artificial neural network

The Internet of Things is a network of physical devices and equipment that includes sensors, software, and other technologies for exchanging data with other devices and systems over the Internet. The spread of the Internet of Things in the fields of smart health, smart agriculture, smart city, smart home, has revolutionized human life. Given the importance of the Internet of Things, identifying anomalies and malicious traffic is essential to maintaining privacy, network stability, and blocking unwanted behaviors. Due to the limited resources on IoT devices, traditional methods cannot be used directly to secure IoT devices and networks. To solve this problem, an artificial neural network-based identification method and in-depth learning has been developed to identify malformations and malicious traffic about which there is no predefined information. The data set used in this method is a combination of malicious and healthy traffic collected from related sources and feature extraction manually. Deep artificial neural network was applied to the data set and preprocessed and the results were analyzed with some conventional machine learning algorithms. The results show that the model designed using neural network and deep learning is able to detect anomalies and malicious traffic in the Internet of Things with an accuracy rate of more than 98.9% and an accuracy rate of 99.3%. In addition, the detection speed is 1.7 times faster than machine learning algorithms.

Deflecting missile’s radar guidance or missile’s seeker by chaff is a common and effective defensive method which is used in military vessels. To counter this defensive measure, methods for recognition targets from chaff have been developed, which generally focus on the special features of chaff or target. These features should be able to perform properly in different operating conditions of the radar or different environmental conditions that change the behavior of the radar. But there is no effective feature that can distinguish target from chaff with appropriate accuracy in all conditions, and different features do not have the same performance in different environmental conditions or radar working parameters such as different waveforms and as a result their performance changes. In this article, by using artificial neural network, a structure is presented for detecting chaff and target in a radar, whose performance in different environmental conditions and waveforms has been better than the existing methods and significantly improved the accuracy of target detection from chaff and led to appropriate accuracy. Also, to improve the performance of the radar with a cognitive approach, its transmitted waveform is optimally selected and changed at each stage. For this purpose, a feedback neural network with LSTM layers has been used, which suggest the optimal waveform according to changes in the environment. The general structure of the proposed method is so that first of all, by using pre-processing on the received radar data, the features of symmetry, Doppler spread and AGCD are extracted, which contain information that separates the target from the chaff. Then, to remove the effect of noise on these features, thresholding is used. Finally, these features are used to correctly distinguish the target from the chaff in a feed-forward neural network with fully connected layers. On the other hand, in each step, by using the waveform suggestion network, the optimal waveform is selected and used for the next moment. Thus, the proposed structure is an intelligent machine that, in addition to recognizing the target from the signal at each moment, determines what the optimal waveform should be at the next moment. At the end, the effectiveness of this method in comparison to the previous methods, that is, thresholding on the characteristics of symmetry, Doppler and AGCD in distinguishing the target from the chaff is evaluated. It is observable the performance of the proposed system has made a significant improvement.

Identifying and classifying marine vehicles is of great significance in military applications. Traditionally, this task was performed by sonar officers based on acoustic signals received by sonar hydrophones. Nowadays, developing systems that perform this task automatically and with minimal human intervention is inevitable. In this study, using different augmentation methods, the size of this dataset is expanded under different scenarios. With the help of signal processing and machine learning knowledge, different features are extracted and next these features are classified using different classification algorithms. Finally, the performance of different scenarios is evaluated by calculating different measures. Results of this study show the success of the Mel-Frequency Cepsteral Coeffiecent (MFCC) feature extraction method when using various classification algorithms, including Artificial Neural Network (ANN) and also the failure of Convolutional Neural Network (CNN) when training with small datasets. The overall comparison reveals that due to the small number of samples in the dataset ANN performs the best.

As source and channel coding are performed independently from each other without any feedback, source coding tries to remove redundancy of the information whereas channel coding tries to increase reliability of transmitted data. As known, channel capacity restricts volume of the transmitted data and so depending to conditions, it is required to have a tradeoff between source and channel coding. The aim of this paper is to improve the quality of the synthesized video by increasing the robustness against channel errors in fixed transmission rate. In other words, the robustness of the transmitted video frames increases without any increment in bit rate. This results in improvement in the quality of the synthesized video. In the proposed method, the transmitted information is considerably compressed using neural network with Huffman coding in H. 264. Then a secondary channel coding whose rate depends on the amount of the compression is applied on the compressed information. This causes that the proposed method is able to increase channel coding rate and therefore provides higher protection for the transmitted information and more robustness against channel errors. The obtained results by the proposed method are compared to the other methods for different source coding rates and SNRs.