For realizing of long text information hiding and covert communication, a binary watermark sequence was obtained firstly from a text file and encoded by a redundant encoding method. Then, two neighboring blocks were s...For realizing of long text information hiding and covert communication, a binary watermark sequence was obtained firstly from a text file and encoded by a redundant encoding method. Then, two neighboring blocks were selected at each time from the Hilbert scanning sequence of carrier image blocks, and transformed by 1-level discrete wavelet transformation (DWT). And then the double block based JNDs (just noticeable difference) were calculated with a visual model. According to the different codes of each two watermark bits, the average values of two corresponding detail sub-bands were modified by using one of JNDs to hide information into carrier image. The experimental results show that the hidden information is invisible to human eyes, and the algorithm is robust to some common image processing operations. The conclusion is that the algorithm is effective and practical.展开更多
Discriminative region localization and efficient feature encoding are crucial for fine-grained object recognition.However,existing data augmentation methods struggle to accurately locate discriminative regions in comp...Discriminative region localization and efficient feature encoding are crucial for fine-grained object recognition.However,existing data augmentation methods struggle to accurately locate discriminative regions in complex backgrounds,small target objects,and limited training data,leading to poor recognition.Fine-grained images exhibit“small inter-class differences,”and while second-order feature encoding enhances discrimination,it often requires dual Convolutional Neural Networks(CNN),increasing training time and complexity.This study proposes a model integrating discriminative region localization and efficient second-order feature encoding.By ranking feature map channels via a fully connected layer,it selects high-importance channels to generate an enhanced map,accurately locating discriminative regions.Cropping and erasing augmentations further refine recognition.To improve efficiency,a novel second-order feature encoding module generates an attention map from the fourth convolutional group of Residual Network 50 layers(ResNet-50)and multiplies it with features from the fifth group,producing second-order features while reducing dimensionality and training time.Experiments on Caltech-University of California,San Diego Birds-200-2011(CUB-200-2011),Stanford Car,and Fine-Grained Visual Classification of Aircraft(FGVC Aircraft)datasets show state-of-the-art accuracy of 88.9%,94.7%,and 93.3%,respectively.展开更多
With the rapid expansion of social media,analyzing emotions and their causes in texts has gained significant importance.Emotion-cause pair extraction enables the identification of causal relationships between emotions...With the rapid expansion of social media,analyzing emotions and their causes in texts has gained significant importance.Emotion-cause pair extraction enables the identification of causal relationships between emotions and their triggers within a text,facilitating a deeper understanding of expressed sentiments and their underlying reasons.This comprehension is crucial for making informed strategic decisions in various business and societal contexts.However,recent research approaches employing multi-task learning frameworks for modeling often face challenges such as the inability to simultaneouslymodel extracted features and their interactions,or inconsistencies in label prediction between emotion-cause pair extraction and independent assistant tasks like emotion and cause extraction.To address these issues,this study proposes an emotion-cause pair extraction methodology that incorporates joint feature encoding and task alignment mechanisms.The model consists of two primary components:First,joint feature encoding simultaneously generates features for emotion-cause pairs and clauses,enhancing feature interactions between emotion clauses,cause clauses,and emotion-cause pairs.Second,the task alignment technique is applied to reduce the labeling distance between emotion-cause pair extraction and the two assistant tasks,capturing deep semantic information interactions among tasks.The proposed method is evaluated on a Chinese benchmark corpus using 10-fold cross-validation,assessing key performance metrics such as precision,recall,and F1 score.Experimental results demonstrate that the model achieves an F1 score of 76.05%,surpassing the state-of-the-art by 1.03%.The proposed model exhibits significant improvements in emotion-cause pair extraction(ECPE)and cause extraction(CE)compared to existing methods,validating its effectiveness.This research introduces a novel approach based on joint feature encoding and task alignment mechanisms,contributing to advancements in emotion-cause pair extraction.However,the study’s limitation lies in the data sources,potentially restricting the generalizability of the findings.展开更多
Image compression consists of two main parts: encoding and decoding. One of the important problems of the fractal theory is the long encoding implementation time, which hindered the acceptance of fractal image compres...Image compression consists of two main parts: encoding and decoding. One of the important problems of the fractal theory is the long encoding implementation time, which hindered the acceptance of fractal image compression as a practical method. The long encoding time results from the need to perform a large number of domain-range matches, the total encoding time is the product of the number of matches and the time to perform each match. In order to improve encoding speed, a hybrid method combining features extraction and self-organization network has been provided, which is based on the feature extraction approach the comparison pixels by pixels between the feature of range blocks and domains blocks. The efficiency of the new method was been proved by examples.展开更多
