切换中文
Latest Issue
Online First List of Issues Document Types
Vol. 23 ， Issue 9 ， 2024
-
Review of Named Entity Recognition Research AI Introduction
“在自然语言处理领域，命名实体识别技术取得新进展，专家综述了传统与深度学习方法，为未来研究指明方向。”Abstract：Named entity recognition is an important foundational tool in application fields such as machine translation and question answering systems， and has always been a research hotspot in the field of natural language processing. Firstly， introduce the definition of named entity recognition， and organize the commonly used implementation tools， datasets， and evaluation criteria in named entity recognition tasks； Then， based on their development history， existing named entity recognition methods are summarized and divided into traditional methods and deep learning methods； Next， summarize the key ideas， advantages， and disadvantages of various methods for named entity recognition， provide the main process of deep learning based named entity methods， and summarize them in the order of the process； Finally， looking forward to the future research directions of named entity recognition， providing ideas for subsequent research.Keywords：natural language processing;named entity recognition;machine learning;deep learning- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在图像生成领域，生成对抗网络通过博弈提升图像生成质量，专家梳理了其应用方法，为图像生成研究提供新思路。”Abstract：Since its proposal in 2014， generative adversarial networks have greatly promoted the progress of image generation research. Through the mutual game between two neural networks， it gradually improves the ability to distinguish between real images and generate images， as well as the ability to generate realistic images， ultimately achieving a Nash equilibrium between the two parties. Briefly introduce the generation of adversarial networks and sort out their application methods in the field of image generation around the issue of generating images containing targets. They are divided into five categories： direct method， iterative method， hierarchical method， decoupling method， and 3D modeling method. Focus on the research progress of generative adversarial networks in generating images containing targets， and prospect the development direction of image generation methods containing objects， in order to provide reference for relevant researchers in image generation research.Keywords：image generation;generative adversarial network;target image;decoupling;artificial intelligence
- 54
- |
- 4
- |
- 0
Published：2024-09-25 -
Review of Dialogue Systems Based on Deep Neural Networks AI Introduction
“在自然语言处理领域，深度学习技术推动了对话系统研究的显著进步。专家梳理了关键技术，分析了模型应用，指出问题并提出解决方案。”Abstract：With the rise of deep learning technologies， significant advancements have been achieved in the field of natural language processing（NLP）， particularly in the domain of dialogue systems. This paper begins by providing an overview of the fundamental processes involved in dialogue systems. Subsequently， it comprehensively reviews deep learning-based techniques for dialogue systems， encompassing three key categories： convolutional neural network（CNN）， recurrent neural network（RNN）， and attention mechanism（AM）. The paper introduces the principles of these models， and then provides an in-depth analysis and comparison of the applications， characteristics， and advantages and disadvantages of various basic models and their derivative models in dialogue tasks form the perspectives of information extraction， dialogue state tracking， and dialogue generation. Finally， this paper enumerates persisting challenges within dialogue tasks， and proposes feasible solutions.Keywords：deep learning;natural language processing;attention mechanism;dialogue system;neural network- 13
- |
- 1
- |
- 0
Published：2024-09-25 -
Detection of Fake News on Social Media： Basic Theories， Methods and Research Prospects AI Introduction
“在社交媒体领域，假新闻检测技术取得突破，为维护信息真实性提供新思路。”Abstract：The rapid development of social media platforms not only greatly enhances the accessibility of information， but also accelerates the spread of fake news. The explosive growth of fake news not only damages social stability， but also erodes public trust in the media. In the field of natural language processing， fake news detection is a crucial and challenging task. To this end， first provide a definition of fake news and analyze its characteristics in depth； Then， analyze and evaluate the existing fake news detection methods from four perspectives： news content， social context， communication networks， and hybrid methods. Introduce the performance of relevant models， commonly used datasets， and evaluation indicators； Finally， summarize and analyze the current problems in fake news detection research， and propose possible future research directions.Keywords：social network;fake news;natural language processing;early detection;explainability- 11
- |
- 3
- |
- 0
Published：2024-09-25 - “在推荐系统领域，研究者提出了LFDMF模型，通过LSTM和深度矩阵分解，有效提升了推荐准确性。”Abstract：Aiming at the problem that most recommendation algorithms in real recommendation scenarios ignore the timeliness factors of dynamic changes in user preferences， resulting in limited model performance， a recommendation fusion model based on LSTM and deep matrix factorization （Long Short-Term Memory Fusion Deep Matrix Factorization， LFDMF） is proposed. The model uses generalized matrix factorization to learn nonlinear low-order features between users and items， uses multi-layer perceptron to learn nonlinear high-order features between users and items， and obtains users ' long-term dynamic preferences. LSTM 's strong fitting ability to time series is used to obtain users ' short-term dynamic preferences. In order to verify the effectiveness and feasibility of the LFDMF model， comparative experiments are carried out on the public datasets MovieLens-1M and Pinterest. The simulation results show that the HR @ 10 and NDCG @ 10 indexes of the LFDMF model are improved by 0.103 4 and 0.132 2，0.118 1 and 0.101 8， respectively， compared with the traditional MF algorithm. Compared with the DMF model， it is improved by 0.022 8 and 0.032 3， 0.016 9 and 0.013 5， respectively，the recommendation performance is significantly improved.Keywords：recommendation fusion;generalized matrix factorization;multilayer perceptron;skip connections;long short-term memory
