通讯机构:
[Zhang, F ] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
关键词:
Face anti -spoofing;Presentation attack;Disentangled representation;Deep learning;Face recognition
摘要:
Face recognition systems have been widely applied in security-related areas of our daily life. However, they are vulnerable to face spoofing attacks. Specifically, an attacker can fool a face recognition system into making false decisions, by presenting spoof face information (such as printed photos, replayed videos, etc.), rather than live face, to the face recognition system. Therefore, Face Anti-Spoofing (FAS) is critical for the security operation of a face recognition system.Deep learning-based FAS approaches show the best performance among existing FAS approaches. The basic idea of deep learning-based FAS approaches is to learn statistical representations capable of distin-guishing spoof faces from live ones, and then leverage the learned representations for live and spoof face classifications. Therefore, the learned representations play a key role in the performance of FAS. However, most existing approaches learn representations from representation-entangled spaces, in which critical and irrelevant representations for live and spoof face classifications are entangled with each other, thereby bringing a negative influence on the performance of a FAS system.To address the issue, we introduced a Twin Autoencoder Disentanglement (TAD) framework. Our TAD framework utilizes adversarial learning and a reconstruction strategy to disentangle both critical and irrelevant representations into two mutually independent representation spaces. In addition, to further suppress irrelevant representations that may remain in the critical representation space, we design a multi-branch supervision architecture (MSA) and embed it into TAD. MSA achieves the goal via imposing depth supervision and pattern supervision to the critical representation space. i.e., learning spatial rep-resentation (face depth information) and texture representation (face spoof pattern information).Experimental results on four typical public datasets, OULU-NPU, SiW, Replay-Attack, and CASIA-MFSD, demonstrate that our proposed TAD approach successfully disentangles critical and irrelevant represen-tations, and the two disentangled representations are more interpretable than state-of-the-art FAS meth-ods. The codes are available at https://github.com/TAD-FAS/TAD.& COPY; 2023 Elsevier B.V. All rights reserved.
摘要:
With the increasing popularity of online fruit sales, accurately predicting fruit yields has become crucial for optimizing logistics and storage strategies. However, existing manual vision-based systems and sensor methods have proven inadequate for solving the complex problem of fruit yield counting, as they struggle with issues such as crop overlap and variable lighting conditions. Recently CNN-based object detection models have emerged as a promising solution in the field of computer vision, but their effectiveness is limited in agricultural scenarios due to challenges such as occlusion and dissimilarity among the same fruits. To address this issue, we propose a novel variant model that combines the self-attentive mechanism of Vision Transform, a non-CNN network architecture, with Yolov7, a state-of-the-art object detection model. Our model utilizes two attention mechanisms, CBAM and CA, and is trained and tested on a dataset of apple images. In order to enable fruit counting across video frames in complex environments, we incorporate two multi-objective tracking methods based on Kalman filtering and motion trajectory prediction, namely SORT, and Cascade-SORT. Our results show that the Yolov7-CA model achieved a 91.3% mAP and 0.85 F1 score, representing a 4% improvement in mAP and 0.02 improvement in F1 score compared to using Yolov7 alone. Furthermore, three multi-object tracking methods demonstrated a significant improvement in MAE for inter-frame counting across all three test videos, with an 0.642 improvement over using yolov7 alone achieved using our multi-object tracking method. These findings suggest that our proposed model has the potential to improve fruit yield assessment methods and could have implications for decision-making in the fruit industry.
通讯机构:
[Yang, H ] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430040, Peoples R China.
关键词:
text classification;convolutional neural networks (CNN);attention mechanism;multi-layer feature fusion
摘要:
Text classification is one of the fundamental tasks in natural language processing and is widely applied in various domains. CNN effectively utilizes local features, while the Attention mechanism performs well in capturing content-based global interactions. In this paper, we propose a multi-layer feature fusion text classification model called CAC, based on the Combination of CNN and Attention. The model adopts the idea of first extracting local features and then calculating global attention, while drawing inspiration from the interaction process between membranes in membrane computing to improve the performance of text classification. Specifically, the CAC model utilizes the local feature extraction capability of CNN to transform the original semantics into a multi-dimensional feature space. Then, global attention is computed in each respective feature space to capture global contextual information within the text. Finally, the locally extracted features and globally extracted features are fused for classification. Experimental results on various public datasets demonstrate that the CAC model, which combines CNN and Attention, outperforms models that solely rely on the Attention mechanism. In terms of accuracy and performance, the CAC model also exhibits significant improvements over other models based on CNN, RNN, and Attention.