Sepsis poses a serious threat to health of children in pediatric intensive care unit.The mortality from pediatric sepsis can be effectively reduced through in-time diagnosis and therapeutic intervention.The bacillicul...Sepsis poses a serious threat to health of children in pediatric intensive care unit.The mortality from pediatric sepsis can be effectively reduced through in-time diagnosis and therapeutic intervention.The bacilliculture detection method is too time-consuming to receive timely treatment.In this research,we propose a new framework:a deep encoding network with cross features(CF-DEN)that enables accurate early detection of sepsis.Cross features are automatically constructed via the gradient boosting decision tree and distilled into the deep encoding network(DEN)we designed.The DEN is aimed at learning sufficiently effective representation from clinical test data.Each layer of the DEN fltrates the features involved in computation at current layer via attention mechanism and outputs the current prediction which is additive layer by layer to obtain the embedding feature at last layer.The framework takes the advantage of tree-based method and neural network method to extract effective representation from small clinical dataset and obtain accurate prediction in order to prompt patient to get timely treatment.We evaluate the performance of the framework on the dataset collected from Shanghai Children's Medical Center.Compared with common machine learning methods,our method achieves the increase on F1-score by 16.06%on the test set.展开更多
The detection of brain disease is an essential issue in medical and research areas.Deep learning techniques have shown promising results in detecting and diagnosing brain diseases using magnetic resonance imaging(MRI)...The detection of brain disease is an essential issue in medical and research areas.Deep learning techniques have shown promising results in detecting and diagnosing brain diseases using magnetic resonance imaging(MRI)images.These techniques involve training neural networks on large datasets of MRI images,allowing the networks to learn patterns and features indicative of different brain diseases.However,several challenges and limitations still need to be addressed further to improve the accuracy and effectiveness of these techniques.This paper implements a Feature Enhanced Stacked Auto Encoder(FESAE)model to detect brain diseases.The standard stack auto encoder’s results are trivial and not robust enough to boost the system’s accuracy.Therefore,the standard Stack Auto Encoder(SAE)is replaced with a Stacked Feature Enhanced Auto Encoder with a feature enhancement function to efficiently and effectively get non-trivial features with less activation energy froman image.The proposed model consists of four stages.First,pre-processing is performed to remove noise,and the greyscale image is converted to Red,Green,and Blue(RGB)to enhance feature details for discriminative feature extraction.Second,feature Extraction is performed to extract significant features for classification using DiscreteWavelet Transform(DWT)and Channelization.Third,classification is performed to classify MRI images into four major classes:Normal,Tumor,Brain Stroke,and Alzheimer’s.Finally,the FESAE model outperforms the state-of-theart,machine learning,and deep learning methods such as Artificial Neural Network(ANN),SAE,Random Forest(RF),and Logistic Regression(LR)by achieving a high accuracy of 98.61% on a dataset of 2000 MRI images.The proposed model has significant potential for assisting radiologists in diagnosing brain diseases more accurately and improving patient outcomes.展开更多
Multimodal sentiment analysis aims to understand emotions from text,speech,and video data.However,current methods often overlook the dominant role of text and suffer from feature loss during integration.Given the vary...Multimodal sentiment analysis aims to understand emotions from text,speech,and video data.However,current methods often overlook the dominant role of text and suffer from feature loss during integration.Given the varying importance of each modality across different contexts,a central and pressing challenge in multimodal sentiment analysis lies in maximizing the use of rich intra-modal features while minimizing information loss during the fusion process.In response to these critical limitations,we propose a novel framework that integrates spatial position encoding and fusion embedding modules to address these issues.In our model,text is treated as the core modality,while speech and video features are selectively incorporated through a unique position-aware fusion process.The spatial position encoding strategy preserves the internal structural information of speech and visual modalities,enabling the model to capture localized intra-modal dependencies that are often overlooked.This design enhances the richness and discriminative power of the fused representation,enabling more accurate and context-aware sentiment prediction.Finally,we conduct comprehensive evaluations on two widely recognized standard datasets in the field—CMU-MOSI and CMU-MOSEI to validate the performance of the proposed model.The experimental results demonstrate that our model exhibits good performance and effectiveness for sentiment analysis tasks.展开更多
