International Journal of Pattern Recognition and Artificial Intelligence

Aggregate depth residual network (ResNeXt) can not only improve the accuracy without increasing the parameter complexity, but also reduce the number of super parameters. It is one of the popular convolutional neural network models for image recognition. Maize diseases have a great impact on maize yield, quality and farmers’ income. Rapid and effective identification of the severity of maize diseases plays an important role in accurate control and accurate drug use. The general ResNeXt model has large spots in extracting image features, but the spots of corn diseases are small and the extracted features are not obvious, which affects the recognition accuracy. Therefore, an improved ResNeXt model is proposed to recognize the occurrence degree of corn diseases. First, the original data of maize disease degree are classified according to national standards. Second, the original data are extended through data enhancement. Third, the original ResNeXt101 model is improved. The first layer convolution kernel is changed to three 3 * 3 convolution kernels, and the cardinality is adjusted to 64. Finally, the improved model is verified. The recognition accuracy of corn disease degree is 89.667%, which is 0.98% higher than the original model. Through testing on 276 actually collected corn disease images, the recognition accuracy is 90.22%. Therefore, this method is feasible for the diagnosis of corn disease degree and can provide an important basis for accurate prevention and control.

In order to quantitatively analyze the proportions of independent components in mixtures, it is necessary to extract line spectra corresponding to those components from the spectrum signal of some mixture and evaluate the amplitude of the spectral lines. Multiple factors cause the drift and tilt of a spectrum signal’s baseline, such as environment noises, instrument bias, and sample size, which affect the identification and quantitative analysis of the line spectra superimposed on the baseline. Therefore, the baseline of a spectrum signal should be removed before the line spectra are identified. A baseline correction algorithm based on Catastrophe Point detection and Lipschitz exponent’s analysis is proposed in this paper. With the algorithm, the strong spectral lines are identified and removed, and then the spectral baseline is evaluated without the interference of strong spectrum signals. First, catastrophe points are located based on the local modulus maxima theory of wavelet coefficients. Second, according to the Lipschitz exponent theory, the strong spectral peaks’ regions are identified and removed by a smoothing filter. Then the slowly varying spectrum is segmented adaptively and fitted by the least square fitting method. After the segments are attached and the boundaries are smoothed, the baseline of the spectrum is acquired and extracted finally. The algorithm is more accurate than classical ones because identifying the baseline is implemented after strong peaks are removed, so their influences to baseline extracting are eliminated. The results of experiments show that the algorithm is accurately performed for the spectrum signal of a gas mixture, $S{F}_6$.

With the rapid development of online e-commerce, traditional collaborative filtering algorithms have the disadvantages of data set reduction and sparse matrix filling cannot meet the requirements of users. This paper takes handicrafts as an example to propose the design and application of handicraft recommendation system based on an improved hybrid algorithm. Based on the theory of e-commerce system, through the traditional collaborative filtering algorithm of users, the personalized e-commerce system of hybrid algorithm is designed and analyzed. The personalized e-commerce system based on hybrid algorithm is further proposed. The component model of the business recommendation system and the specific steps of the improved hybrid algorithm based on user information are given. Finally, an experimental analysis of the improved hybrid algorithm is carried out. The results show that the algorithm can effectively improve the effectiveness and exemption of recommending handicrafts. What’s more, it can reduce the user item ratings of candidate set and improve accuracy of the forecast recommendation.

In order to solve the problem of dataset expansion in deep learning tasks such as image classification, this paper proposed an image generation model called Class Highlight Generative Adversarial Networks (CH-GANs). In order to highlight image categories, accelerate the convergence speed of the model and generate true-to-life images with clear categories, first, the image category labels were deconvoluted and integrated into the generator through $1\times 1$ convolution. Second, a novel discriminator that cannot only judge the authenticity of the image but also the image category was designed. Finally, in order to quickly and accurately classify strip steel defects, the lightweight image classification network GhostNet was appropriately improved by modifying the number of network layers and the number of network channels, adding SE modules, etc., and was trained on the dataset expanded by CH-GAN. In the comparative experiments, the average FID of CH-GAN is 7.59; the accuracy of the improved GhostNet is 95.67% with 0.19M parameters. The experimental results prove the effectiveness and superiority of the methods proposed in this paper in the generation and classification of strip steel defect images.