期刊:
IEEE Transactions on Fuzzy Systems,2023年:1-15 ISSN:1063-6706
作者机构:
[Kun Hu] School of Computer Science, The University of Sydney, Camperdown, NSW, Australia;[Jun Bai] School of Information Technology, Deakin University, Melbourne, Australia;[Yuanyan Tang] Faculty of Science and Technology, University of Macau, Macau, China;[Shan Zeng; Xiangjun Duan; Wei Tao] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, China
摘要:
This paper proposes a novel Soft Multi-Prototype clustering algorithm (SMP) for high-dimensional data clustering with noisy and complex structural patterns. SMP integrates dimensionality reduction, multi-prototype clustering, and multi-prototype merge clustering under a two-layer Semi-Non-negative Matrix Factorization (Semi-NMF) architecture. Specifically, the first Semi-NMF layer performs multi-prototype clustering, which solves the problem that a single prototype cannot represent complex data structures. Meanwhile, the multi-prototype fuzzy clustering constraints ensure that the multi-prototypes better characterize the original data structure. The second Semi-NMF layer performs multi-prototype merge clustering to mitigate the issues of heavy computation burden and poor anti-noise performance of the spectral clustering algorithm. The introduction of the Laplace graph matrix regularization constraint in this layer assists SMP in completing the merging of multi-prototypes with complex data structures. Comprehensive experiments demonstrate that the proposed method outperforms the state-of-the-art algorithms.
期刊:
2023 2nd Asia Conference on Electrical, Power and Computer Engineering (EPCE),2023年:151-157
作者机构:
[Yang Wenzhuo; Liu Shuo] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan
摘要:
To enhance the parameters of the kernel function in the conventional support vector machine (SVM) - σ and penalty factor C - an advanced whale optimization algorithm (IWOA) is introduced within this article to optimize the SVM parameter model (IWOA-SVM). The IWOA algorithm is employed to augment the optimization capability of the original whale optimization algorithm, focusing on three key aspects: Firstly, the chaotic circle mapping technique is utilized to produce the initial positions of the initial whale population, which serves as a foundation for population diversity during the algorithm's global search process; Secondly, an adaptable weight parameter is integrated into the spiral ascent phase of the whale, which reinforces the local exploration capability of IWOA, accelerates its rate of convergence and augments the precision of the algorithm; Lastly, the Cauchy mutation perturbation is employed to alter the current optimal solution, thereby averting the algorithm from being confined to a local optima state. The optimization of parameters for the SVM kernel function is achieved through the Improved Whale Optimization Algorithm, by tuning the kernel function's parameters such as σ and penalty factor C, and then verified on the UCI dataset. In comparison to conventional SVM, Particle Swarm Optimization SVM, Genetic Algorithm Optimization SVM, and Original Whale Algorithm Optimization SVM, IWOA-SVM demonstrates the highest classification accuracy, indicating its effectiveness as an SVM parameter optimization algorithm.
摘要:
This paper proposes an anchor-free wheat ear detection method using ObjectBox with attention. First, on base of the backbone of ObjectBox, convolutional block attention module is used to improve the connection of each feature in the channel and space and enhance the feature extraction ability of the network. Second, in the neck part, ConvNeXtBlock is used to better fuse or extract the feature map given by the backbone. Last, the non-maximum suppression algorithm is improved to remove the center redundant detection box. The experimental results on the public global wheat head detection dataset show that the proposed method has an mean Average Precision (mAP) of 96.0%, an Precision of 94.5%, an Recall of 92.2% and
$$F_{1}$$
score of 93.3%. Compared with the original ObjectBox model, the improvement for mAP, Precision, Recall and
$$F_{1}$$
score is 2.0%, 1.3%, 2.9% and 2.1%, respectively. Compared with other existing wheat ear detection methods, it has higher detection accuracy.
摘要:
Inspired by an earlier work, which showed that the Hankel matrix constructed by a global positioning system (GPS) coordinate time series has a low rank, this study proposes a low-rank matrix approximation based on a nonconvex log-sum function minimization for extracting seasonal signals from GPS coordinate time series. Compared to a convex nuclear norm minimization, the nonconvex log-sum function is more approximate for the rank function. Unlike a nuclear norm minimization, which can be solved by singular value thresholding, obtaining the solution of a nonconvex log-sum function minimization is challenging. By using the iterative reweighted nuclear norm algorithm, the nonconvex low-rank matrix approximation problem is addressed in this paper by iteratively updating the weighted nuclear norm minimization problem, which has a globally optimal solution under certain conditions. Correspondingly, the residuals were analyzed using a maximum likelihood estimation. Both the proposed method and the weighted nuclear norm minimization approach were compared with classical methods in the presence of time-correlated noise. Compared to the classical seasonally extracted methods, the proposed method provides a guaranteed root mean square error performance in the presence of time-correlated noise. The experimental analysis results of both the simulated time series and real station data demonstrated that the proposed method outperformed the weighted nuclear norm minimization approach, which outperformed the classical approaches under different noise levels.