Automatic crack detection of cement pavement chiefly benefits from the rapid development of deep learning,with convolutional neural networks(CNN)playing an important role in this field.However,as the performance of cr...Automatic crack detection of cement pavement chiefly benefits from the rapid development of deep learning,with convolutional neural networks(CNN)playing an important role in this field.However,as the performance of crack detection in cement pavement improves,the depth and width of the network structure are significantly increased,which necessitates more computing power and storage space.This limitation hampers the practical implementation of crack detection models on various platforms,particularly portable devices like small mobile devices.To solve these problems,we propose a dual-encoder-based network architecture that focuses on extracting more comprehensive fracture feature information and combines cross-fusion modules and coordinated attention mechanisms formore efficient feature fusion.Firstly,we use small channel convolution to construct shallow feature extractionmodule(SFEM)to extract low-level feature information of cracks in cement pavement images,in order to obtainmore information about cracks in the shallowfeatures of images.In addition,we construct large kernel atrous convolution(LKAC)to enhance crack information,which incorporates coordination attention mechanism for non-crack information filtering,and large kernel atrous convolution with different cores,using different receptive fields to extract more detailed edge and context information.Finally,the three-stage feature map outputs from the shallow feature extraction module is cross-fused with the two-stage feature map outputs from the large kernel atrous convolution module,and the shallow feature and detailed edge feature are fully fused to obtain the final crack prediction map.We evaluate our method on three public crack datasets:DeepCrack,CFD,and Crack500.Experimental results on theDeepCrack dataset demonstrate the effectiveness of our proposed method compared to state-of-the-art crack detection methods,which achieves Precision(P)87.2%,Recall(R)87.7%,and F-score(F1)87.4%.Thanks to our lightweight crack detectionmodel,the parameter count of the model in real-world detection scenarios has been significantly reduced to less than 2M.This advancement also facilitates technical support for portable scene detection.展开更多
An approach for color image segmentation is proposed based on the contributions of color features to segmentation rather than the choice of a particular color space. The determination of effective color features depen...An approach for color image segmentation is proposed based on the contributions of color features to segmentation rather than the choice of a particular color space. The determination of effective color features depends on the analysis of various color features from each tested color image via the designed feature encoding. It is different from the pervious methods where self organized feature map (SOFM) is used for constructing the feature encoding so that the feature encoding can self organize the effective features for different color images. Fuzzy clustering is applied for the final segmentation when the well suited color features and the initial parameter are available. The proposed method has been applied in segmenting different types of color images and the experimental results show that it outperforms the classical clustering method. The study shows that the feature encoding approach offers great promise in automating and optimizing the segmentation of color images.展开更多
In the natural environment,non-stationary background noise affects the animal sound recognition directly.Given this problem,a new technology of animal sound recognition based on energy-frequency(E-F)feature is propose...In the natural environment,non-stationary background noise affects the animal sound recognition directly.Given this problem,a new technology of animal sound recognition based on energy-frequency(E-F)feature is proposed in this paper.The animal sound is turned into spectrogram to show the energy,time and frequency characteristics.The sub-band frequency division and sub-band energy division are carried out on the spectrogram for extracting the statistical characteristic of energy and frequency,so as to achieve sub-band power distribution(SPD)and sub-band division.Radon transform(RT)and discrete wavelet transform(DWT)are employed to obtain the important projection coefficients,and the energy values of sub-band frequencies are calculated to extract the sub-band frequency feature.The E-F feature is formed by combining the SPD feature and sub-band energy value feature.The classification is achieved by support vector machine(SVM)classifier.The experimental results show that the method can achieve better recognition effect even when the SNR is below10 dB.展开更多
在一些修船企业建立的修船结算系统和电子价格库中,人工匹配结算编码步骤易出错且耗时长,直接影响结算效率。为解决该问题,提出一种基于多特征融合的修船结算编码智能匹配复合模型。采用来自变换器的双向编码器表示(Bidirectional Encod...在一些修船企业建立的修船结算系统和电子价格库中,人工匹配结算编码步骤易出错且耗时长,直接影响结算效率。为解决该问题,提出一种基于多特征融合的修船结算编码智能匹配复合模型。采用来自变换器的双向编码器表示(Bidirectional Encoder Representations from Transformers,BERT)模型将工程内容文本表示为词向量,采用卷积神经网络(Convolutional Neural Network,CNN)模型提取文本的局部特征,采用双向长短期记忆网络结合注意力机制(Bidirectional Long Short-Term Memory with Attention Mechanism,BiLSTM-Attention)模型提取上下文特征,得到对应的结算编码。试验结果表明,所提出的复合模型在整体准确率方面实现显著提升,充分证明该复合模型在处理复杂文本分类任务中的优势。展开更多
文摘For realizing of long text information hiding and covert communication, a binary watermark sequence was obtained firstly from a text file and encoded by a redundant encoding method. Then, two neighboring blocks were selected at each time from the Hilbert scanning sequence of carrier image blocks, and transformed by 1-level discrete wavelet transformation (DWT). And then the double block based JNDs (just noticeable difference) were calculated with a visual model. According to the different codes of each two watermark bits, the average values of two corresponding detail sub-bands were modified by using one of JNDs to hide information into carrier image. The experimental results show that the hidden information is invisible to human eyes, and the algorithm is robust to some common image processing operations. The conclusion is that the algorithm is effective and practical.
