Short-text classification is an important and challenging task in natural language processing. Compared with long text, short text suffers from insufficient semantic information. Based on a lot of theoretical and experimental analysis, a two-channel short-text classification method, DA-STCM, is proposed. This method implements word correction by introducing the Aliyun service and introduces an attention mechanism in the channels to achieve weight assignment. Experimental results show that our proposed method is feasible in short-text classification, and the classification accuracy is significantly improved.

Chinese calligraphy is a significant aspect of traditional culture, as it involves the art of writing Chinese characters. Despite the development of numerous deep learning models for generating calligraphy characters, the resulting outputs often suffer from issues related to stroke accuracy and stylistic consistency. To address these problems, an end-to-end generation model for Chinese calligraphy characters based on dense blocks and a capsule network is proposed. This model aims to solve issues such as redundant and broken strokes, twisted and deformed strokes, and dissimilarity with authentic ones. The generator of the model employs self-attention mechanisms and densely connected blocks to reduce redundant and broken strokes. The discriminator, on the other hand, consists of a capsule network and a fully connected network to reduce twisted and deformed strokes. Additionally, the loss function includes perceptual loss to enhance the similarity between the generated calligraphy characters and the authentic ones. To demonstrate the validity of the proposed model, we conducted comparison and ablation experiments on the datasets of Yan Zhenqing's regular script, Deng Shiru's clerical script, and Wang Xizhi's running script. The experimental results show that, compared to the comparison model, the proposed model improves SSIM by 0.07 on average, reduces MSE by 1.95 on average, and improves PSNR by 0.92 on average, which proves the effectiveness of the proposed model.

In the context of online text-picture relic exhibitions, two exploratory experiments were conducted to investigate the role of integrated/separate display, background information, and annotation type in learning tangible heritage. Using ceramics as an example, we tracked the eye movement of subjects under different display forms and tested whether they obtained the relevant information. Experiment 1 (N = 48) adopted a 2 (integrated/separate display) × 2 (with/without background information) design and Experiment 2 (N = 93) investigated distinct types of annotation (no annotation, indicative/direct/picture/contour shape annotation). We found that the following. (1) In the segmented relic display, the usage time, fixation count, and total fixation duration of relic names were lower than those in the integrated case. The probability that subjects would learn comparatively was also lower in the separate display. However, the performance on retention or transfer tests did not differ depending on the integrated/separate display. After reading the background information, subjects paid less attention to relic names but had better knowledge transfer performance. (2) The viewers' attention allocation to the materials was not significantly influenced by the annotations. Mere visual annotations did not provide an advantage for information acquisition. By contrast, indicative verbal annotation required relatively more time for better target information memory, and the direct verbal cue consumed the least time. Based on the results, we discussed the application scenario of multimedia learning principles and potential recommendations for designing online relic displays. Copyright © 2024 held by the owner/author(s). Publication rights licensed to ACM.

The self-supervised speech representation (S3R) has succeeded in many downstream tasks, such as speaker recognition and voice conversion thanks to its high-level information. Voice conversion (VC) is a task to convert the source speech into a target speaker's voice. Though S3R features effectively encode content and speaker information, spectral features contain low-level acoustic information that is complementary to the S3R. As a result, solely relying on the S3R features for VC may not be optimal. In order to seek speech representation carrying both high-level learned information and low-level spectral details for VC, we proposed a three-level attention to combine Mel-spectrogram (Mel) and S3R, denoted as Mel-S3R. In particular, S3R features are high-level learned representations extracted by a pre-trained network with self-supervised learning. Whereas Mel is the spectral feature representing the acoustic information. Then the proposed Mel-S3R is used as the input of any-to-any VQ-VAE-based VC and the experiments are performed as a downstream task. Objective metrics and subjective listening tests have demonstrated that the proposed Mel-S3R speech representation facilitates the VC framework to achieve robust performance in terms of both speech quality and speaker similarity. © 2023 Elsevier B.V.

近年来,随着空间感知能力和点云大规模处理技术的愈加成熟,基于激光的三维扫描技术已经广泛应用于实时导航、虚拟现实、建筑信息模型建模等各个领域中,而点云分割技术是面向这些应用领域的必要技术之一。点云分割是对点云模型进行有效利用的关键技术,点云分割技术实现了对点云模型的分而治之,无论是识别还是分类都需要先对点云进行分割提取,提高点云分割精度与效率是目前主要的研究方向。本文提出基于图卷积神经网络的点云特征提取网络,使用标签数据的优化方法与语义-实例分割结合的点云分割方法,针对目前深度网络在大场景点云数据上遇到的数据规模大和点云分布不均的问题进行改进。提高了点云分割的准确率,并减少训练中的内存消耗与运行时间。主要研究工作包括:（1）提出了一种基于图卷积神经网络的点云特征提取方法。针对大场景点云无特定顺序并且点云分布不均的特性,应用K临近的邻域采样法,对采样点的距离信息进行排序,计算输入点云的拉普拉斯矩阵,并参考空洞卷积技术对采样法进行了改进,扩大网络的感受域;针对图卷积容易产生梯度消失的问题,本文加入了残差块,提高了网络抗过平滑的能力。（2）提出了一种基于标签传播的点云标签优化方法。针对深度学习对大数据量的需求问题,对实例标签和语义标签进行标签优化,利用得到的优化标签矩阵与原数据集的标记标签矩阵共同影响点云分割的结果。（3）提出了一种点云分割的改进方法。针对点云数据的语义标签与实例标签的相似性,将相似参数结合到语义分割与实例分割的损失函数中,同时进行点云的语义分割与实例分割,通过调整相似性参数占比来影响分割结果。通过在公开数据集S3DIS的点云分割实验证明,针对点云分割问题,语义分割与实例分割相结合能取得点云分割精度上的提升,与Point Net网络和Point Net++网络的分割结果进行了对比。

