Research on Multi-Channel Semantic Fusion Classification Model

In this work, we propose a multi-channel semantic fusion convolutional neural network (SFCNN) to solve the problem of emotional ambiguity caused by the change of contextual order in sentiment classification task. Firstly, the emotional tendency weights are evaluated on the text word vector through the improved emotional tendency attention mechanism. Secondly, the multi-channel semantic fusion layer is leveraged to combine deep semantic fusion of sentences with contextual order to generate deep semantic vectors, which are learned by CNN to extract high-level semantic features. Finally, the improved adaptive learning rate gradient descent algorithm is employed to optimize the model parameters, and completes the sentiment classification task. Three datasets are used to evaluate the effectiveness of the proposed algorithm. The experimental results show that the SFCNN model has the high steady-state precision and generalization performance.

Download Full-text

Improving Document-Level Sentiment Classification Using Importance of Sentences

Entropy ◽

10.3390/e22121336 ◽

2020 ◽

Vol 22 (12) ◽

pp. 1336

Author(s):

Gihyeon Choi ◽

Shinhyeok Oh ◽

Harksoo Kim

Keyword(s):

Sentiment Analysis ◽

Deep Neural Networks ◽

Sentiment Classification ◽

Classification Model ◽

Classification Task ◽

Analysis Model ◽

Proposed Model ◽

Previous State ◽

Restaurant Reviews ◽

Document Level

Previous researchers have considered sentiment analysis as a document classification task, in which input documents are classified into predefined sentiment classes. Although there are sentences in a document that support important evidences for sentiment analysis and sentences that do not, they have treated the document as a bag of sentences. In other words, they have not considered the importance of each sentence in the document. To effectively determine polarity of a document, each sentence in the document should be dealt with different degrees of importance. To address this problem, we propose a document-level sentence classification model based on deep neural networks, in which the importance degrees of sentences in documents are automatically determined through gate mechanisms. To verify our new sentiment analysis model, we conducted experiments using the sentiment datasets in the four different domains such as movie reviews, hotel reviews, restaurant reviews, and music reviews. In the experiments, the proposed model outperformed previous state-of-the-art models that do not consider importance differences of sentences in a document. The experimental results show that the importance of sentences should be considered in a document-level sentiment classification task.

Download Full-text

Adversarial Feature Alignment: Avoid Catastrophic Forgetting in Incremental Task Lifelong Learning

Neural Computation ◽

10.1162/neco_a_01232 ◽

2019 ◽

Vol 31 (11) ◽

pp. 2266-2291 ◽

Cited By ~ 4

Author(s):

Xin Yao ◽

Tianchi Huang ◽

Chenglei Wu ◽

Rui-Xiao Zhang ◽

Lifeng Sun

Keyword(s):

Lifelong Learning ◽

Historical Memory ◽

Fine Tuning ◽

Model Parameters ◽

Semantic Features ◽

Task Sequence ◽

Knowledge Distillation ◽

And Performance ◽

High Level ◽

Feature Alignment

Humans are able to master a variety of knowledge and skills with ongoing learning. By contrast, dramatic performance degradation is observed when new tasks are added to an existing neural network model. This phenomenon, termed catastrophic forgetting, is one of the major roadblocks that prevent deep neural networks from achieving human-level artificial intelligence. Several research efforts (e.g., lifelong or continual learning algorithms) have proposed to tackle this problem. However, they either suffer from an accumulating drop in performance as the task sequence grows longer, or require storing an excessive number of model parameters for historical memory, or cannot obtain competitive performance on the new tasks. In this letter, we focus on the incremental multitask image classification scenario. Inspired by the learning process of students, who usually decompose complex tasks into easier goals, we propose an adversarial feature alignment method to avoid catastrophic forgetting. In our design, both the low-level visual features and high-level semantic features serve as soft targets and guide the training process in multiple stages, which provide sufficient supervised information of the old tasks and help to reduce forgetting. Due to the knowledge distillation and regularization phenomena, the proposed method gains even better performance than fine-tuning on the new tasks, which makes it stand out from other methods. Extensive experiments in several typical lifelong learning scenarios demonstrate that our method outperforms the state-of-the-art methods in both accuracy on new tasks and performance preservation on old tasks.

