یادگیری عمیق

Open Access Article

1 - A Dynamic Sequential Approach Using Deep Learning to Improve the Performance of Biometrics Match on Card Systems
Mohammad Sabri Mohammad Moin Farbod Razzazi

20.1001.1.16823745.1399.18.1.15.5

Nowadays, the threats such as terrorism and cybercrime are extremely increased, therefore, the identity authentication process is very substantial for the national security of a country. In this paper, we propose a novel multimodal authentication system with sequential More

Nowadays, the threats such as terrorism and cybercrime are extremely increased, therefore, the identity authentication process is very substantial for the national security of a country. In this paper, we propose a novel multimodal authentication system with sequential structure based on deep learning. In the proposed method, feature vectors are extracted automatically through deep network with an end to end architecture. A multi biometric system using two fingerprint and a face is implemented and evaluated. The results demonstrate that, the authentication is done by fingerprints in 91.42% cases and only for 8.58% cases the face modal is required. In addition, the proposed method is more accurate than first and second fingerprint by 35% and 30% at FMR=0.001, respectively. As a result, we augmented the accuracy of the system and at the same time reduced the acquisition and matching time. This conducts to the improvement of user convenience and security of the service provider, simultaneously. The achievements of this work can be used to increase the effectiveness of authentication process and can play an important role in the acceptability of real world applications. Manuscript profile

Open Access Article

2 - Spatio-Temporal Prediction of Vegetation Dynamics Based on Remote Sensing Data Using Deep Learning
Elham Zangeneh H. Mashayekhi Saeed Gharachelo

20.1001.1.16823745.1401.20.4.2.2

Understanding and analyzing spatial-temporal data changes is very important in various applications, including the protection and development of natural resources. In the past studies, Markov process and comparison-based methods were mainly used to predict the changes o More

Understanding and analyzing spatial-temporal data changes is very important in various applications, including the protection and development of natural resources. In the past studies, Markov process and comparison-based methods were mainly used to predict the changes of vegetation indices, whose accuracy still needs improvement. Although time series analysis has been used to predict some indices, the method to extract these indices from remote sensing data and model their sequences with deep learning is rarely observed. In this article, a method for predicting changes in plant indices based on deep learning is presented. The research data includes Landsat satellite images from 2000 to 2018, related to four seasons in the north and east of Shahrood city in Semnan province. The time span of the extracted images makes it possible to predict changes in vegetation cover. Vegetation indices extracted from the data set are NDVI, SAVI and RVI. After performing atmospheric corrections on the images, the desired indicators are extracted and then the data is converted into a time series. Finally, the modeling of the sequence of these data is performed by the Short-Long-Term Memory (LSTM) network. The results of the experiments show that the deep network is able to predict future values with high accuracy. The amount of the model error without additional data is 0.03 for the NDVI index, 0.02 for the SAVI index, and 0.06 for the RVI index. Manuscript profile

Open Access Article

3 - Detecting Human Activities Based on Motion Sensors in IOT Using Deep Learning
Abbas Mirzaei fatemeh faraji

20.1001.1.16823745.1400.19.4.10.3

Control of areas and locations and motion sensors in the Internet of Things requires continuous control to detect human activities in different situations, which is an important challenge, including manpower and human error. Permanent human control of IoT motion sensors More

Control of areas and locations and motion sensors in the Internet of Things requires continuous control to detect human activities in different situations, which is an important challenge, including manpower and human error. Permanent human control of IoT motion sensors also seems impossible. The IoT is more than just a simple connection between devices and systems. IoT information sensors and systems help companies get a better view of system performance. This study presents a method based on deep learning and a 30-layer DNN neural network for detecting human activity on the Fordham University Activity Diagnostic Data Set. The data set contains more than 1 million lines in six classes to detect IoT activity. The proposed model had almost 90% and an error rate of 0.22 in the evaluation criteria, which indicates the good performance of deep learning in activity recognition. Manuscript profile

Open Access Article

4 - An Intelligent Vision System for Automatic Forest Fire Surveillance
Mohammad Sadegh Kayhanpanah Behrooz Koohestani

20.1001.1.16823745.1401.20.1.3.7

Fighting forest fires to avoid their potential dangers as well as protect natural resources is a challenge for researchers. The goal of this research is to identify the features of fire and smoke from the unmanned aerial vehicle (UAV) visual images for classification, o More