兵马俑是我国优秀历史文化的重要载体。随着光学感知和计算能力的提升,基于三维扫描的数字化建模广泛应用于兵马俑的保护和展示中。三维扫描构建的兵马俑稠密点云冗余数据多,数据量大,降低了处理、传输和展示的效率,现有点云简化方法往往强调执行效率而忽视了点云特征的保持。本文针对兵马俑点云在应用和展示中对特征保持的要求,提出了一种基于深度学习的三维点云简化方法,通过二维图像特征线提取三维点云的特征点,对非特征点进行均匀化采样,设计并开发了原型系统,在相同简化率上提升了兵马俑的特征保持效果。主要研究工作包括:（1）提出了一种基于深度神经网络2D转3D特征点的提取方法。针对点云简化中易损失关键信息点和繁琐计算开销的问题,采用了一种基于卷积神经网络的VGG16模型,获得图像中的特征线,根据本文提出的二维图像特征线与三维空间特征点的映射关系,提取三维点云的特征点。有效地降低了计算的复杂程度,获得了点云中的特征点。（2）提出了一种自适应的非特征点简化方法。针对简化模型易丢失几何外形和存在空白区域的问题,采取了基于最远点采样与均匀采样的方法,当简化率较高时,采取最远点采样,当简化率较低时,采取均匀采样。实验证明本方法不仅避免了大规模的孔洞,而且保持了应有的几何外形。（3）设计了一套基于特征线的点云简化原型系统。针对兵马俑点云存储和展示需求,实现了点云数据读取、特征点提取和点云简化等功能。实验结果表明,简化系统功能设计合理,能够有效地实现三维点云的特征线提取和简化要求,在降低兵马俑点云数据量的同时较好地保证表面几何特征,提升了点云存储、传输和展示效率。

博物馆网站打破了实体博物馆的时空界限,因此具有比实体博物馆更广泛的用户群体,更加多样化的用户需求,这使得分众成为博物馆网站建设的重要内容。博物馆网站的分众可以理解为馆方根据一定的指标将网站用户分为不同群体,并从不同群体的需求出发,提供相对应的内容与服务。本文通过对博物馆网站分众进行再思考,对现有博物馆网站常见的分众模式进行了总结,分析现有博物馆网站分众模式存在的分类不合理、提供的内容与服务针对性不强等问题,并提出建议,馆方应当结合人工智能领域的相关技术进行分众框架的架构,并不断完善其内容与服务等。

In 3D face analysis research, automated classification to recognize gender and ethnicity has received an increasing amount of attention in recent years. Feature extraction and feature calculation have a fundamental role in the process of classification construction. In particular, the challenge of 3D low-quality face data, including inconsistent mesh numbers or holes, makes it difficult to extract and calculate facial features. To overcome this challenge, we propose a shape descriptor Scaling invariant harmonic wave kernel signature (SIHWKS) that is robust to scaling, topology and sampling and can effectively describe the global and local properties of 3D shapes simultaneously by involving two energy parameters. We extract a local nose region in the center of the face using isogeodesic stripes replacing full facial information, which has a lower probability of occlusion and lower calculation complexity. Actually, the local nose region is constrained by the skull so that it has high distinction gender and ethnicity property and stability property that are robust to 3D facial expression for gender and ethnicity classification. Compared with gender and ethnicity classification based on 2D deep-learning methods influenced by texture information, the proposed method does not require complex processes for model training and only considers the geometric information of the 3D nose region. In addition, to estimate the effectiveness of our point descriptor SIHWKS for gender and ethnicity classification, we compare our SIHWKS with four existing descriptors – global point signature (GPS), heat kernel signature (HKS), wave kernel signature (WKS) and harmonic wave kernel signature (HWKS) – on four databases, namely, FRGC2.0, Bosphorus3D, Facewarehouse and Asian Mongolian craniofacial. Finally, we perform experiments comparing our method with other recent existing classification methods. The experimental results show that our proposed method can achieve a higher accuracy rate for gender and ethnicity classification. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.