- 7
- |
- 0
- |
- 0
Published：2024-09-25 -
Steganography Model and System for Automatic Generated Poem AI Introduction
“在信息安全领域，中国古诗词的自动生成技术取得突破，专家设计了基于BERT和Attention机制的隐写算法，为信息隐写提供新方案。”Abstract：Chinese ancient poetry has the characteristics of neat structure， rich emotions， and appropriate length， making it suitable as a natural carrier for generative information hiding. However， due to its implicit semantics， profound meaning， and diverse styles， the automatic generation of poetry hiding is extremely challenging. With the emergence of big language generation models， the quality of poetry generation has been greatly improved， but there have been no reports on applying big language models to implicit writing in poetry generation. To this end， the BERT sentiment analysis model was introduced and combined with the Attention mechanism to design a hidden writing algorithm and model for Seq2Seq to automatically generate poetry. The corresponding hardware system and GUI interface were constructed based on PN40. Under the guidance of theme words and the constraints of rhythm/emotion/mutual information， multi-mode implicit poetry was generated and quickly output on the system. The experimental results show that the proposed model generates implicit poems with clear themes and consistent emotions. The BLEU evaluation value is as high as 44.3%， and the average accuracy of sentiment analysis is above 85%， greatly enhancing the perception and statistical concealment of implicit poems and accelerating the application of generative information steganography.Keywords：poetry;large language model;sentiment analysis;concealability;information steganography- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Study on Knowledge Graph Question Answering Methods Incorporating External Knowledge AI Introduction
“在自然语言处理领域，DEK-KGQA方法通过融合外部知识，提升了问答推理能力，实验显示其在常识问答任务中效果显著。”Abstract：Knowledge graph question answering is one of the hot research areas in the field of natural language processing. Existing methods face two main challenges： difficulty in understanding complex natural language questions and limited semantic interpretation of entity representations. To address these challenges， a knowledge graph question answering method called DEK-KGQA is proposed， which integrates external knowledge. First， a joint graph is constructed by combining the question knowledge graph subgraph and the QA context. Then， the relevance scores of nodes in the joint graph are calculated using pre-trained language models. Finally， external knowledge is introduced to enhance information interaction and reasoning ability during the question answering process. Experimental validation is conducted on the CommonsenseQA dataset， comparing the proposed method with existing methods. The results demonstrate that the proposed method achieves better performance in commonsense question answering tasks， validating its effectiveness. In addition， ablation experiments are conducted to evaluate the impact of each component on the overall performance.Keywords：knowledge graph question answering;QA context;pre-trained language model;external knowledge- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Distribution Network Line Loss Analysis Based on Autoencoder and Time Domain Convolutional Neural Network Algorithm AI Introduction
“在配电网线损预测领域，研究人员提出了一种基于TCN-BiGRU的改进算法，有效提升了线损预测的准确性和实时性，为电网节能降损提供了科学依据。”Abstract：In the complex distribution network environment， there are problems of insufficient accuracy and real-time calculation of line loss. This paper proposes an improved TCN-BiGRU distribution line loss prediction method based on recurrent neural network autoencoder. The TCN neural network model， which is good at processing time series， was selected as the backbone feature extraction network， and the BiGRU unit was integrated into the TCN to effectively solve the problem of gradient vanishing. On this basis， combined with the recurrent neural network autoencoder， the line loss outliers were classified and labeled unsupervised， and the causes of the line loss rate anomalies were predicted through the softmax loss function and the corresponding loss reduction measures were formulated. The experimental results show that compared with the traditional distribution network line loss prediction method， the root mean square error of the proposed line loss prediction method is reduced by 0.036 99 and 0.004 02 compared with the traditional EMD-LSTM and PSO-CNN algorithms， respectively， and the accuracy of line loss cause analysis is improved by 1.500% and 5.841% compared with the ResNet50 and DBN-DNN algorithms， respectively. It provides a scientific reference for energy saving and loss reduction in the distribution network after the distributed power generation is connected and the realization of the dual carbon goal of the power grid.Keywords：recurrent neural network autoencoder;TCN-BiGRU line loss prediction algorithm;smart grid;analysis of the causes of abnormal line loss;prediction of line loss in the substation area- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在环境监测领域，研究者利用随机森林算法和长短期记忆网络，建立了蓝藻水华预测模型，显著提高了预测精度。”