作者机构:
[Chen, Zhenhong; Xu, Peihua; Wang, Biqiang; Cheng, Chi] Hubei Meteorol Serv Ctr, Wuhan 430079, Peoples R China.;[Xu, Peihua] Cent China Normal Univ, Fac Artificial Intelligence Educ, Sch Educ Informat Technol, Wuhan 430079, Peoples R China.;[Zhang, Maoyuan] Cent China Normal Univ, Hubei Prov Key Lab Artificial Intelligence & Smart, Wuhan 430079, Peoples R China.;[Liu, Renfeng] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
通讯机构:
[Maoyuan Zhang] H;Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning, Central China Normal University, Wuhan 430079, China<&wdkj&>Author to whom correspondence should be addressed.
关键词:
discrete wavelet transform;autoencoder;bidirectional LSTM;wind power forecasting
摘要:
Due to the increasing proportion of wind power connected to the grid, day-ahead wind power prediction plays a more and more important role in the operation of the power system. This paper proposes a day-ahead wind power short-term prediction model based on deep learning (DWT_AE_BiLSTM). Firstly, discrete wavelet transform (DWT) is used to denoise the data, then an autoencoder (AE) technology is used to extract the data features, and finally, bidirectional long short-term memory (BiLSTM) is used for prediction. To verify the effectiveness of the proposed DWT_AE_BiLSTM model, we studied three different power stations and compared their performance with the shallow neural network model. Experimental analysis shows that this model is more competitive in forecasting accuracy and stability. Compared with the BP model, the proposed model has increased by 3.86%, 3.22% and 3.42% in three wind farms, respectively.
摘要:
Phthalides are a class of unique compounds such as ligustilide, butylphthalide and butyldenephthalide, which have shown to possess multiple bioactivities in new drug research and development. Phthalides are naturally distributed in different plants that have been utilized as herbal treatments for various ailments with a long history in Asia, Europe and North America. Their extensive biological activity has led to a dramatic increase in the study of phthalide compounds in recent years. This review summarizes the latest research progress of plant-derived phthalides, and a total of 133 phthalide compounds are described based on the characteristics of chemical structures. Pharmacological properties of plant-derived phthalides are associated with hemorheological improvement, vascular function modulation and central nervous system protection. Potential treatments for a variety of diseases mainly including cardio-cerebrovascular disorders and neurological complications such as Alzheimer's disease are also concluded. In addition, key metabolic pathways have been clearly elucidated. Further investigations on the molecular mechanisms involved in biological activity are recommended for offering new insights into profound comprehension of phthalide applications.
摘要:
Leaf segmentation is crucial for plant recognition, especially for tree species identification. In natural environments, leaf segmentation can be very challenging due to the lack of prior information about leaves and the variability of backgrounds. In typical applications, supervised algorithms often require pixel-level annotation of regions, which can be labour-intensive and limited to identifying plant species using pre-labelled samples. On the other hand, traditional unsupervised image segmentation algorithms require specialised parameter tuning for leaf images to achieve optimal results. Therefore, this paper proposes an unsupervised leaf segmentation method that combines mutual information with neural networks to better generalise to unknown samples and adapt to variations in leaf shape and appearance to distinguish and identify different tree species. First, a model combining a Variational Autoencoder (VAE) and a segmentation network is used as a pre-segmenter to obtain dynamic masks. Secondly, the dynamic masks are combined with the segmentation masks generated by the mask generator module to construct the initial mask. Then, the patcher module uses the Mutual Information Minimum (MIM) loss as an optimisation objective to reconstruct independent regions based on this initial mask. The process of obtaining dynamic masks through pre-segmentation is unsupervised, and the entire experimental process does not involve any label information. The experimental method was performed on tree leaf images with a naturally complex background using the publicly available Pl@ntLeaves dataset. The results of the experiment showed that compared to existing excellent methods on this dataset, the IoU (Intersection over Union) index increased by 3.9%.
