Jinan University

Macalester College

Shenzhen Lanyou Technology Co., Ltd.

Feiran Huang

Jian Weng

Zhiquan Liu

Tingrong Zhi

Shuyuan Lin

Zhibo Zhou

Yuanchen Bei

Jingjing Guo

Youqiong Xiong

Jinming Zhong

Zhenghang Yang

Xiaocheng Lin

Junhong Wen

Pengtao Xia

Guan Liu

Zihong Yang

Senting Ma

Yongli Hu

Yijie Zhang

Yong Ma

A traffic anomaly detection method and system based on improved BERT integrating contrastive learning is provided, the method includes: obtaining traffic data and preprocessing the data; building an improved BERT model including an embedding layer and 12 Transformer encoder networks; performing weight sharing operation among a first 6 Transformer encoder networks and a last 6 Transformer encoder networks, respectively; building a classification network; building a total loss function based on a cross-entropy loss and a contrastive loss; performing unsupervised pre-training on the improved Bidirectional Encoder Representations from Transformers, BERT model; fine-tune training the improved BERT model; updating model parameters through backpropagation to obtain a trained improved BERT model; feed test traffic data into the trained improved BERT model; and obtain a traffic detection outcome. The present disclosure enhances the generalization ability of the model while maintaining stability and accuracy.

Traffic anomaly detection method and system based on improved BERT integrating contrastive learning

A full-link data security protection method and a system are provided. The method includes: at a data creation and collection stage: building a data security identification; at a data transmission and storage stage: dividing the ciphertext file into blocks to generate ciphertext components; calculating a virtual index and a data label; transmitting the ciphertext components to a distributed hash table (DHT) network; uploading a tuple including the virtual index, the data block, and the data label to a cloud server; at a data processing and exchange stage: applying re-encryption based on a re-encryption key generation algorithm; performing decryption to obtain the signed identifier and a secret value; acquiring a tuple having a ciphertext component associated with the virtual index. Attribute-based proxy re-encryption is used to achieve fine-grained access control for the cloud storage. In the data destruction stage, the DHT network automatic updating utility is leveraged to realize data self-destructing.

Full-link data security protection method and system

Disclosed is a method for scheduling cars at a narrow curve based on a parking space for temporary avoidance. The method includes: fusing car data obtained by laser radar and millimeter wave radar, so as to predict car states and driving trajectories of meeting cars; computing an expected meeting point of the cars through prediction of the driving trajectories of the cars; and executing, on the basis of the expected meeting point, different car avoidance strategies according to road conditions and a number of parking spaces for temporary avoidance, so as to implement car avoidance and scheduling of car meeting at the narrow curve.

Method for scheduling cars at narrow curve based on parking space for temporary avoidance

The present invention relates to the field of artificial intelligence, and in particular, to a camera pose estimation method and system, electronic equipment and a readable medium. The camera pose estimation method includes: acquiring an initial matching set between a first image and a second image, where the first image and the second image are images from different angles for a same scene; performing a mismatch removal operation on the initial matching set based on an optimization network to obtain an optimized matching set, where the optimization network is constructed based on a multi-stage geometric semantic attention network; and acquiring a camera pose result based on the optimized matching set. By removing mismatches, feature matching results between the first image and the second image are more accurate, and thus a more accurate result can be obtained when performing camera pose estimation.

Camera pose estimation method and system, electronic equipment and readable medium

This invention discloses a multimodal data heterogeneous Transformer-based asset recognition method, system, and device, the method including: collecting various-modal information of an asset, including text information and image information; building an ALBERT model, a ViT model, and a CLIP model; by the ALBERT model, extracting a text information feature; by the ViT model, extracting an image information feature; by the CLIP model, extracting image-text matching information feature; by different channels, applying asset type recognition to information in different modalities; outputting classification information from the different channels; by the CLIP model, generating asset void information; and discriminatively fusing the classification information from the different channels with the matching degree between the image information and the text information obtained by the CLIP model, and outputting final asset class information. This invention realizes comprehensive discrimination by drawing from multiple modalities to improve the accuracy of asset recognition.