Abstract：In order to monitor and analyze the constraining effect of meteorological factors on the outbreak of blue-green algae blooms， establish the response relationship between blue-green algae blooms and meteorological factors， and predict changes in blue-green algae biomass and distribution. Quantitatively evaluate the importance and contribution rate of meteorological factors through random forest algorithm， select meteorological factors， and construct a blue-green algae bloom prediction model using long short-term memory network. Selecting the annual values of normalized vegetation index in Dianchi Lake from 2000 to 2021 and meteorological data from Kunming Station during the same period as the research objects， this study explores the feasibility of using the long short-term memory network RF-LSTM based on the random forest algorithm to predict blue-green algae blooms， with normalized vegetation index as the indicator of changes in blue-green algae blooms. The results showed that compared with the single structure long short-term memory network model and the single algorithm random forest model， the RF-LSTM model's annual value prediction passed the 0.01 significance test， and the simulation accuracy reached 90.9%. The random forest algorithm is beneficial for understanding the relationship between blue-green algae blooms and meteorological factors， selecting meteorological factors with predictive performance， and thus improving the predictive ability of long short-term memory network models.Keywords：cyanobacterial blooms;meteorological factor;random forest;long and short term memory network;Dianchi Lake
- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Research on Low-Resource Speech Recognition Based on ADTDNN AI Introduction
“在语音识别领域，研究者提出了一种新方法，通过融合Attention机制与DTDNN，有效提升了低资源环境下的识别精度。”Abstract：A speech recognition approach has been proposed to address the problem of reduced recognition accuracy and poorer generalization performance due to insufficient training data in low-resource conditions. This method leverages convolutional neural networks to extract feature information. It combines the attention mechanism with delayed time-delay neural networks， referred to as ADTDNN， enhancing the model's ability to capture key information in sequences within low-resource environments. The approach employs linking temporal classification to streamline the recognition process of the model. Additionally， a Transformer is utilized as the language model. Experimental results on the Aishell-1 dataset demonstrate that the ADTDNN-based speech recognition model in low-resource settings reduces word error rates by 3.7% and 1% compared to mainstream end-to-end models like LAS and Transformer， respectively.Keywords：speech recognition;time delay neural networks;Transformer;data enhancement;low resource- 10
- |
- 0
- |
- 0
Published：2024-09-25 -
Open-Domain Question Answering Task Enhanced by Multiple Documents Refinement and Fusion AI Introduction
“在自然语言处理领域，研究者提出了RFMD方法，通过多文档精排与融合，有效提升了开放域问答的准确性。”Abstract：Open-domain question answering （OpenQA） is a challenging task in natural language processing，the conventional machine learning and deep learning techniques are commonly used to retrieve many candidate document fragments related to the question from the raw corpus for answer extraction. However， the candidate document fragments retrieved by current methods tend to include considerable noise and irrelevant information to the question， and the previous OpenQA model falls short in accurately responding to questions that necessitate multiple document fragments as correlative evidence. Therefore， this paper proposes an open-domain question answering method based on refinement and fusion of multiple documents（RFMD）. Specifically， RFMD designs a Transformer-based document refinement module during the retrieval stage to reduce noise information in the candidate documents. In the reading comprehension stage， RFMD employs a text generation-focused question answering module. By constructing a global attention mechanism across document fragments， it integrates information from multiple relevant document fragments to accurately answer questions that require multiple document fragments as supporting evidence. RFMD achieves EM scores of 45.8% and 63.4% on the NaturalQuestions and TriviaQA datasets respectively， verifying the effectiveness and superiority of the model in OpenQA tasks.Keywords：open-domain question answering;pre-training model;generative model;similarity score;Prompt design- 7
- |
- 0
- |
- 0
Published：2024-09-25 - “在电网行业，一项新研究通过多模态心理评测算法，显著提升了作业前心理状态的评估准确性，准确率提高至65.66%。”Abstract：To solve the intelligent psychological evaluation task for high-risk industry employees in the power grid before work， a multimodal psychological evaluation algorithm is proposed for employees in the power grid industry， which includes expressions， sounds， and walking posture. Firstly， construct a dataset of employees in the power grid industry， extracting facial RGB image sequences， audio ComparE feature sets， and human skeleton keypoint sequences from videos； Secondly， residual networks and bidirectional long short-term memory networks are used to extract facial visual features， audio features are extracted in time windows， and gait features are extracted in spatiotemporal graph convolutional networks， respectively， to obtain the optimal single modal models； Finally， a deep learning training method for polarity loss function and a multimodal fusion algorithm based on attention mechanism are proposed to obtain the optimal multimodal psychological state evaluation model by fusing the output features of a single modal model. The experiment shows that multimodal fusion can significantly improve the accuracy of psychological evaluation compared with single-mode system， and the accuracy rate of the four classification tasks of psychological labels reaches 65.66%. Compared with the model based on facial expression， voice， and gait， the effect of multimodal fusion is increased by 18.04%， 21.22%， and 13.28% respectively.Keywords：multimodality;deep learning;attention mechanism;psychological evaluation;power grid industry