基金supported,in part,by the National Nature Science Foundation of China under Grant 62272236,62376128 and 62306139the Natural Science Foundation of Jiangsu Province under Grant BK20201136,BK20191401.
文摘Discriminative region localization and efficient feature encoding are crucial for fine-grained object recognition.However,existing data augmentation methods struggle to accurately locate discriminative regions in complex backgrounds,small target objects,and limited training data,leading to poor recognition.Fine-grained images exhibit“small inter-class differences,”and while second-order feature encoding enhances discrimination,it often requires dual Convolutional Neural Networks(CNN),increasing training time and complexity.This study proposes a model integrating discriminative region localization and efficient second-order feature encoding.By ranking feature map channels via a fully connected layer,it selects high-importance channels to generate an enhanced map,accurately locating discriminative regions.Cropping and erasing augmentations further refine recognition.To improve efficiency,a novel second-order feature encoding module generates an attention map from the fourth convolutional group of Residual Network 50 layers(ResNet-50)and multiplies it with features from the fifth group,producing second-order features while reducing dimensionality and training time.Experiments on Caltech-University of California,San Diego Birds-200-2011(CUB-200-2011),Stanford Car,and Fine-Grained Visual Classification of Aircraft(FGVC Aircraft)datasets show state-of-the-art accuracy of 88.9%,94.7%,and 93.3%,respectively.
文摘With the rapid expansion of social media,analyzing emotions and their causes in texts has gained significant importance.Emotion-cause pair extraction enables the identification of causal relationships between emotions and their triggers within a text,facilitating a deeper understanding of expressed sentiments and their underlying reasons.This comprehension is crucial for making informed strategic decisions in various business and societal contexts.However,recent research approaches employing multi-task learning frameworks for modeling often face challenges such as the inability to simultaneouslymodel extracted features and their interactions,or inconsistencies in label prediction between emotion-cause pair extraction and independent assistant tasks like emotion and cause extraction.To address these issues,this study proposes an emotion-cause pair extraction methodology that incorporates joint feature encoding and task alignment mechanisms.The model consists of two primary components:First,joint feature encoding simultaneously generates features for emotion-cause pairs and clauses,enhancing feature interactions between emotion clauses,cause clauses,and emotion-cause pairs.Second,the task alignment technique is applied to reduce the labeling distance between emotion-cause pair extraction and the two assistant tasks,capturing deep semantic information interactions among tasks.The proposed method is evaluated on a Chinese benchmark corpus using 10-fold cross-validation,assessing key performance metrics such as precision,recall,and F1 score.Experimental results demonstrate that the model achieves an F1 score of 76.05%,surpassing the state-of-the-art by 1.03%.The proposed model exhibits significant improvements in emotion-cause pair extraction(ECPE)and cause extraction(CE)compared to existing methods,validating its effectiveness.This research introduces a novel approach based on joint feature encoding and task alignment mechanisms,contributing to advancements in emotion-cause pair extraction.However,the study’s limitation lies in the data sources,potentially restricting the generalizability of the findings.
文摘Image compression consists of two main parts: encoding and decoding. One of the important problems of the fractal theory is the long encoding implementation time, which hindered the acceptance of fractal image compression as a practical method. The long encoding time results from the need to perform a large number of domain-range matches, the total encoding time is the product of the number of matches and the time to perform each match. In order to improve encoding speed, a hybrid method combining features extraction and self-organization network has been provided, which is based on the feature extraction approach the comparison pixels by pixels between the feature of range blocks and domains blocks. The efficiency of the new method was been proved by examples.