Download Full-text

A Parallel Convolutional Neural Network for Pedestrian Detection

Electronics ◽

10.3390/electronics9091478 ◽

2020 ◽

Vol 9 (9) ◽

pp. 1478

Author(s):

Mengya Zhu ◽

Yiquan Wu

Keyword(s):

Pedestrian Detection ◽

Autonomous Driving ◽

Feature Representation ◽

Model Parameters ◽

Detection Accuracy ◽

Semantic Features ◽

Practical Application ◽

Lightweight Framework ◽

Human Activity Analysis ◽

High Level

Pedestrian detection is a crucial task in many vision-based applications, such as video surveillance, human activity analysis and autonomous driving. Recently, most of the existing pedestrian detection frameworks only focus on the detection accuracy or model parameters. However, how to balance the detection accuracy and model parameters, is still an open problem for the practical application of pedestrian detection. In this paper, we propose a parallel, lightweight framework for pedestrian detection, named ParallelNet. ParallelNet consists of four branches, each of them learns different high-level semantic features. We fused them into one feature map as the final feature representation. Subsequently, the Fire module, which includes Squeeze and Expand parts, is employed for reducing the model parameters. Here, we replace some convolution modules in the backbone with Fire modules. Finally, the focal loss is led into the ParallelNet for end-to-end training. Experimental results on the Caltech–Zhang dataset and KITTI dataset show that: Compared with the single-branch network, such as ResNet and SqueezeNet, ParallelNet has improved detection accuracy with fewer model parameters and lower Giga Floating Point Operations (GFLOPs).

Download Full-text

Sentiment classification based on weak tagging information and imbalanced data

Intelligent Data Analysis ◽

10.3233/ida-205408 ◽

2021 ◽

Vol 25 (3) ◽

pp. 555-570

Author(s):

Chuantao Wang ◽

Xuexin Yang ◽

Linkai Ding

Keyword(s):

Deep Learning ◽

Imbalanced Data ◽

Online Reviews ◽

Sentiment Classification ◽

Classification Task ◽

Model Parameters ◽

Learning Models ◽

Algorithm Optimization ◽

Model Training ◽

The Impact

Sentiment classification aims to solve the problem of automatic judgment of sentiment polarity. In the sentiment classification task of text data, such as online reviews, traditional deep learning models are dedicated to algorithm optimization but ignore the characteristics of imbalanced distribution of the number of classified samples and the inclusion of weak tagging information such as ratings and tags. Based on the traditional deep learning model, the method of random oversampling and cost sensitivity is used to increase the contribution of a minority of samples to the model loss function and avoid the model biasing to the majority of samples. The model training is divided into two stages. In the first stage, a large amount of weak tagging data is used to train the model, therefore a model that captures the sentiment semantics of the data is obtained. After that, the model parameters trained in the first stage are used as the initial parameters of the second stage model training, and only a small amount of tagging data is used to continue training the model to reduce the impact of noise, thus reducing the use of manual tagging samples. The experimental results show that the method is considerably better than traditional deep learning models in the sentiment classification task of hotel review data.

Download Full-text

Extraction of Arecanut Planting Distribution Based on the Feature Space Optimization of PlanetScope Imagery

Agriculture ◽

10.3390/agriculture11040371 ◽

2021 ◽

Vol 11 (4) ◽

pp. 371

Author(s):

Yu Jin ◽

Jiawei Guo ◽

Huichun Ye ◽

Jinling Zhao ◽

Wenjiang Huang ◽

...

Keyword(s):

Random Forest ◽

Satellite Imagery ◽

Feature Space ◽

Kappa Coefficient ◽

Classification Model ◽

Support Vector ◽

Textural Feature ◽

Monitoring Accuracy ◽

Areca Catechu ◽

High Level

The remote sensing extraction of large areas of arecanut (Areca catechu L.) planting plays an important role in investigating the distribution of arecanut planting area and the subsequent adjustment and optimization of regional planting structures. Satellite imagery has previously been used to investigate and monitor the agricultural and forestry vegetation in Hainan. However, the monitoring accuracy is affected by the cloudy and rainy climate of this region, as well as the high level of land fragmentation. In this paper, we used PlanetScope imagery at a 3 m spatial resolution over the Hainan arecanut planting area to investigate the high-precision extraction of the arecanut planting distribution based on feature space optimization. First, spectral and textural feature variables were selected to form the initial feature space, followed by the implementation of the random forest algorithm to optimize the feature space. Arecanut planting area extraction models based on the support vector machine (SVM), BP neural network (BPNN), and random forest (RF) classification algorithms were then constructed. The overall classification accuracies of the SVM, BPNN, and RF models optimized by the RF features were determined as 74.82%, 83.67%, and 88.30%, with Kappa coefficients of 0.680, 0.795, and 0.853, respectively. The RF model with optimized features exhibited the highest overall classification accuracy and kappa coefficient. The overall accuracy of the SVM, BPNN, and RF models following feature optimization was improved by 3.90%, 7.77%, and 7.45%, respectively, compared with the corresponding unoptimized classification model. The kappa coefficient also improved. The results demonstrate the ability of PlanetScope satellite imagery to extract the planting distribution of arecanut. Furthermore, the RF is proven to effectively optimize the initial feature space, composed of spectral and textural feature variables, further improving the extraction accuracy of the arecanut planting distribution. This work can act as a theoretical and technical reference for the agricultural and forestry industries.