Fighting forest fires to avoid their potential dangers as well as protect natural resources is a challenge for researchers. The goal of this research is to identify the features of fire and smoke from the unmanned aerial vehicle (UAV) visual images for classification, object detection, and image segmentation. Because forests are highly complex and nonstructured environments, the use of the vision system is still having problems such as the analogues of flame characteristics to sunlight, plants, and animals, or the smoke blocking the images of the fire, which causes false alarms. The proposed method in this research is the use of convolutional neural networks (CNNs) as a deep learning method that can automatically extract or generate features in different layers. First, we collect data and increase them according to data augmentation methods, and then, the use of a 12-layer network for classification as well as transfer learning method for segmentation of images is proposed. The results show that the data augmentation method used due to resizing and processing the input images to the network to prevent the drastic reduction of the features in the original images and also the CNNs used can extract the fire and smoke features in the images well and finally detect and localize them. Manuscript profile

Open Access Article

5 - Efficient Recognition of Human Actions by Limiting the Search Space in Deep Learning Methods
m. koohzadi N. Moghadam

20.1001.1.16823745.1400.19.3.6.7

The efficiency of human action recognition systems depends on extracting appropriate representations from the video data. In recent years, deep learning methods have been proposed to extract efficient spatial-temporal representations. Deep learning methods, on the other More

The efficiency of human action recognition systems depends on extracting appropriate representations from the video data. In recent years, deep learning methods have been proposed to extract efficient spatial-temporal representations. Deep learning methods, on the other hand, have a high computational complexity for development over temporal domain. Challenges such as the sparsity and limitation of discriminative data, and highly noise factors increase the computational complexity of representing human actions. Therefore, creating a high accurate representation requires a very high computational cost. In this paper, spatial and temporal deep learning networks have been enhanced by adding appropriate feature selection mechanisms to reduce the search space. In this regard, non-online and online feature selection mechanisms have been studied to identify human actions with less computational complexity and higher accuracy. The results showed that the non-linear feature selection mechanism leads to a significant reduction in computational complexity and the online feature selection mechanism increases the accuracy while controlling the computational complexity. Manuscript profile

Open Access Article

6 - Iranian Dastgah Music Recognition Based on Notes Sequence Extraction and Use of LSTM Networks
سینا غضنفری پور M. Khademi Abbas Ebrahimi moghadam

20.1001.1.16823745.1401.20.2.6.2

Iranian "Dastgah" music classification by computer is a very interesting yet complex and challenging topic for those who are interested in Iranian Dastgah music. The aforementioned problem is important, firstly, due to its many applications in different areas such as co More

Iranian "Dastgah" music classification by computer is a very interesting yet complex and challenging topic for those who are interested in Iranian Dastgah music. The aforementioned problem is important, firstly, due to its many applications in different areas such as composing and teaching music, and secondly, because of the needs of ordinary people to computer to detect the Dastgah. This paper presents a method for recognition of the genre (Dastgah) and subgenre (sub-Dastgah) of Iranian music based on sequential note extraction, hierarchical classification, and the use of LSTM networks. In the proposed method, the music track is first classified into one of the three general categories. The first category includes only "Mahour" Dastgah, the second category includes "Shour" and "Nava", and the third category includes "Homayoun", "Segah" and "Chahargah". Then, for each category, depending on its type, a different number of classifiers are applied until one of the 6 Dastgah and 11 sub-Dastgah of Iranian music are recognized. This research is not limited to any particular style of playing or instruments, it is also not affected by neither the speed nor the techniques of player. The labeled tracks in the "Arg" database, which is created for this research, are solo. However, some of them are also played by percussion instruments (such as the Tombak) along with melodic instruments. The results show that recognition of 6 main Dastgah and 11 sub-Dastgah have been approved by an average accuracy of 74.5% and 66.35%, respectively, which is more promising compared to other few similar studies. Manuscript profile

Open Access Article

7 - Generation of Persian sentences By Generative Adversarial Network
Nooshin riahi Sahar Jandaghy

20.1001.1.16823745.1402.21.1.10.1

Text generation is a field of natural language processing. Text generation enables the system to produce comprehensive, .grammatically correct texts like humans. Applications of text generation include image Captioning, poetry production, production of meteorological re More