文摘Sepsis poses a serious threat to health of children in pediatric intensive care unit.The mortality from pediatric sepsis can be effectively reduced through in-time diagnosis and therapeutic intervention.The bacilliculture detection method is too time-consuming to receive timely treatment.In this research,we propose a new framework:a deep encoding network with cross features(CF-DEN)that enables accurate early detection of sepsis.Cross features are automatically constructed via the gradient boosting decision tree and distilled into the deep encoding network(DEN)we designed.The DEN is aimed at learning sufficiently effective representation from clinical test data.Each layer of the DEN fltrates the features involved in computation at current layer via attention mechanism and outputs the current prediction which is additive layer by layer to obtain the embedding feature at last layer.The framework takes the advantage of tree-based method and neural network method to extract effective representation from small clinical dataset and obtain accurate prediction in order to prompt patient to get timely treatment.We evaluate the performance of the framework on the dataset collected from Shanghai Children's Medical Center.Compared with common machine learning methods,our method achieves the increase on F1-score by 16.06%on the test set.
基金supported by financial support from Universiti Sains Malaysia(USM)under FRGS Grant Number FRGS/1/2020/TK03/USM/02/1the School of Computer Sciences USM for their support.
文摘The detection of brain disease is an essential issue in medical and research areas.Deep learning techniques have shown promising results in detecting and diagnosing brain diseases using magnetic resonance imaging(MRI)images.These techniques involve training neural networks on large datasets of MRI images,allowing the networks to learn patterns and features indicative of different brain diseases.However,several challenges and limitations still need to be addressed further to improve the accuracy and effectiveness of these techniques.This paper implements a Feature Enhanced Stacked Auto Encoder(FESAE)model to detect brain diseases.The standard stack auto encoder’s results are trivial and not robust enough to boost the system’s accuracy.Therefore,the standard Stack Auto Encoder(SAE)is replaced with a Stacked Feature Enhanced Auto Encoder with a feature enhancement function to efficiently and effectively get non-trivial features with less activation energy froman image.The proposed model consists of four stages.First,pre-processing is performed to remove noise,and the greyscale image is converted to Red,Green,and Blue(RGB)to enhance feature details for discriminative feature extraction.Second,feature Extraction is performed to extract significant features for classification using DiscreteWavelet Transform(DWT)and Channelization.Third,classification is performed to classify MRI images into four major classes:Normal,Tumor,Brain Stroke,and Alzheimer’s.Finally,the FESAE model outperforms the state-of-theart,machine learning,and deep learning methods such as Artificial Neural Network(ANN),SAE,Random Forest(RF),and Logistic Regression(LR)by achieving a high accuracy of 98.61% on a dataset of 2000 MRI images.The proposed model has significant potential for assisting radiologists in diagnosing brain diseases more accurately and improving patient outcomes.
基金supported by the Collaborative Tackling Project of the Yangtze River Delta SciTech Innovation Community(Nos.2024CSJGG01503,2024CSJGG01500)Guangxi Key Research and Development Program(No.AB24010317)Jiangxi Provincial Key Laboratory of Electronic Data Control and Forensics(Jiangxi Police College)(No.2025JXJYKFJJ002).
文摘Multimodal sentiment analysis aims to understand emotions from text,speech,and video data.However,current methods often overlook the dominant role of text and suffer from feature loss during integration.Given the varying importance of each modality across different contexts,a central and pressing challenge in multimodal sentiment analysis lies in maximizing the use of rich intra-modal features while minimizing information loss during the fusion process.In response to these critical limitations,we propose a novel framework that integrates spatial position encoding and fusion embedding modules to address these issues.In our model,text is treated as the core modality,while speech and video features are selectively incorporated through a unique position-aware fusion process.The spatial position encoding strategy preserves the internal structural information of speech and visual modalities,enabling the model to capture localized intra-modal dependencies that are often overlooked.This design enhances the richness and discriminative power of the fused representation,enabling more accurate and context-aware sentiment prediction.Finally,we conduct comprehensive evaluations on two widely recognized standard datasets in the field—CMU-MOSI and CMU-MOSEI to validate the performance of the proposed model.The experimental results demonstrate that our model exhibits good performance and effectiveness for sentiment analysis tasks.