Download Full-text

AI-based localization and classification of skin disease with erythema

Scientific Reports ◽

10.1038/s41598-021-84593-z ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Ha Min Son ◽

Wooho Jeon ◽

Jinhyun Kim ◽

Chan Yeong Heo ◽

Hye Jin Yoon ◽

...

Keyword(s):

Neural Network ◽

Skin Diseases ◽

Classification Model ◽

Screening Tests ◽

Sensitivity Score ◽

Common Skin ◽

Novel Method ◽

Improved Performance ◽

High Level

AbstractAlthough computer-aided diagnosis (CAD) is used to improve the quality of diagnosis in various medical fields such as mammography and colonography, it is not used in dermatology, where noninvasive screening tests are performed only with the naked eye, and avoidable inaccuracies may exist. This study shows that CAD may also be a viable option in dermatology by presenting a novel method to sequentially combine accurate segmentation and classification models. Given an image of the skin, we decompose the image to normalize and extract high-level features. Using a neural network-based segmentation model to create a segmented map of the image, we then cluster sections of abnormal skin and pass this information to a classification model. We classify each cluster into different common skin diseases using another neural network model. Our segmentation model achieves better performance compared to previous studies, and also achieves a near-perfect sensitivity score in unfavorable conditions. Our classification model is more accurate than a baseline model trained without segmentation, while also being able to classify multiple diseases within a single image. This improved performance may be sufficient to use CAD in the field of dermatology.

Download Full-text

The use of remote sensing satellite using deep learning in emergency monitoring of high-level landslides disaster in Jinsha River

The Journal of Supercomputing ◽

10.1007/s11227-020-03604-4 ◽

2021 ◽

Author(s):

Leijin Long ◽

Feng He ◽

Hongjiang Liu

Keyword(s):

Remote Sensing ◽

Southwest China ◽

Influence Factors ◽

Classification Error ◽

Model Parameters ◽

Detection Accuracy ◽

Remote Sensing Images ◽

Jinsha River ◽

Detection Model ◽

High Level

AbstractIn order to monitor the high-level landslides frequently occurring in Jinsha River area of Southwest China, and protect the lives and property safety of people in mountainous areas, the data of satellite remote sensing images are combined with various factors inducing landslides and transformed into landslide influence factors, which provides data basis for the establishment of landslide detection model. Then, based on the deep belief networks (DBN) and convolutional neural network (CNN) algorithm, two landslide detection models DBN and convolutional neural-deep belief network (CDN) are established to monitor the high-level landslide in Jinsha River. The influence of the model parameters on the landslide detection results is analyzed, and the accuracy of DBN and CDN models in dealing with actual landslide problems is compared. The results show that when the number of neurons in the DBN is 100, the overall error is the minimum, and when the number of learning layers is 3, the classification error is the minimum. The detection accuracy of DBN and CDN is 97.56% and 97.63%, respectively, which indicates that both DBN and CDN models are feasible in dealing with landslides from remote sensing images. This exploration provides a reference for the study of high-level landslide disasters in Jinsha River.

Download Full-text

An Efficient Module for Instance Segmentation Based on Multi-Level Features and Attention Mechanisms

Applied Sciences ◽

10.3390/app11030968 ◽

2021 ◽

Vol 11 (3) ◽

pp. 968

Author(s):

Yingchun Sun ◽

Wang Gao ◽

Shuguo Pan ◽

Tao Zhao ◽

Yahui Peng

Keyword(s):

Feature Extraction ◽

Spatial Structure ◽

Semantic Feature ◽

Semantic Features ◽

Segmentation Method ◽

Spatial Dimensions ◽

Feature Pyramid ◽

Multi Level ◽

High Level ◽

Instance Segmentation

Recently, multi-level feature networks have been extensively used in instance segmentation. However, because not all features are beneficial to instance segmentation tasks, the performance of networks cannot be adequately improved by synthesizing multi-level convolutional features indiscriminately. In order to solve the problem, an attention-based feature pyramid module (AFPM) is proposed, which integrates the attention mechanism on the basis of a multi-level feature pyramid network to efficiently and pertinently extract the high-level semantic features and low-level spatial structure features; for instance, segmentation. Firstly, we adopt a convolutional block attention module (CBAM) into feature extraction, and sequentially generate attention maps which focus on instance-related features along the channel and spatial dimensions. Secondly, we build inter-dimensional dependencies through a convolutional triplet attention module (CTAM) in lateral attention connections, which is used to propagate a helpful semantic feature map and filter redundant informative features irrelevant to instance objects. Finally, we construct branches for feature enhancement to strengthen detailed information to boost the entire feature hierarchy of the network. The experimental results on the Cityscapes dataset manifest that the proposed module outperforms other excellent methods under different evaluation metrics and effectively upgrades the performance of the instance segmentation method.

Download Full-text