) and Pix2pixGAN. Experimental results indicate that the sinogram inpainting results of RED-GAN are better than those of the comparison methods under all the three conditions. Under the 1/4 sparse-view sampling condition, the proposed network has the most obvious advantages. In the sinogram domain, the proposed network has the Root Mean Square Error (RMSE) decreased by 7.2%, Peak Signal-to-Noise Ratio (PSNR) increased by 1.5% and Structural Similarity (SSIM) increased by 1.4%; in the reconstructed image domain, the proposed network has the RMSE decreased by 5.4%, PSNR increased by 1.6% and SSIM increased by 1.0%. It can be seen that RED-GAN is suitable for high-quality CBCT reconstruction and has potential application value in the field of fast low-dose CBCT scanning.
Concerning the problems of poor classification of Capsule Network (CapsNet) on complex datasets and large number of parameters in the routing process, a Capsule Network based on Multipath feature (MCNet) was proposed, including a novel capsule feature extractor and a novel capsule pooling method. By the capsule feature extractor, the features of different layers and locations were extracted in parallel from multiple paths, and then the features were encoded into capsule features containing more semantic information. In the capsule pooling method, the most active capsules at each position of the capsule feature map were selected, and the effective capsule features were represented by a small number of capsules. Comparisons were performed on four datasets (CIFAR-10, SVHN, Fashion-MNIST, MNIST) with models such as CapsNet. Experimental results show that MCNet has the classification accuracy of 79.27% on CIFAR-10 dataset and the number of trainable parameters of 6.25×106; compared with CapsNet, MCNet has the classification accuracy improved by 8.7%, and the number of parameters reduced by 46.8%. MCNet can effectively improve the classification accuracy while reducing the number of trainable parameters.
Aiming at the insufficient deep network model training problem caused by limited rotary machine sensor signal samples, a fault diagnosis model combining improved residual convolutional auto-encoding network and class adaption method was proposed to deal with the data with small sample size. Firstly, paired samples were created by a small number of labeled source domain data and target domain data, and an improved one-dimensional residual convolutional auto-encoding network was designed to extract features from two types of original vibration signals with different distributions. Secondly, the Maximum Mean Discrepancy (MMD) was used to reduce the distribution difference, and the data space of the same fault category from two domains was mapped to a common feature space. Finally, the accurate fault diagnosis was realized. Experimental results show that the proposed model is able to effectively improve the fault diagnosis accuracy of the target domain vibration data with few labels under different working conditions compared with the fine-tuning and domain adaptation methods.
The recognition of spam is one of the main tasks in natural language processing. The traditional methods are based on text features or word frequency, which recognition accuracies mainly depend on the presence or absence of specific keywords. When there are no keywords or errors in recognizing keywords in the spam, the traditional methods have poor recognition performance. Neural network-based methods were proposed. Recognition training and testing were conducted on complex spam. The spams that cannot be recognized by traditional methods were collected and the same amount of normal information was randomly selected from spam messages, advertisement and spam email datasets to form three new datasets without duplicate data. Three models were proposed based on convolutional neural network and recurrent neural network and tested on three new datasets for spam recognition. The experimental results show that the neural network-based models learned better semantic features from the text and achieved the accuracies of more than 98% on all three datasets, which are significantly higher than those of the traditional methods, such as Naive Bayes (NB), Random Forest (RF) and Support Vector Machine (SVM). The experimental results also show that different neural networks are suitable for text classification with different lengths. The models composed of recurrent neural networks are good at recognizing text with sentence length, the models composed of convolutional neural networks are good at recognizing text with paragraph length, and the models composed of both neural networks are good at recognizing text with chapter length.
To study the performance and application prospects of novel intelligent optimization algorithms, six bionic intelligent optimization algorithms proposed in the past few years were analyzed, concluding Harris Hawks Optimization (HHO) algorithm, Equilibrium Optimizer (EO), Marine Predators Algorithm (MPA), Political Optimizer (PO), Slime Mould Algorithm (SMA), and Heap-Based Optimizer (HBO). Their performance and applications in different constrained engineering optimization problems were compared and analyzed. Firstly, the basic principles of six optimization algorithms were introduced. Secondly, the optimization tests were performed on ten standard benchmark functions for six optimization algorithms. Thirdly, six optimization algorithms were applied to solve three engineering optimization problems with constraints. Experimental results show that the convergence accuracy of PO is the best for the optimization of unimodal and multimodal test functions and can reach the theoretical optimal value zero many times. The EO and MPA are better for solving constrained engineering problems with fast optimization speed, high stability and standard deviation of a small order of magnitude. Finally, the improvement methods and development potentials of six optimization algorithms were analyzed.
To improve the accuracy and efficiency of absolute gas emission prediction, a new algorithm based on Chaos Immune Particle Swarm Optimization (CIPSO) and General Regression Neural Network (GRNN) was proposed. In this algorithm, CIPSO was employed to dynamically optimize the smooth factor of GRNN to reduce the impact of artificial factors in GRNN model construction, and then the optimized network was adopted to establish gas emission prediction model. The simulation experiment results on gas emission data of a coal mine show that the model is of faster convergence and higher prediction accuracy than other prediction models based on BP and Elman neural network. It is proved that the proposed method is feasible and effective.
Defect detection is an important part of fabric quality control. To make the detection algorithm possess good commonality and high detection accuracy, a dual-scale fabric defect detection algorithm based on sparse coding was proposed. The algorithm combined the advantage of high stability under large-scale and the advantage of high detection sensitivity under small-scale. At first, the dictionaries under large and small scales were obtained through a small-scale over-complete dictionary training method. Then, the projection of detection image block on the over-complete dictionary was used to extract detection characteristics. Finally, the detection results under dual-scale were fused by the means of distance fusion. The algorithm overcame the disadvantage of large computation because of the introduction of dual-scale while using small-scale over-complete dictionary and downsampling the detection image under large-scale. TILDA Textile Texture Data base was used in the experiment. The experimental results show that the algorithm can effectively detect defects on plain, gingham and striped fabric, the comprehensive detection rate achieves 95.9%. And its moderate amount of calculation can satisfy the requirement of industrial real-time detection, so it does have much value of practical application.