- 15
- |
- 1
- |
- 0
Published：2024-09-25 -
A Fault Diagnosis and Prediction System for Rotating Machinery Based on Self-Attention Encoding and Decoding Structure AI Introduction
“在旋转机械故障诊断领域，专家构建了基于自注意力编解码结构的在线诊断和预测系统，为优化维护计划提供解决方案。”Abstract：Due to harsh operating conditions and high load requirements， rotating machinery faults can lead to high maintenance costs and unnecessary downtime. It is necessary to develop an efficient and accurate rotating machinery fault online diagnosis and prediction system to help enterprises quickly identify faults， predict future events， and optimize maintenance plans. The construction state matrix represents the vibration signal generated by rotating machinery， that is， a continuous time series is divided into several Windows， and each window is converted into an image. Image features are extracted and processed by a series codec of a specific structure to classify vibration patterns in a training data set. The reliability and effectiveness of the online fault diagnosis and prediction system for rotating machinery based on self-attention codec structure are verified by simulation experiments. The state characteristic database of rotating machinery can accurately diagnose and predict the faults of rotating machinery. The system can help businesses optimize maintenance schedules and reduce downtime and maintenance costs.Keywords：fault diagnosis;rotating machinery;Transformer;convolutional neural network- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Educational Named Entity Recognition Integrating Word Information and Self-Attention Mechanism AI Introduction
“在教育领域，WBBAC模型通过融合词信息与自注意力，显著提升了命名实体识别的精度，有效应对了语料不足的挑战。”Abstract：To solve the problems of low accuracy and severe lack of corpus in named entity recognition tasks in the education field， a named entity recognition model WBBAC that integrates word information and self attention is proposed. This model utilizes a BERT pre trained language model to enhance the semantic representation of word vectors and introduces word frequency information into them. The word vectors are concatenated with each other as inputs to a bidirectional long short-term memory network， which further searches for internal connections within the sequence through a self attention layer. Finally， the optimal sequence is obtained through CRF decoding. Create a computer composition principle dataset based on the characteristics of the course text and annotate it. Conduct experiments on the Resume dataset and the computer composition principle dataset， and the F1 values of the WBBAC model are 95.65% and 73.94%， respectively. The experimental results show that compared with the baseline model， the WBBAC model has a higher F1 value， effectively solving the problem of insufficient annotated data in named entity recognition tasks in the education field.Keywords：named entity recognition;word information;self-attention mechanism;education domain;BERT- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在短文本分类领域，研究者基于预训练语言模型，提出了一种新的标签感知注意力方法，有效提升了分类精度。”Abstract：Aiming at the problem that current short text classification only uses classification labels as the basis for judging classification results， and ignores the semantic information contained in the classified label text， a short text classification method based on label aware attention is proposed， which is based on a large-scale pre trained language model. This method represents text data in distributed vector form through large-scale pre trained language models to obtain richer semantic information； At the same time， incorporating classification label information into the text data training process， using attention mechanisms to make the text data perceive and classify the most relevant information； Using CNN networks and max pooling layers to extract local word level vector features， in order to better address semantic issues such as double negation and comparative negation in English texts； Using residual connections to fuse sentence level vectors with word level vectors effectively alleviates the problem of text information decay. Tests were conducted on three common English datasets， R8， R52， and MR， and the experimental results showed that the proposed method achieved accuracies of 98.51% and 97.10% on the R8 and R52 datasets， respectively， which are better than DeBERTa and BertGCN.Keywords：short text classification;CNN;label awareness;attention;pre-trained
- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在数据科学领域，PSON算法通过定义少数类样本影响力，显著提升了分类准确率。”Abstract：Oversampling is a commonly used method to solve the problem of imbalanced class distribution in a dataset by synthesizing new samples of the same class. A PSON algorithm based on neighborhood concept is proposed to address the issue of imbalanced sample distribution in the dataset. This algorithm defines the influence of each minority class sample and oversamples the minority class samples based on different influences to obtain a balanced dataset. Classification tests were conducted on datasets obtained from 8 oversampling algorithms on 50 datasets. The Wilcoxon symbol rank test was used to compare 7 classification performance indicators， and the results showed that the use of PSON algorithm significantly improved classification accuracy.Keywords：imbalanced dataset;over-sampling;classification;reverse neighbors
- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Research on Optimization Problem of Agricultural Supplies Distribution Based on Learning Algorithm AI Introduction
“在农资供应链优化领域，专家建立了DCVRPTW模型，探索了DRL-SIA算法，验证了其在降低物流成本方面的有效性。”Abstract：A mathematical model for vehicle path planning （DCVRPTW） was established to optimize the delivery path of agricultural inputs supply chain orders， taking into account constraints such as the comprehensive range， maximum load capacity， and time window of new energy trucks. The model comprehensively optimizes the fixed and transportation costs of vehicles， and proposes a swarm intelligence optimization algorithm framework based on deep reinforcement learning （DRL-SIA）. An intelligent agent is a decision-maker who selects the best action from the action pool based on the environmental state as input to change the environment and obtain environmental rewards. The DRL-SIA algorithm combines trained agents with swarm intelligence algorithms to replace the original algorithm for decision selection， thereby improving optimization speed and accuracy. The experiment shows that the optimal solution of the proposed algorithm is superior to other algorithms in all cases， verifying that the algorithm can effectively reduce logistics transportation costs in the agricultural material supply chain.Keywords：deep reinforcement learning;vehicle routing planning;swarm intelligence optimization algorithm;agricultural supply chain;evolutionary computation- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在物流拣选领域，研究者构建了多人协同拣选模型，采用改进算法优化拣选效率，为提升拣选速度提供新方案。”Abstract：Aiming at a variety of small picking problems of parts， in the multi-person collaborative picking mode， the task allocation is unreasonable， the picking time varies greatly， and the picking link is easy to timeout. A multi-person collaborative picking model aiming at the shortest picking time is constructed， and the improved K-means algorithm and genetic algorithm are used to solve the model. Aiming at the shortcomings of the traditional K-means algorithm clustering results， the number of picking points contained in each cluster varies greatly. The picking time of each cluster is used as an index to transform the cluster where the picking points belong. The genetic algorithm is used to perform path planning and picking time calculation on the clustering results to obtain the optimal clustering results. Taking the parts picking process of a security equipment manufacturing enterprise as the research object， the effectiveness of the algorithm is verified by comparing with the picking time obtained by simple batching.Keywords：task allocation;K-means algorithm;genetic algorithm;path planning
- 7
- |
- 0
- |
- 0
Published：2024-09-25 - “在食品工业领域，专家利用支持向量机算法和R语言，建立了葡萄酒质量检测模型，精度高达96.46%，为质量控制提供新途径。”Abstract：With the continuous development of our country's economy， wine， as a delightful medium of life， has entered the public dining table. However， the quality inspection of wine is still mainly based on the tasting of wine tasters， which can no longer meet the needs of the large-scale and intelligent development of the food industry. Therefore， based on the support vector machine algorithm， the physicochemical indicators of wine were modeled， and the Box plot method was implemented using R language to handle outliers. At the same time， the support vector machine parameters of the RBF kernel were optimized， resulting in a wine quality detection model with an accuracy of 96.46%. This provides an effective approach for wine quality control.Keywords：support vector machine;R language;Box-plot method;wine quality testing
- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Safe Reinforcement Learning Method Based on Robust Cross-Entropy and Gradient Optimization AI Introduction
“在智能体安全执行任务领域，研究者提出了一种新算法，通过模型预测控制框架和优化方法，有效提升了智能体的安全性和效率。”Abstract：Ensuring security and efficiency when intelligent agents perform tasks in complex environments is a major challenge. Traditional reinforcement learning methods use model free reinforcement learning to solve intelligent decision-making problems， constantly trial and error to find the optimal strategy using a large amount of data， ignoring the training cost and security risks of the agent， and therefore cannot effectively ensure the safety of decision-making. To this end， safety constraints are added to the actions of intelligent agents in the model predictive control framework， and a safety reinforcement learning algorithm is designed to obtain the safest action control sequence. At the same time， in response to the problems of high computational complexity and low efficiency in the cross entropy method， as well as the problem of falling into local optima in the gradient optimization method， a combination of robust cross entropy and gradient optimization methods is used to optimize the action control sequence to improve algorithm safety and solving efficiency. The experiment shows that the proposed method can effectively improve the convergence speed compared to the robust cross entropy method， and has the best safety performance compared to other optimization algorithms without sacrificing much performance.Keywords：reinforcement learning;robust cross-entropy;gradient optimization;safety- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Research on Improved Moving Edge Computing Task Unloading Algorithm Based on Deep Reinforcement Learning AI Introduction
