检索结果-内蒙古大学图书馆

A three-branch deep neural network for diagnosing respiratory sounds

Neural Computing and Applications 2024年第35期36卷 22611-22631页

作者： Imani, Maryam Ghassemian, Hassan Image Processing and Information Analysis Lab Faculty of Electrical and Computer Engineering Tarbiat Modares University Tehran Iran

Finding an accurate model is essential for classification of respiratory pathologies through extraction and fusion of respiratory sounds’ features. To handle the unlabeled data, a sequence of autoencoders are used for data augmentation. A deep neural network framework is proposed to extract three types of features: 1-Features based on the human auditory system, using Mel-frequency cepstral coefficients (MFCCs), 2-Temporal features contained in the sequence of sound signal using the long short-term memory (LSTM) network, and 3-The nonlinear and complex relationship among temporal characteristics in neighborhood regions using a two-dimensional (2D) convolutional neural network (CNN) applied to the sound sequence converted to a 2D time array. Three branches are fused through some fully connected layers. The ICBHI 2017 sound database is used in two cases of 6 and 3 pathological classes. In the case of 6-classes database, 98.72% overall accuracy, 98.46% kappa coefficient, 99.66% sensitivity, 98.70% specificity and 99.70% F1-Sscore is provided. In the case of 3-classes database, 97.18% overall accuracy, 95.77% kappa coefficient, 98.31% sensitivity, 99.16% specificity and 98.93% F1-score are achieved. In contrast to the many researchers that are applying CNN on the Mel spectrogram of the audios, CNN is applied to the 2D form of the time series to find the hidden relationship among the time instants in local regions. Moreover, fusion of these hidden features with the LSTM and MFCC features leads to an accurate multiple classification. This work proposes a new deep learning framework for fusion of three types of sound features, which leads to a significant improvement in multiple respiratory infection diagnosing. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Hyperspectral image Classification: Spatial and Spectral Feature Fusion with Masked Autoencoders 20

Unsupervised Hyperspectral Image Classification: Spatial and...

引用

20th CSI International Symposium on Artificial Intelligence and Signal processing, AISP 2024

作者： Mirsharji, Seyed Ali Ghassemian, Hassan Tarbiat Modares University Image Processing and Information Analysis Lab. Faculty of Electrical and Computer Engineering Tehran Iran

ISBN: (纸本)9798350383942

In this study, we present an innovative unsupervised hyperspectral image classification method using a dual-branch architecture that merges spatial and spectral feature extraction. Our unique approach employs masked autoencoders, significantly outperforming traditional methods with an impressive overall accuracy of 97.1%. The paper details the model's performance evaluation, offers visual insights into its classification capabilities, and compares it with existing techniques, demonstrating its effectiveness and potential for advancing remote sensing applications. © 2024 IEEE.

关键词： image classification

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Change Detection in SAR images Using a Six-Branch CNN and Adaptive Window Approach 31

Unsupervised Change Detection in SAR Images Using a Six-Bran...

引用

31st International Conference on Electrical Engineering, ICEE 2023

作者： Kakoolvand, Abbas Imani, Maryam Ghassemian, Hassan Tarbiat Modares University Image Processing and Information Analysis Lab Faculty of Electrical and Computer Engineering Tehran Iran

ISBN: (纸本)9798350312560

Change detection is one of the important and hot topics in remote sensing. Adaptive windowing approaches can preserve image details while reduce noise in the process of change detection. In the proposed method, two difference images (DIs) are obtained by using the adaptive window approach. Some fake labels are provided from these DIs. Neural networks have a good performance in image processing. To use this advantage, a six-branch convolutional neural network (CNN) is trained using the fake labels. Using these six branches with utilizing the adaptive windows, the network can preserve image detail while reduce noise. The proposed method based on three criteria (PCC, Kappa, and F1 score) has the best results in two datasets and has results close to the best method in other datasets. © 2023 IEEE.

关键词： Synthetic aperture radar

来源：评论

学校读者我要写书评

暂无评论

Fusion of Multi-Level CNN With LBP Features For Facial Emotion Recognition 31

Fusion of Multi-Level CNN With LBP Features For Facial Emoti...