Multimodal data heterogeneous transformer-based asset recognition method, system, and device

A multi-source image correspondence method and system based on heterogeneous model fitting is provided, the method includes the following steps: constructing a multi-orientation phase consistency model, fusing phase consistency, image amplitude, and orientation detection feature points, constructing logarithmic polar coordinate descriptors with variable-size bins using sub-region grids and orientation histograms, effectively estimating model parameters through heterogeneous model fitting, accumulating matching pairs from different heterogeneous models that meet a preset joint position offset transformation error, outputting a final matching pair, and completing multi-source image correspondence. The present disclosure alleviate the influence of nonlinear radiation distortion by constructing the multi-orientation phase consistency model, constructing logarithmic polar coordinate descriptors with variable-size bins by sub-region grids and orientation histograms, removing an abnormal matching relationship in multi-source images with the heterogeneous model fitting method, thereby improving the accuracy and robustness of feature detection and improving multi-source image correspondence performance.

Multi-source image correspondence method and system based on heterogeneous model fitting

The embodiment of the invention discloses a diversified recommendation method for news based on graph neural network and its device, which comprises the followings: word segmentation processing is performed on the target news text to obtain a word segmentation set, and vectorization processing is performed on the word segmentation set, so that each word segmentation in the word segmentation set has a word segmentation embedding vector; vectorization processing is performed on a target user group, so that each user in the target user group has a user embedding vector; the word segmentation embedding vector and the user embedding vector are input into the attention mechanism algorithm to obtain the news embedding vector; a graph neural network is constructed according to the user embedding vector and the news embedding vector, and the recommendation model is trained to obtain diversified recommendation results. In the embodiment, firstly, a word segmentation tool is used to segment each news text, and then the embedding vector of news is obtained by using attention mechanism, user interest and word segmentation information, and the corresponding expression is obtained by graph neural network, and the model is optimized, and finally the diversified recommendation results are obtained by calculation.

Diversified recommendation method for news based on graph neural network and its device

A specific target-oriented social media tweet sentiment analysis method includes: preprocessing social media tweet data to obtain a target text and a specific target; passing the target text and the specific target through an embedding layer to obtain target text word vectors and obtain a specific target word vector, respectively; passing the target text word vectors through a self-attention structure to obtain a self-attention result; combining the self-attention result with the specific target word vector and passing through a cross-attention structure to obtain cross-attention results; concatenating the cross-attention results to obtain an attention representation matrix; and passing the attention representation matrix through a pooling layer, a fully connected layer and a softmax layer to obtain a sentiment tendency of the specific target. The method, based on an attention mechanism in the transformer structure, fuses a local attention mechanism and an attention mechanism containing position information, the analysis accuracy is improved.

Specific target-oriented social media tweet sentiment analysis method

A text sentiment analysis method based on multi-level graph pooling includes steps of: preprocessing a target text; taking collocate point mutual information between word nodes as an edge weight between the word nodes, and building a graph for each text independently; constructing a multi-level graph pooling model, of which a gated graph neural network layer transfers low-level information, a first graph self-attention pooling layer performs an initial graph pooling operation and uses a Readout function to extract low-level features, a second graph self-attention pooling layer performs a graph pooling operation again, performs a pruning update on the graph structure by calculating attention scores of nodes in the graph and uses a Readout function to extract high-level features; obtaining a multi-level final vector representation through a feature fusion function; and selecting a sentiment category corresponding to a maximum probability value as a final sentiment category output of the text.

Text sentiment analysis method based on multi-level graph pooling

A social recommendation method based on a multi-feature heterogeneous graph neural network is provided and includes: extracting attribute information of users and topics to code; processing user coding information and topic coding information through a multi-layer perceptron to obtain initial feature vectors of the users and the topics; establishing a heterogeneous graph by taking the users and the topics as nodes; inputting the heterogeneous graph into a heterogeneous graph neural network to perform information transmission in combination with an attention mechanism, and updating the feature vectors; and performing similarity calculation on the feature vectors of the users, and selecting the user and the topic with the highest similarity with the feature vector of the user for recommendation. Social information can be mined more comprehensively, features of users and interested topics of the users can be deeply fused, and recommendation accuracy and user experience can be improved.

Growing community of inventors

Guangzhou, China

Feiran Huang