جستجوی مقالات مرتبط با کلیدواژه « neural network » در نشریات گروه « برق »

Deep learning method is used to predict the future value of load demand. Based on obtained results, a new model based on the forward-backward load shifting and unnecessary load shedding is presented. As well, to increase energy efficiency, excess renewable energy has been used to produce green hydrogen. For this purpose, GAMS optimization software has been used for optimal operation of the microgrid in the presence of renewable energy sources, battery, diesel generator, aqua electrolyzer, and fuel cell considering demand side management (DSM) restrictions. The obtained results from the proposed model of the considered microgrid show that the huge amount of excess electricity can be saved to enhance energy efficiency. This issue increases green hydrogen production that can be used for fuel cell consumption. As well, the proposed model provides lower cost of operation cost.  In addition, the diesel generator consumes lower diesel fuel.

Hyper-spectral image classification is a popular topic in the field of remote sensing.Hyperspectral images (HSI) have rich spectral information and spatial information. Traditional hyperspectral image (HSI) classification methods typically use the spectral features and do not make full use of the spatial or other features of the HSI. In general, the classification approaches classify input data by considering the spectral information of the data to produce a classification map in order to discriminate different classes of interest. The pixel-wise classification approaches classify each pixel autonomously without considering information about spatial structures, further enhancement of classification results can be obtain by considering spatial dependences between pixels. However, how to fuse and utilize spectral-spatial features more efficiently is a challenging task. So the combination of spectral information and spatial information has become an effective means to obtain good classification results. Specifically, firstly, the principal component analysis (PCA) algorithm is used to extract the first principal component in the original hyperspectral image. Secondly, the   residual network Gabor, GLCM and MP   are introduced for each band to extract the spatial information of the image. Thirdly, the image is classified by using SVM to get the final classification result. In this paper, we have used the neural network classifier in the classification of hyperspectral images by integrating spectral and spatial properties in two methods stack and the method based on binary graphs. In spite of   the traditional stack method, the use of local binary graph method to properly integrate spectral and spatial information is a desirable method for the simultaneous use of spectral information along with spatial information (Feature Fusion) in hyperspectral image classification. In each of these methods, the neural network classifier is applied to the spectral and spatial features separately and then compared with the performance of the support vector machine classifier in similar conditions. The classification results show that the proposed method can outperform other traditional   classification techniques

Nowadays the most commonly used method in various tasks of machine learning and artificial intelligence are neural networks. In spite of their different uses, neural networks and Deep neural networks (DNNs) have some vulnerabilities. A little distortion or adversarial perturbation in the input data for both additive and non-additive cases can be led to change the output of the trained model, and this could be a kind of DNN vulnerability. Despite the imperceptibility of the mentioned disturbance for human beings, DNN is vulnerable to these changes. Creating and applying any malicious perturbation named “attack”, penetrates DNNs and makes them incapable of doing the duty assigned to them. In this paper different attack approaches were categorized based on the signal applied in the attack procedure. Some approaches use the gradient signal for detecting the vulnerability of DNN and try to create a powerful attack. The other ones create a perturbation in a blind situation and change a portion of the input to create a potential malicious perturbation. Adversarial attacks include both black-box and White-box situations. White-box situation focuses on training loss function and the architecture of the model but black box situation focuses on the approximation of the main model and dealing with the restriction of the input-output model request. Making a deep neural network resilient against attacks is named “defense”. Defense approaches are divided into three categories. One of them tries to modify the input, the other one makes some changes in the developed model and also changes the loss function of the model. In the third defense approach some networks are first used for purification and refinement of the input before passing it to the main network. Furthermore, an analytical approach was presented for the entanglement and disentanglement representation of inputs of the trained model. The gradient is a very powerful signal usually used in learning and an attacking approaches. Besides, adversarial training is a well-known approach in changing a loss function method to defend against adversarial attacks. In this study the most recent research on the vulnerability of DNN through a critical literature review was presented. Literature and our experiments indicate that the projected gradient descent (PGD) and AutoAttack methods are successful approaches in the l2 and l∞  bounded attacks, respectively. Furthermore, our experiments indicate that AutoAttack is much more time-consuming than the other methods. In the defense concept, different experiments were conducted to compare different attacks in the adversarial training approaches. Our experimental results indicate that the PGD is much more efficient in adversarial training than the fast gradient sign method (FGSM) and its deviations like MIFGSM and covers a wider range of generalizations of the trained model on predefined datasets. Furthermore, AutoAttack integration with adversarial training works well, but it is not efficient in low epoch numbers. Aside from that, it has been proven that adversarial training is time-consuming. Furthermore, we released our code for researchers or individuals interested in extending or evaluating predefined models for standard and adversarial machine learning projects. A more detailed description of the framework can be found at https://github.com/khalooei/Robustness-framework .

The purpose of this paper's prediction of survival or death within 30 days is based on a cerebral hemorrhage. Timely and correct diagnosis and treatment of cerebral hemorrhage are essential. If the patient's death is predicted during these thirty days, the treating physician should use intensive care and more treatment for the patient. Cerebral hemorrhages require immediate treatment and rapid and accurate diagnosis. In this article, using the volume of cerebral hemorrhage and the patient's age and using the neural network of support vector machine (SVM), it is predicted what percentage of people with cerebral hemorrhage survive and what percentage die. Parameters of cerebral hemorrhage volume and, age of patients, neural network input are considered. The network's output is the survival or death of patients with cerebral hemorrhage over the next thirty days. The data we used included the bleeding volume and age of 66 patients with lobar hemorrhage, 76 patients with deep bleeding, nine patients with Pontine hemorrhage and 11 patients with cerebellar hemorrhage. All bleeding models are considered as input to the support vector machine neural network. The overall accuracy of the designed support vector machine neural network is 93%. Regardless of the type of cerebral hemorrhage, the survival or death of people with cerebral hemorrhage within 30 days is predicted.

Historical documents are always of interest to historians and linguists. Important documents are usually digitized by segmentation and identification methods. Digitization of documents is very important for research on these documents and their protection. This article proposes a general classification and recognition framework for the images of Persian historical documents. First, pre-processing is done on documents by removing noises, removing skew, removing stamps, etc., and the document image becomes a two-level image. In the second step, a method of dividing the text of the document into lines is proposed. In the third stage, a method of dividing lines into sub-words of Persian script is presented and the sub-words of these documents are extracted, then deep networks are used to train frequent sub-words and recognize them, and the results are reported based on different criteria.

Due to developments have been occurred in ECCM techniques of radars, pulse separation methods in ESM systems rely on intera-pulse modulation instead of analyzing pulse descriptive words in traditional methods. Extraction of pulse modulation is a suitable method but in the case of multi-mode radars, the number of targets is overestimated by changing the intra-pulse modulation. The purpose of this paper is to detect multi-mode radars with various types of internal modulation in a dense radar environment. The proposed solution is to add multi-mode radars detection to the existing pulse separation methods at the post processing stage. This method involves providing an appropriate framework for examining the separated pulse string by defining and selecting similarity criteria from the extended PDW. In this method, at first the distinguishing features of each radar are extracted and the similarity criteria of each feature are calculated to check the similarity between the two pulse strings. Input data contains information separated from real radars received by ESM system. Due to the uncertainties of each criteria, similarity scores are computed through the fuzzy roles and normalization and training dataset is formed. The data table is then used to train a perceptron neural network. The trained network can detect multimode radars automatically. To test the network, a section of the data table is applied to the network and trained network succeed in 100% of test data to distinguish multi-mode radars from the distinct radars.