引用

31st International Conference on Electrical Engineering, ICEE 2023

作者： Bahmanabady, Ehsan Imani, Maryam Ghassemian, Hassan Tarbiat Modares University Image Processing and Information Analysis Lab Faculty of Electrical and Computer Engineering Tehran Iran

ISBN: (纸本)9798350312560

A facial emotion recognition framework is proposed in this work. The convolutional neural network (CNN) has high ability in extraction of hierarchical spatial features from low level texture characteristics to high level contextual features. A simple CNN model with three layers is suggested in this paper where the features extracted in all layers containing multi-level features are activated and used for classification. In addition, the local binary pattern (LBP) descriptor is used to extract discriminative features from the spatial structure of the input image. Therefore, four feature sources are provided by multi levels of CNN and the LBP descriptor. Each feature source is used for facial emotion recognition by applying to the support vector machine (SVM) classier. Finally, the majority voting rule is used for decision fusion to provide the final emotional label of each given face image. The proposed method with 84% overall accuracy, 83% weighted F1-score and 81% kappa coefficient provides the best performance compared to LBP, multi-level CNN and two-dimensional principal component analysis (2DPCA) methods. © 2023 IEEE.

关键词： Emotion Recognition

来源：评论

学校读者我要写书评

暂无评论

Building Detection Using Very High Resolution SAR images with Multi-Direction Based on Weighted-Morphological Indexes 12

Building Detection Using Very High Resolution SAR Images wit...

引用

12th Iranian/2nd International Conference on Machine Vision and image processing, MVIP 2022

作者： Amjadipour, Fateme Ghassemian, Hassan Imani, Maryam Tarbiat Modares University Image Processing and Information Analysis Lab Tehran Iran

ISBN: (纸本)9781665412162

Today, technological advancement in production of radar images can be seen with high spatial resolution and also the availability of these images' significant growth in interpretation and processing of high-resolution radar images. The building extraction from urban areas is one of the most challenging applications in VHR SAR image, which is used to estimate the population and urban development. Detection of individual buildings in the urban context is highly considered by researchers due to complexity of interpreting radar images in these fields. On the other hand, one of the main issues in the complexity of the scatters received from buildings is change in direction of the building relative to the horizon, which is correlated with the look angle. Other influential parameters are geometric distortions, which include layover and shadow effects. In some cases, the effect of shadow is an auxiliary parameter in detection of these targets that increases accuracy of the detection. In this paper, we intend to extract the building from high spatial resolution SAR images using fuzzy fusion of two morphological indicators, SI and DI, which represent the shadow and bright area, respectively. Due to the effect of SAR imaging geometry on ground targets, different sizes and directions of structural elements were applied to the image. The use of indicators weights with different sizes is proposed in this work. The Detection Ratio of experiment of TerraSAR-X image has a result of 95.3%. © 2022 IEEE.

关键词： Morphology

来源：评论

学校读者我要写书评

暂无评论

Deep Curriculum Learning for PolSAR image Classification 12

Deep Curriculum Learning for PolSAR Image Classification

引用

12th Iranian/2nd International Conference on Machine Vision and image processing, MVIP 2022

作者： Mousavi, Hamidreza Imani, Maryam Ghassemian, Hassan Image Processing and Information Analysis Lab Faculty of Electrical and Computer Engineering Tehran Iran

ISBN: (纸本)9781665412162

Following the great success of curriculum learning in the area of machine learning, a novel deep curriculum learning method proposed in this paper, entitled DCL, particularly for the classification of fully polarimetric synthetic aperture radar (PolSAR) data. This method utilizes the entropy-alpha target decomposition method to estimate the degree of complexity of each PolSAR image patch before applying it to the convolutional neural network (CNN). Also, an accumulative mini-batch pacing function is used to introduce more difficult patches to CNN. Experiments on the widely used data set of AIRSAR Flevoland reveal that the proposed curriculum learning method can not only increase classification accuracy but also lead to faster training convergence. © 2022 IEEE.

关键词： Synthetic aperture radar

来源：评论

学校读者我要写书评

暂无评论

Optimizing LaneSegNet for Real-Time Lane Topology Prediction in Autonomous Vehicles

arXiv

引用

arXiv 2024年

作者： Stevens, William Urs, Vishal Selvaraj, Karthik Torres, Gabriel Lakhanpal, Gaurish Delp, Edward J. Zoltowski, Carla B. Purdue University West Lafayette Indiana Image Processing and Analysis Lab