摘要:
Sound event detection is sensitive to the network depth, and the increase of the network depth will lead to a decrease in the event detection ability. However, event localization has a deeper requirement for the network depth. In this paper, the accuracy of the joint task of event detection and localization is improved by decoupling SELD-TCN. The joint task is reflected in the early fusion of primary features and the enhancement of the generalization ability of the sound event detection branch as the DOA branch mask, while the advanced feature extraction and recognition of the two branches are carried out in different ways separately. The primary features extracted by resnet16-dilated instead of CNN-Pool. The SED branch adopts linear temporal convolution to realize sound event detection by imitating the linear classifier, and ED-TCN is used for the localization detection branch. The joint training of the DOA branch and the SED branch will affect each other badly. Using the most appropriate way for both branches and masking the DOA branch with the SED branch can improve the performance of both. In the TUT Sound Events 2019 dataset, the DOA error achieved an error effect of 6.73, 8.8 and 30.7 with no overlapping source data, with two and three overlapping sources, respectively. The SED accuracy has been significantly improved, and the DOA error has been significantly reduced.
期刊:
Journal of Fixed Point Theory and Applications,2023年25(2):1-31 ISSN:1661-7738
通讯作者:
Wang, CH
作者机构:
[Wang, Chunhua; Wang, CH] Cent China Normal Univ, Sch Math & Stat, Wuhan, Peoples R China.;[Wang, Chunhua; Wang, CH] Cent China Normal Univ, Hubei Key Lab Math Sci, Wuhan, Peoples R China.;[Wang, Qingfang] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.;[Yang, Jing] Jiangsu Univ Sci & Technol, Sch Sci, Zhenjiang 212003, Peoples R China.
通讯机构:
[Wang, CH ] C;Cent China Normal Univ, Sch Math & Stat, Wuhan, Peoples R China.;Cent China Normal Univ, Hubei Key Lab Math Sci, Wuhan, Peoples R China.
摘要:
We study the following nonlinear critical elliptic equation
$$\begin{aligned} -\Delta u+\epsilon Q(y)u=u^{\frac{N+2}{N-2}},\;\;\; u>0\;\;\;\hbox { in } {\mathbb {R}}^N, \end{aligned}$$
where
$$\epsilon >0$$
is small and
$$N\ge 5.$$
Assuming that Q(y) is periodic in
$$y_1$$
with period 1 and has a local minimum at 0 satisfying
$$Q(0)>0,$$
we prove the existence and local uniqueness of infinitely many bubbling solutions of it. This local uniqueness result implies that some bubbling solutions preserve the symmetry of the potential function Q(y), i.e., the bubbling solution whose blow-up set is
$$\{(jL,0,\ldots ,0):j=0,\pm 1, \pm 2,\ldots , \pm m\}$$
must be periodic in
$$y_{1}$$
provided that
$$\epsilon $$
goes to zero and L is any positive integer, where m is the number of the bubbles which is large enough but independent of
$$\epsilon .$$
摘要:
In recent years, the domain of diagnosing plant afflictions has predominantly relied upon the utilization of deep learning techniques for classifying images of diseased specimens; however, these classification algorithms remain insufficient for instances where a single plant exhibits multiple ailments. Consequently, we view the region afflicted by the malady of rice leaves as a minuscule issue of target detection, and then avail ourselves of a computational approach to vision to identify the affected area. In this paper, we advance a proposal for a Dense Higher-Level Composition Feature Pyramid Network (DHLC-FPN) that is integrated into the Detection Transformer (DETR) algorithm, thereby proffering a novel Dense Higher-Level Composition Detection Transformer (DHLC-DETR) methodology which can effectively detect three diseases: sheath blight, rice blast, and flax spot. Initially, the proposed DHLC-FPN is utilized to supersede the backbone network of DETR through amalgamation with Res2Net, thus forming a feature extraction network. Res2Net then extracts five feature scales, which are coalesced through the deployment of high-density rank hybrid sampling by the DHLC-FPN architecture. The fused features, in concert with the location encoding, are then fed into the transformer to produce predictions of classes and prediction boxes. Lastly, the prediction classes and the prediction boxes are subjected to binary matching through the application of the Hungarian algorithm. On the IDADP datasets, the DHLC-DETR model, through the utilization of data enhancement, elevated mean Average Precision (mAP) by 17.3% in comparison to the DETR model. Additionally, mAP for small target detection was improved by 9.5%, and the magnitude of hyperparameters was reduced by 324.9 M. The empirical outcomes demonstrate that the optimized structure for feature extraction can significantly enhance the average detection accuracy and small target detection accuracy of the model, achieving an average accuracy of 97.44% on the IDADP rice disease dataset.