Brain tumor is a severe nervous disorder that causes damage to health and often leads to death. Therefore, it is significant to classify the brain tumor at an early stage as it increases the survival rate of patients. One of the commonly employed imaging modalities for brain tumor classification is Magnetic Resonance Imaging (MRI). However, it is relatively complex to perform the brain tumor classification process due to the variations of type, shape, size and tumor location. To overcome such issues and classify the tumor more accurately, a deep learning classifier named Deep Maxout network is developed to classify the tumor into different grades. Based on the classification result, the features connected with the tumor grades are effectively acquired to make the survival prediction process. Deep learning is an effective and robust classifier model employed to perform the tumor classification or detection process with the MRI modality. Here, the survival prediction of tumor patients is carried out by the Deep Long Short-Term Memory (LSTM) classifier. Accordingly, the proposed method achieved higher performance using accuracy, sensitivity, specificity and prediction error with the values of 0.9434, 0.9324, 0.9202 and 0.0579.

It is an obvious fact that drivers’ drowsiness is more likely to cause traffic accidents. Recently, driver drowsiness detection has drawn considerable attention. In this paper, a novel drowsiness detection scheme is proposed, which can recognize drivers’ drowsiness actions through their facial expressions. First, a drowsiness action recognition model based on 3D-CNN is proposed, which can effectively distinguish drivers’ drowsiness actions and nondrowsiness actions. Second, a fusion algorithm of the two input streams is proposed, which can fuse gray image sequence and optical image sequence containing target motion information. Finally, the proposed model is evaluated on National Tsinghua University Driver Drowsiness Detection (NTHU-DDD) dataset. The experimental results show that the algorithm performs better than other algorithms, and its accuracy reaches 86.64%.

The sensors or electronic components are vulnerable to interference in the camera’s imaging process, usually leading to random directional stripes. Therefore, a method of stripe noise removal based on compressed sensing is proposed. First, the measurement matrix of the image with stripe noise is established, which makes the stripe images equivalent to the observation of the original image. Second, the relationships between the corresponding coefficients of adjacent scales are defined. On this basis, the bivariate threshold function is set in the curvelet sparse domain to represent the features of images. Finally, the Landweber iteration algorithm of alternating convex projection and filtering operation is achieved. Furthermore, to accelerate the noise removal at the initial stage of iteration and preserve the image details later, the exponential threshold function is utilized. This method does not need many samples, which is different from the current deep learning method. The experimental results show that the proposed algorithm represents excellent performance in removing the stripes and preserving the texture details. In addition, the PSNR of the denoised image has been dramatically improved compared with similar algorithms.

3D human pose estimation describes estimating 3D articulation structure of a person from an image or a video. The technology has massive potential because it can enable tracking people and analyzing motion in real time. Recently, much research has been conducted to optimize human pose estimation, but few works have focused on reviewing 3D human pose estimation. In this paper, we offer a comprehensive survey of the state-of-the-art methods for 3D human pose estimation, referred to as pose estimation solutions, implementations on images or videos that contain different numbers of people and advanced 3D human pose estimation techniques. Furthermore, different kinds of algorithms are further subdivided into sub-categories and compared in light of different methodologies. To the best of our knowledge, this is the first such comprehensive survey of the recent progress of 3D human pose estimation and will hopefully facilitate the completion, refinement and applications of 3D human pose estimation.

Pose estimation is the basis and key of human motion recognition. In the two-dimensional human pose estimation based on image, in order to reduce the adverse effects of mutual occlusion among multiple people and improve the accuracy of motion recognition, a structurally symmetrical two-dimensional multi-person pose estimation model combined with face detection is proposed in this paper. First, transfer learning is used to initialize each sub-branch network model. Then, MTCNN is used for face detection to predict the number of people in the image. According to the number of people, the image is input into the improved two-branch OpenPose network. What is more, the double judgment algorithm is proposed to correct the false detection of MTCNN. The experimental results show that compared with TensorPose, which is the latest improved method based on OpenPose, the Average Precision (AP) (Intersection over Union $\mathrm{\text{(IoU)}}=0.5$) on the validation set is 8.8 higher. Furthermore, compared with OpenPose, the mean AP ($\mathrm{\text{IoU}}=0.5:0.95$) is 1.7 higher on the validation set and is 1.3 higher on the Test-dev test set.

The number of existing person re-identification datasets is limited, and there are a series of changes such as illumination, background occlusion and pose among each dataset, which makes it difficult for the existing methods to learn robust feature representation, leading to a decline in recognition performance. To solve these problems, a person re-identification method combining style and pose generation is proposed in this paper. First with the impact of camera style differences in collecting images from different cameras, a style transformation method based on generative adversarial network is introduced into a person re-identification model, and cyclic generative adversarial networks (CycleGAN) is used to realize style transfer and reduce the influence of camera differences. Second, in view of the problem that when pedestrian pose changes greatly, easy to ignore identity-sensitive related information, AlphaPose is introduced to implement pose estimation. Combining style and posture for the first time and using improved deep convolution generative adversarial network (DCGAN) structure enrich the input sample information and generate unified style pose image; while using new synthetic data to train person re-identification network model improves the recognition performance of the model. Finally, further introducing random erasure method during data enhancement, in order to reduce the overfitting phenomena, improves the generalization ability of the network simultaneously and solves partial occlusion. The experimental results show that the proposed method outperforms typical style-based or pose-based methods. The accuracy of rank-1 and mAP on Market-1501 dataset is 90.4% and 74.5%, respectively, which are 2.28% and 5.78% higher, respectively. To a certain extent, the performance of person re-identification is improved.