Text generation is a field of natural language processing. Text generation enables the system to produce comprehensive, .grammatically correct texts like humans. Applications of text generation include image Captioning, poetry production, production of meteorological reports and environmental reports, production of business reports, automatic text summarization, .With the appearance of deep neural networks, research in the field of text generation has change to use of these networks, but the most important challenge in the field of text generation using deep neural networks is the data is discrete, which has made gradient inability to transmit. Recently, the use of a new approach in the field of deep learning, called generative adversarial networks (GANs) for the generation of image, sound and text has been considered. The purpose of this research is to use this approach to generate Persian sentences. In this paper, three different algorithms of generative adversarial networks were used to generate Persian sentences. to evaluate our proposed methods we use BLEU and self-BLEU because They compare the sentences in terms of quality and variety. Manuscript profile

Open Access Article

8 - Provide a Personalized Session-Based Recommender System with Self-Attention Networks
Azam Ramazani A. Zareh

20.1001.1.16823745.1401.20.3.3.1

Session-based recommender systems predict the next behavior or interest of the user based on user behavior and interactions in a session, and suggest appropriate items to the user accordingly. Recent studies to make recommendations have focused mainly on the information More

Session-based recommender systems predict the next behavior or interest of the user based on user behavior and interactions in a session, and suggest appropriate items to the user accordingly. Recent studies to make recommendations have focused mainly on the information of the current session and ignore the information of the user's previous sessions. In this paper, a personalized session-based recommender model with self-attention networks is proposed, which uses the user's previous recent sessions in addition to the current session. The proposed model uses self-attention networks (SANs) to learn the global dependencies among all session items. First, SAN is trained based on anonymous sessions. Then for each user, the sequences of the current session and previous sessions are given to the network separately, and by weighted combining the ranking results from each session, the final recommended items are obtained. The proposed model is tested and evaluated on real-world Reddit dataset in two criteria of accuracy and mean reciprocal rank. Comparing the results of the proposed model with previous approaches indicates the ability and effectiveness of the proposed model in providing more accurate recommendations. Manuscript profile

Open Access Article

9 - Stock Price Movement Prediction Using Directed Graph Attention Network
Alireza Jafari Saman Haratizadeh

20.1001.1.16823745.1402.21.1.13.4

Prediction of the future behavior of the stock market has always attracted researchers' attention as an important challenge in the field of machine learning. In recent years deep learning methods have been successfully applied in this domain to improve prediction perfor More

Prediction of the future behavior of the stock market has always attracted researchers' attention as an important challenge in the field of machine learning. In recent years deep learning methods have been successfully applied in this domain to improve prediction performance. Previous studies have demonstrated that aggregating information from related stocks can improve the performance of prediction. However, the capacity of modeling the stocks relations as directed graphs and the power of sophisticated graph embedding techniques such as Graph Attention Networks have not been exploited so far for prediction in this domain. In this work, we introduce a framework called DeepNet that creates a directed graph representing how useful the data from each stock can be for improving the prediction accuracy of any other stocks. DeepNet then applies Graph Attention Network to extract a useful representation for each node by aggregating information from its neighbors, while the optimal amount of each neighbor's contribution is learned during the training phase. We have developed a novel Graph Attention Network model called DGAT that is able to define unequal contribution values for each pair of adjacent nodes in a directed graph. Our evaluation experiments on the Tehran Stock Exchange data show that the introduced prediction model outperforms the state-of-the-art baseline algorithms in terms of accuracy and MCC measures. Manuscript profile

Open Access Article

10 - Social Networks Embedding Based on the Employment of Community Recognition and Latent Semantic Feature Extraction Approaches
Mohadeseh Taherparvar Fateme Ahmadi abkenari Peyman bayat

The purpose of embedding social networks, which has recently attracted a lot of attention, is to learn to display in small dimensions for each node in the network while maintaining the structure and characteristics of the network. In this paper, we propose the effect of More

The purpose of embedding social networks, which has recently attracted a lot of attention, is to learn to display in small dimensions for each node in the network while maintaining the structure and characteristics of the network. In this paper, we propose the effect of identifying communities in different situations such as community detection during or before the process of random walking and also the effect of semantic textual information of each node on network embedding. Then two main frameworks have been proposed with community and context aware network embedding and community and semantic feature-oriented network embedding. In this paper, in community and context aware network embedding, the detection of communities before the random walk process, is performed through using the EdMot non-overlapping method and EgoNetSplitter overlapping method. However, in community and semantic feature-oriented network embedding, the recognition of communities during a random walk event is conducted using a Biterm topic model. In all the proposed methods, text analysis is examined and finally, the final display is performed using the Skip-Gram model in the network. Experiments have shown that the methods proposed in this paper work better than the superior network embedding methods such as Deepwalk, CARE, CONE, and COANE and have reached an accuracy of nearly 0.9 and better than other methods in terms of edge prediction criteria in the network. Manuscript profile