“在大数据时代，移动终端用户激增，万物互联带来便利，但也引发数据地理位置分散问题，影响服务质量。研究者构建移动边缘计算平台任务卸载模型，采用深度强化学习算法优化策略，实验显示改进算法在能耗、时延、网络使用量方面表现更优。”Abstract：In the era of big data， the scale of mobile terminal users continues to expand， and the Internet of everything brings great convenience to people. At the same time， there is also the problem of geographic dispersion of a large amount of data， which brings great challenges to the QoS of user service. In this paper， a task unloading model based on the three-layer service architecture of the mobile edge computing platform is first built. Combined with the actual application scenario of the MEC platform， the deep reinforcement learning algorithm is improved by using the same policy experience playback and entropy regularization， and the task unloading strategy of the MEC platform is optimized. Experiments are designed to compare and analyze the three indexes of energy consumption， delay and network usage of the three traditional algorithms and the improved algorithm， and verify that the improved algorithm has better performance in reducing energy consumption， delay and network usage.Keywords：deep reinforcement learning;edge computing task offloading;same strategy experience replay;entropy regularity- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Research on Tobacco Retail Store Regulatory Route Optimization Algorithm Based on Graph Attention Mechanism for Node Selection AI Introduction
“在城市烟草监管领域，最新研究提出了GA-SGPO算法，有效提升了监管路径规划的效率和准确性。”Abstract：Since the tobacco retail stores in cities are dense， traditional path planning algorithms for solving the optimal supervision path will consume a lot of time， and cannot guarantee the effect within the specified time. In addition， existing methods seldom consider the network characteristics and the explainability of the candidate subset.This study proposes a graph attention-based node selection and path optimization algorithm （GA-SGPO）， which iteratively selects the optimal coordinate node subset and performs calculation on the subset to reduce computation time. In addition， the structural similarity between nodes is calculated to reduce the sparsity of training samples.The experimental data includes the coordinates of 40，000 retail stores in Dongguan City. The experimental results show that the GA-SGPO model ensures the solution accuracy while the solution time is reduced by an average of 48%.The GA-SGPO can significantly save computational time and is closer to practical application scenarios. The attention mechanism and node similarity calculation can provide visualization basis for optimal node selection.Keywords：graph attention;shortest path optimization;tobacco regulation;node structure similarity;node selection- 13
- |
- 0
- |
- 0
Published：2024-09-25 -
Research on Personal Credit Risk Assessment Based on Improved Active Generative Oversampling AI Introduction
“在个人信用风险评估领域，专家提出了一种改进的ACGAN模型，通过Wasserstein距离和Focal loss增强分类器性能，显著提升了F1、AUC及G-means指标。”Abstract：Aiming at the problems of sample imbalance and category overlap in personal credit risk assessment， an improved active generative oversampling model is proposed. Firstly， based on the auxiliary classifier generative adversarial network （ACGAN） framework， Wasserstein distance is introduced to improve the true false discrimination loss function， and gradient penalty is added to prevent pattern collapse； Secondly， Focal loss is used instead of traditional cross entropy loss to enhance the ability to identify difficult samples； Finally， the proposed model is used to oversample imbalanced data to improve classifier performance. Experiments on real credit data show that the model improves the classifier's classification performance indicators F1， AUC， and G-means by 11.2%， 1.7%， and 12.8%， respectively. It achieves significant results in enhancing sample diversity， reducing class overlap， and improving the classifier's classification performance on imbalanced datasets.Keywords：deep learning;unbalanced data;class overlap;ACGAN;Focal Loss;Wasserstein distance- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Office Smoking Behavior Detection Based on Improved YOLOv5 Model AI Introduction
“最新研究突破：YOLOv5s+模型在办公室吸烟行为检测中，平均精度提升至81.8%，显著优于原YOLOv5。”Abstract：To solve the problems of low accuracy and easy false detection of small targets in current smoking behavior detection， an improved YOLOv5 recognition model YOLOv5s+is proposed. This model combines the backbone network of YOLOv5 with BoTNet to improve the feature extraction ability of the model， enabling it to detect smaller target objects； At the same time， the feature fusion part is improved by applying a weighted bidirectional feature pyramid BiFPN in the neck of the network model to efficiently fuse shallow position information and deep high-level semantic information， effectively improving detection accuracy. Integrate publicly available online datasets and self-made datasets into an office smoking experimental dataset， and compare the detection performance of the YOLOv5s+ model with the original YOLOv5 model on this dataset. The experimental results show that the average accuracy （mAP） of the improved model YOLOv5s+ is 81.8%， with an accuracy of 82.8% and a recall rate of 83.9%. Compared with the original model， it has improved by 5.4%， 4.1%， and 6.4%， respectively， and has achieved good detection of office smoking behavior.Keywords：deep learning;YOLOv5;smoking detection;feature fusion- 22