基金supported by the National Natural Science Foundation of China(No.62176034)the Science and Technology Research Program of Chongqing Municipal Education Commission(No.KJZD-M202300604)the Natural Science Foundation of Chongqing(Nos.cstc2021jcyj-msxmX0518,2023NSCQ-MSX1781).
文摘Automatic crack detection of cement pavement chiefly benefits from the rapid development of deep learning,with convolutional neural networks(CNN)playing an important role in this field.However,as the performance of crack detection in cement pavement improves,the depth and width of the network structure are significantly increased,which necessitates more computing power and storage space.This limitation hampers the practical implementation of crack detection models on various platforms,particularly portable devices like small mobile devices.To solve these problems,we propose a dual-encoder-based network architecture that focuses on extracting more comprehensive fracture feature information and combines cross-fusion modules and coordinated attention mechanisms formore efficient feature fusion.Firstly,we use small channel convolution to construct shallow feature extractionmodule(SFEM)to extract low-level feature information of cracks in cement pavement images,in order to obtainmore information about cracks in the shallowfeatures of images.In addition,we construct large kernel atrous convolution(LKAC)to enhance crack information,which incorporates coordination attention mechanism for non-crack information filtering,and large kernel atrous convolution with different cores,using different receptive fields to extract more detailed edge and context information.Finally,the three-stage feature map outputs from the shallow feature extraction module is cross-fused with the two-stage feature map outputs from the large kernel atrous convolution module,and the shallow feature and detailed edge feature are fully fused to obtain the final crack prediction map.We evaluate our method on three public crack datasets:DeepCrack,CFD,and Crack500.Experimental results on theDeepCrack dataset demonstrate the effectiveness of our proposed method compared to state-of-the-art crack detection methods,which achieves Precision(P)87.2%,Recall(R)87.7%,and F-score(F1)87.4%.Thanks to our lightweight crack detectionmodel,the parameter count of the model in real-world detection scenarios has been significantly reduced to less than 2M.This advancement also facilitates technical support for portable scene detection.
文摘An approach for color image segmentation is proposed based on the contributions of color features to segmentation rather than the choice of a particular color space. The determination of effective color features depends on the analysis of various color features from each tested color image via the designed feature encoding. It is different from the pervious methods where self organized feature map (SOFM) is used for constructing the feature encoding so that the feature encoding can self organize the effective features for different color images. Fuzzy clustering is applied for the final segmentation when the well suited color features and the initial parameter are available. The proposed method has been applied in segmenting different types of color images and the experimental results show that it outperforms the classical clustering method. The study shows that the feature encoding approach offers great promise in automating and optimizing the segmentation of color images.
基金Supported by the National Natural Science Foundation of China(No.61075022)
文摘In the natural environment,non-stationary background noise affects the animal sound recognition directly.Given this problem,a new technology of animal sound recognition based on energy-frequency(E-F)feature is proposed in this paper.The animal sound is turned into spectrogram to show the energy,time and frequency characteristics.The sub-band frequency division and sub-band energy division are carried out on the spectrogram for extracting the statistical characteristic of energy and frequency,so as to achieve sub-band power distribution(SPD)and sub-band division.Radon transform(RT)and discrete wavelet transform(DWT)are employed to obtain the important projection coefficients,and the energy values of sub-band frequencies are calculated to extract the sub-band frequency feature.The E-F feature is formed by combining the SPD feature and sub-band energy value feature.The classification is achieved by support vector machine(SVM)classifier.The experimental results show that the method can achieve better recognition effect even when the SNR is below10 dB.
文摘在一些修船企业建立的修船结算系统和电子价格库中,人工匹配结算编码步骤易出错且耗时长,直接影响结算效率。为解决该问题,提出一种基于多特征融合的修船结算编码智能匹配复合模型。采用来自变换器的双向编码器表示(Bidirectional Encoder Representations from Transformers,BERT)模型将工程内容文本表示为词向量,采用卷积神经网络(Convolutional Neural Network,CNN)模型提取文本的局部特征,采用双向长短期记忆网络结合注意力机制(Bidirectional Long Short-Term Memory with Attention Mechanism,BiLSTM-Attention)模型提取上下文特征,得到对应的结算编码。试验结果表明,所提出的复合模型在整体准确率方面实现显著提升,充分证明该复合模型在处理复杂文本分类任务中的优势。