Open Access Article

11 - A Fast and Lightweight Network for Road Lines Detection Using Mobile-Net Architecture and Different Loss Functions
Pejman Goudarzi milad Heydari Mehdi Hosseinpour

By using the line detection system, the relative position of the self-driving cars compared to other cars, the possibility of leaving the lane or an accident can be checked. In this paper, a fast and lightweight line detection approach for images taken from a camera ins More

By using the line detection system, the relative position of the self-driving cars compared to other cars, the possibility of leaving the lane or an accident can be checked. In this paper, a fast and lightweight line detection approach for images taken from a camera installed in the windshield of cars is presented. Most of the existing methods consider the problem of line detection in the form of classification at the pixel level. These methods despite having high accuracy, suffer from two weaknesses of having the high computational cost and not paying attention to the general lines content information of the image (as a result, they cannot detect if there is an obstacle). The proposed method checks the presence of lines in each row by using the row-based selection method. Also, the use of Mobile-net architecture has led to good results with fewer learning parameters. The use of three different functions as cost functions, with different objectives, has resulted in obtaining excellent results and considering general content information along with local information. Experiments conducted on the TuSimple video image collection show the suitable performance of the proposed approach both in terms of efficiency and especially in terms of speed. Manuscript profile

Open Access Article

12 - Semantic Word Embedding Using BERT on the Persian Web
shekoofe bostan Ali-Mohammad Zare-Bidoki mohamad reza pajohan

20.1001.1.16823745.1402.21.2.11.4

Using the context and order of words in sentence can lead to its better understanding and comprehension. Pre-trained language models have recently achieved great success in natural language processing. Among these models, The BERT algorithm has been increasingly popular More

Using the context and order of words in sentence can lead to its better understanding and comprehension. Pre-trained language models have recently achieved great success in natural language processing. Among these models, The BERT algorithm has been increasingly popular. This problem has not been investigated in Persian language and considered as a challenge in Persian web domain. In this article, the embedding of Persian words forming a sentence was investigated using the BERT algorithm. In the proposed approach, a model was trained based on the Persian web dataset, and the final model was produced with two stages of fine-tuning the model with different architectures. Finally, the features of the model were extracted and evaluated in document ranking. The results obtained from this model are improved compared to results obtained from other investigated models in terms of accuracy compared to the multilingual BERT model by at least one percent. Also, applying the fine-tuning process with our proposed structure on other existing models has resulted in the improvement of the model and embedding accuracy after each fine-tuning process. This process will improve result in around 5% accuracy of the Persian web ranking. Manuscript profile

Open Access Article

13 - Identification of Cancer-Causing Genes in Gene Network Using Feedforward Neural Network Architecture
مصطفی اخوان صفار abbas ali rezaee

Identifying the genes that initiate cancer or the cause of cancer is one of the important research topics in the field of oncology and bioinformatics. After the mutation occurs in the cancer-causing genes, they transfer it to other genes through protein-protein interact More

Identifying the genes that initiate cancer or the cause of cancer is one of the important research topics in the field of oncology and bioinformatics. After the mutation occurs in the cancer-causing genes, they transfer it to other genes through protein-protein interactions, and in this way, they cause cell dysfunction and the occurrence of disease and cancer. So far, various methods have been proposed to predict and classify cancer-causing genes. These methods mostly rely on genomic and transcriptomic data. Therefore, they have a low harmonic mean in the results. Research in this field continues to improve the accuracy of the results. Therefore, network-based methods and bioinformatics have come to the aid of this field. In this study, we proposed an approach that does not rely on mutation data and uses network methods for feature extraction and feedforward three-layer neural network for gene classification. For this purpose, the breast cancer transcriptional regulatory network was first constructed. Then, the different features of each gene were extracted as vectors. Finally, the obtained vectors were given to a feedforward neural network for classification. The obtained results show that the use of methods based on multilayer neural networks can improve the accuracy and harmonic mean and improve the performance compared to other computational methods. Manuscript profile

Current Issue

Published Issues

Menu

Browse

List of Articles یادگیری عمیق