- |
- 6
- |
- 0
Published：2024-09-25 -
Human Action Recognition Method Based on Graph Attention Network and Directed Graph Neural Network AI Introduction
“在人体动作识别领域，研究者引入图注意网络和有向图卷积，有效提升了动作识别精度。”Abstract：The graph convolutional neural network based on human skeleton data is not easily affected by background environmental noise and has strong robustness， which has become a research focus in the field of human action recognition at present. However， this network assigns the same weight to different neighborhoods in the same order， which limits its ability to capture spatial information correlations. To this end， a graph attention network weighted sum is introduced to sum the features of adjacent nodes， allowing each node to assign different weights based on its adjacent features to enhance feature extraction and learning effectiveness. At the same time， in order to solve the problem of representing the skeleton as an undirected graph， only the relationship between adjacent nodes or edges can be determined， which limits the ability to capture dependency relationships between nodes or edges. Introducing directed graph convolution， utilizing the feature information of first-order and second-order adjacent nodes for graph convolution， not only preserves the directional features of the directed graph， but also expands the perceptual domain of graph convolution， thereby extracting more features. The experiment shows that the proposed method can effectively improve the accuracy of action recognition.Keywords：action recognition;graph neural network;graph attention;directed graph- 13
- |
- 0
- |
- 0
Published：2024-09-25 -
Three-Dimensional Reconstruction of Radar Weather Data Based on Improved Marching Cubes Algorithm AI Introduction
“在气象领域，雷达数据三维重建技术取得突破，NBV-MC算法大幅提升重建效率，为气象数据分析提供新工具。”Abstract：In the field of meteorology， the meteorological data detected by Doppler weather radar is stored in a three-dimensional polar coordinate system with the radar station as the origin， which has characteristics such as irregular shape and large data volume. The Marching Cubes （MC） algorithm is a classic algorithm in 3D reconstruction， but it has the disadvantages of low reconstruction efficiency and inability to directly process meteorological data when applied in the meteorological field. In order to achieve three-dimensional reconstruction of meteorological data， a radar data normalization and state labeling discrimination algorithm NBV-MC is proposed based on the MC algorithm. This algorithm normalizes and preprocesses radar based data files based on their characteristics， constructs trapezoidal hexahedral voxels using radar based data， marks the state of each hexahedral voxel， and dynamically determines whether it needs to be processed when traversing hexahedral voxels. The experimental results show that the NBV-MC algorithm not only solves the problem of being unable to be directly used in the MC algorithm due to the irregularity of meteorological data， but also effectively reduces the number of triangles required to draw contour surfaces while ensuring data authenticity and reconstruction efficiency， and improves reconstruction speed. Compared with the MC algorithm， the reconstruction efficiency of NBV-MC algorithm has increased by more than 77.70%， which is conducive to real-time scene interaction and facilitates meteorological researchers to directly analyze radar data.Keywords：Doppler weather radar data;three-dimensional reconstruction;Marching Cubes algorithm- 24
- |
- 2
- |
- 0
Published：2024-09-25 -
Composite HOG Feature Clustering Method for Extracting Key Frames of Multi-Scene Video AI Introduction
“在视频处理领域，研究者提出了一种基于复合HOG特征的聚类方法，有效提升了多场景视频关键帧提取的效率和准确性。”Abstract：Due to the fact that extracting dynamic multi scene video keyframes directly from frame difference data often results in excessive redundant frames， the directional gradient histogram （HOG） feature has good stability for image brightness and scene changes. Therefore， a composite HOG feature clustering method for extracting keyframes from multi scene videos has been proposed to improve the efficiency of keyframe extraction. Firstly， by extracting the HOG features of video frames and introducing image information entropy， a composite feature vector is constructed to maintain the correlation of data features. Secondly， based on the composite feature vector， the difference data between video frames is calculated to determine the number of video segmentation shots and keyframe extractions； Again， considering both the intra shot frame set and the complete video frame set， select the video frames with high information entropy as the initial clustering centers without repetition to guide the clustering algorithm search direction， and extract video keyframes through K-means clustering. Compared with the traditional K-means clustering method， it was found that the proposed algorithm reduces redundancy by 0.003~0.015， improves precision by 0.14~0.21， reduces clustering time， and has better accuracy and efficiency.Keywords：key frame extraction;video segmentation;HOG feature;composite feature vector;K-means clustering;image entropy- 0
- |
- 0
- |
- 0
Published：2024-09-25 -
Silage Maize Leaf Blight Detection Model in Field Environment AI Introduction