The watermark embedded by traditional methods is easy to be lost under some attacks. To overcome this problem, this study proposes a novel method based on DWT. It adopts a digital audio watermarking state-switching system which optimizes DWT coefficients doubly. Firstly, it combines the quantization-embedding system and the weights of DWT coefficients with SNR to obtain an optimization model for watermarking. Next, the Lagrange principle, Hessian matrix, and minimum energy play three essential roles to obtain the optimal DWT coefficients and weights. Moreover, the almost invariant feature of the optimal weights holds demonstrating resistance to amplitude scaling. Compared with similar algorithms, the experimental results verify that the embedded audio in the proposed method has higher signal-to-noise ratio (SNR) and lower bit error rate (BER). At the same time, it indicates stronger robustness against various attacks, such as re-sampling, amplitude scaling, and mp3 compression.

The original minimum exposure path (MEP) problem in WSNs (wireless sensor networks) requires the starting point and ending point of a moving target to be fixed, thus limiting the evaluation of coverage quality relative to the locations of these points. To resolve this issue, a minimum exposure path problem with various boundaries (VB-MEP) is proposed in this paper. Because the corresponding graph model cannot be established, the original classical methods (grid method and Voronoi diagram method) used to solve the minimum exposure path problem are no longer effective for the VB-MEP problem. This paper first transforms the problem into a hybrid optimization problem with constraint conditions and then considers the characteristics of the transformed mathematical optimization model with high dimensionality, unfixed dimensionality and high nonlinearity so that the deterministic optimization methods are no longer applicable. A hybrid artificial colony solution algorithm that incorporates the actual characteristics of the problem is designed, and a convergence analysis and a proof of the designed algorithm are given. Through simulation experiments based on large-scale node distribution scenarios, it is found that the designed hybrid optimization model with constraints and hybrid artificial bee colony solution algorithm can effectively solve the proposed VB-MEP problem.

Pipeline leakages have plagued pipeline transportation for long time. Therefore, accurately extracting the features of leak signal in the presence of noise, and prompt identification of leak states and leak sizes is essential when leakage occurs. A novel leakage detection method based on the improved adaptive filter, whose parameters were optimized by the particle swarm optimization (PSO), was formulated and applied. The PSO-adaptive filter proved to be an effective signal processing method in contrast with variational mode decomposition (VMD). Its efficiency stems from the fact that the adaptive filter employs the noise collected from the detection environment. Therefore, the filter can adjust its parameters according to the changing situation. What is more, the application of PSO is conducive to automatically set suitable parameters for adaptive filter. After signal denoising, principal component analysis (PCA) was used for feature dimension reduction and selecting optimal features. The features after PCA proved to be more helpful in pattern recognition than the features without PCA. Furthermore, the relationship between the recognition results of leakage sizes and the measurement distance of the sensor was studied. Experimental results show that the method used in this paper can identify the leakage states with the accuracy of 100%. The identification result of leakage size reaches an accuracy of 86.75% under the influence of the measurement distance.

To address the issues of low efficiencies and serious mapping distortions in current mesh parameterization methods, we present a low distortion mesh parameterization mapping method based on proxy function and combined Newton’s method in this paper. First, the proposed method calculates visual blind areas and distortion prone areas of a 3D mesh model, and generates a model slit. Afterwards, the method performs the Tutte mapping on the cut three-dimensional mesh model, measures the mapping distortion of the model, and outputs a distortion metric function and distortion values. Finally, the method sets iteration parameters, establishes a reference mesh, and finds the optimal coordinate points to get a convergent mesh model. When calculating mapping distortions, Dirichlet energy function is used to measure the isometric mapping distortion, and MIPS energy function is used to measure the conformal mapping distortion. To find the minimum value of the mapping distortion metric function, we use an optimal solution method combining proxy functions and combined Newton’s method. The experimental data show that the proposed method has high execution efficiency, fast descending speed of mapping distortion energy and stable optimal value convergence quality. When a texture mapping is performed, the texture is evenly colored, close laid and uniformly lined, which meets the standards in practical applications.