With the increasing prevalence of autonomous vehicles, it is essential for computer vision algorithms to accurately assess road features in real-time. This study explores the LaneSegNet architecture, a new approach to lane topology prediction which integrates topological information with lane-line data to provide a more contextual understanding of road environments. The LaneSegNet architecture includes a feature extractor, lane encoder, lane decoder, and prediction head, leveraging components from ResNet-50, BEVFormer, and various attention mechanisms. We experimented with optimizations to the LaneSegNet architecture through feature extractor modification and transformer encoder-decoder stack modification. We found that modifying the encoder and decoder stacks offered an interesting tradeoff between training time and prediction accuracy, with certain combinations showing promising results. Our implementation, trained on a single NVIDIA Tesla A100 GPU, found that a 2:4 ratio reduced training time by 22.3% with only a 7.1% drop in mean average precision, while a 4:8 ratio increased training time by only 11.1% but improved mean average precision by a significant 23.7%. These results indicate that strategic hyperparameter tuning can yield substantial improvements depending on the resources of the user. This study provides valuable insights for optimizing LaneSegNet according to available computation power, making it more accessible for users with limited resources and increasing the capabilities for users with more powerful resources. © 2024, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Highresolution Remote Sensing image Classification With Limited Training Data

Highresolution Remote Sensing Image Classification With Limi...

引用

Iranian Machine Vision and image processing (MVIP)

作者： Mehdi Ariaei Hassan Ghassemian Maryam Imani Image Processing and Information Analysis Lab Tarbiat Modares University Tehran Iran

Accurate classification of land cover from aerial images is one of the research topics in remote sensing and is also in high demand in industry. However, obtaining labeled data for training different classifiers that heavily depend on supervision is still a challenging and resource-intensive task. Unsupervised methods have emerged as a powerful alternative to overcome the limitations associated with labeled data. Such methods have a high ability to discover hidden patterns and structures in multi-spectral images and have the possibility of classifying various types of land cover without relying on labeled samples. Our research primarily involved the analysis of World-View3 satellite imagery. Our strategy involved creating an advanced pipeline that extracted features using autoencoders. Through this approach, the multispectral images' key characteristics are efficiently extracted. Subsequently, we implement transfer learning to re-train the model with a limited number of labeled data. By applying transfer learning, our pipeline significantly enhances the capability of multispectral image processing, enabling a more comprehensive and accurate interpretation of satellite imagery data. Finally, we evaluate our results not only by providing a confusion matrix but also through a visual comparison between the class map and the RGB composition of the MSI image.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Knowledge-guided EEG Representation Learning 46

Knowledge-guided EEG Representation Learning

引用

46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2024

作者： Kommineni, Aditya Avramidis, Kleanthis Leahy, Richard Narayanan, Shrikanth University of Southern California Signal Analysis and Interpretation Lab Los AngelesCA90089 United States University of Southern California Signal and Image Processing Institute Los AngelesCA90089 United States

ISBN: (纸本)9798350371499

Self-supervised learning has produced impressive results in multimedia domains of audio, vision and speech. This paradigm is equally, if not more, relevant for the domain of biosignals, owing to the scarcity of labelled data in such scenarios. The ability to leverage large-scale unlabelled data to learn robust representations could help improve the performance of numerous inference tasks on biosignals. Given the inherent domain differences between multimedia modalities and biosignals, the established objectives for self-supervised learning may not translate well to this domain. Hence, there is an unmet need to adapt these methods to biosignal analysis. In this work we propose a self-supervised model for EEG, which provides robust performance and remarkable parameter efficiency by using state space-based deep learning architecture. We also propose a novel knowledge-guided pre-training objective that accounts for the idiosyncrasies of the EEG signal. The results indicate improved embedding representation learning and downstream performance compared to prior works on exemplary tasks. Also, the proposed objective significantly reduces the amount of pre-training data required to obtain performance equivalent to prior works. © 2024 IEEE.

关键词： Self-supervised learning

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Classification of Remotely Sensed High resolution images using RP-CNN

Unsupervised Classification of Remotely Sensed High resoluti...

引用

Iranian Machine Vision and image processing (MVIP)

作者： Maedeh Fallahreyhani Hassan Ghassemian Maryam Imani Image Processing and Information Analysis Lab Faculty of Electrical Tarbiat Modares University Tehran Iran

In recent years, various deep learning frameworks have been developed for the classification of remotely sensed images. However, the network models proposed in these frameworks exhibit high complexity and do not yield high classification accuracy when applied to unlabeled scenarios. This paper introduces a Multi spectral image (MSI) classification approach that combines the random patches network with self-supervised branch (RPSS) to extract informative deep features. The proposed method involves convolving image bands with random patches to obtain multi-level deep features. Subsequently, we use panchromatic image (PAN) to extract spatial features. The MS spectral features, the derived RPSS features and spatial features then merged to classify the MSI using a support vector machine (SVM) classifier. The experimental results on real remotely sensed images have been presented.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：