“在农业科技领域，研究人员开发了YOLOv7-MLD模型，实现了青贮玉米枯叶病的精准检测，提高了病害识别的效率和准确性。”Abstract：In order to achieve accurate Detection of Leaf blight of silage Maize， reduce the cost of manual diagnosis in field environment and reduce the impact of the disease， a modified YOLOv7-MLD（Maize Leaf-Blight Detection） model was proposed. Firstly， Diverse Branch Block （DBB） module was added to the backbone of YOLOv7 network to enhance its feature extraction capability. Then a Coordinate Attention module is added to the three output feature layers to enhance the ability of extracting disease features. Finally， the loss function is replaced by CIoU with SIoU to improve the convergence speed and regression accuracy of the bounding box. Experiments were carried out on a subset of maize leaf wilt disease data set， and the results showed that the AP value of YOLOV7-MLD model reached 84.2%， the F1 value increased by 5.9%， the accuracy rate and recall rate increased by 4.3% and 7.3%， respectively， compared with the original YOLOv7. The model can accurately locate and identify the leaf blight of silage maize in the complex field environment， and has very important practical significance for guiding the prevention and control of the early leaf blight of silage maize.Keywords：silage maize;leaf blight;YOLOv7-MLD;object detection;attention mechanism;smart agriculture- 0
- |
- 0
- |
- 0
Published：2024-09-25 - “在遥感影像语义分割领域，专家提出了基于Transformer的端到端图像级标签弱监督网络，有效提升了分割精度和鲁棒性。”Abstract：A Transformer based end-to-end image level weakly supervised semantic segmentation network is proposed to address the complex scene and high annotation cost of remote sensing image semantic segmentation tasks. The network first improves the accuracy and granularity of the class activation map through a multi class label encoding module； Then， the affinity pseudo label generation module is used to further refine the representation of affinity relationships， generating high-precision affinity pseudo labels as segmentation supervision information， thereby improving the ability of weakly supervised networks； Simultaneously designing a mixed label data augmentation module to enhance the composition of remote sensing data； Finally， a mixed loss function with fusion affinity loss is provided to enhance the learning performance of the network. The experimental results on the ISAID dataset show that the model achieves an mIoU of 38.836% in segmentation results using image level labels， demonstrating better robustness and reliability compared to the control network. It has high application value in weakly supervised semantic segmentation of remote sensing images.Keywords：remote sensing image;weakly-supervised semantic segmentation;image-level labels;Transformer
- 16
- |
- 0
- |
- 0
Published：2024-09-25 -
Teaching Exploration and Practice of Ideological and Political Construction in the Course of Data Structure for Information Majors AI Introduction
“在信息类专业数据结构课程中，融入思政元素，提升学生专业理解和学习热情。”Abstract：In order to solve the problems of "hard integration" and "superficiality" in the ideological and political construction of data structure courses in information majors， in accordance with the requirements for the cultivation of engineering students in the "Guidelines for Ideological and Political Construction of Higher Education Curriculum"， we will deeply explore the ideological and political elements related to data structures from the perspectives of engineering ethics education， national craftsmanship spirit， patriotism， and mission responsibility， and integrate them into professional course teaching. The practical results show that after the reform of ideological and political education in the curriculum， students have a clearer grasp of professional knowledge points， a stronger interest in learning， and achieved a silent educational effect. The ideological and political elements mentioned can also provide reference for ideological and political education in other courses of information majorsKeywords：curriculum ideological and political education;data structure;information major;teaching cases- 9
- |
- 1
- |
- 0
Published：2024-09-25 -
Opinions on High Quality Course Construction and Teaching Reform：Basic Elements and Connotation，Challenges and Trends AI Introduction
“在新时代背景下，高校课程建设与教学改革迎来新挑战，专家提出12个核心要素，推动人才培养质量提升。”Abstract：The new era has put forward new requirements for higher education teaching and talent cultivation， and the country and society's thirst for high-quality innovative talents is becoming increasingly urgent. In this context， new trends have emerged in the construction of university curricula， and the reform of curriculum and teaching is constantly deepening. To this end， the basic connotation of curriculum construction and teaching reform is first pointed out， which not only points out the core elements of 12 aspects， but also outlines the internal relationships of each element， and discusses the key elements among them； Then analyze the challenges and development trends faced by curriculum construction and teaching reform in the new era. Practice has shown that university teachers should always regard curriculum construction as the core work of talent cultivation， regard teaching reform as an important means to improve quality， continuously improve the comprehensive ability and level of curriculum teaching， create higher quality first-class courses， and lay a solid foundation for talent cultivation.Keywords：course construction;teaching reform;first-class courses;Golden Course;talent cultivation- 8
- |
- 0
- |
- 0
Published：2024-09-25
- Technical support is provided by Beijing Founder electronics co., LTD 鄂ICP备18021494号-4
鄂公网安备42010602004949号 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0