Search

Currency has become absolutely necessary part in human life. At present recognition of counterfeit and real currency notes is very important. If the application learns to classify the denominations of the currency and detect the counterfeit currency then it will improve the accuracy of the recognition and confidence of the people. Implementation of deep learning techniques on various applications is increasing rapidly, and now many researchers are using deep learning techniques for the applications related to currency notes. In this paper various techniques which have been used for extracting the features from currency note, recognizing the currency note and detecting the counterfeit note have been presented. [ABSTRACT FROM AUTHOR]

In this paper, we develop a risk based predictive analytics model utilizing the available dataset to train the model. To predict the likelihood of a new strain of Covid-19 syndrome in patients, the researchers used a deep learning classifier called Recurrent Neural Network (RNN). The study considers diabetes patients as the respondents and the data is collected from the diabetes patients. The risk of covid-19 effects on diabetes patients are deeply analyzed using RNN. The collected datasets are initially pre-processed and then the features are extracted with final classification using RNN. The experimental analysis is conducted to validate the efficacy of the predictive analytics using RNN. The findings indicate that the suggested RNN outperforms other approaches for forecasting covid-19 risk in diabetic patients. [ABSTRACT FROM AUTHOR]

Text Summarization is one of the most experimental tasks in deep learning that decreases the size of a text while it's important to keep its core and key information. Previous work in this domain is going to give remarkable results by using highly dependent data. Natural language processing-related works are trending due to complications in the implementation. Abstractive text summarization is a critical idea in natural language processing and also it's a trending, hardest problem that varies the solution or outcome based on the dataset. The objective of the paper is to rewrite into well-simplified and core information in text format that is retrieved from files or documents or datasets. Abstractive text summarization using RNN is the methodology going to be implemented in our paper. Attention mechanism and LSTM are going to be add-on advantages to the model. Both the methodologies improve the performance of the model and are also used to solve the complications problem. Keyword extracting is the one of features of this paper. State-of-art is going to be used to verify the model's efficiency. BLEU and ROUGE are going to be used to identify the accuracy and validate the model. The greatest thing about the dataset is the relationship between columns is independent of each other. Finally, the aim of the paper is to generate shortened text from whole huge data (text format). These kinds of problems are often used to solve in sequence-to-sequence passion [ABSTRACT FROM AUTHOR]

The importance of medicinal plants has widely increased throughout the world, especially during the COVID-19 pandemic. It is essential for a layman to identify a medicinal plant and know about its medicinal values. In this view, this paper proposes an approach that helps a person to identify the medicinal values of different plants. The paper focuses on designing a mobile application that can identify the leaf and list its medicinal values with a mere snapshot of it by performing instance segmentation, feature extraction and classification in the backend. The models were trained and tested on a custom dataset generated by collecting around 1500 images spreading over 5 medicinal plant species. The paper uses a deep learning-based segmentation approach for real-time segmentation of the leaf from the complex natural background. The segmented leaf will undergo a feature extraction process and will be given as an input to the traditional Machine Learning algorithms. Manual Feature Extraction and Deep Learning based Feature extraction processes were implemented. Various features including shape, colour and texture were extracted in the manual feature extraction process and a MobileNet pre-trained model was used for deep learning-based feature extraction. The Mask Region-based Convolutional Neural Network (Mask R-CNN) yielded the best performance for real-time segmentation with a Dice score of 0.92 and for the classification MobileNet + Logistic Regression gave the best accuracy of 0.97 and AUC score of 0.99. [ABSTRACT FROM AUTHOR]

The crime rate has been rising at an unprecedented rate, and security has become a big concern in ATM machines. Face detection is the most common biometric technique due to its non-invasive nature. It's been used in a variety of fields, including camera auto focus, attendance, crowd monitoring, object tracking, security, system, etc. Face detection systems uses image processing techniques that are being learned to operate reliably in a variety of conditions, including changes in posture, lighting, skin color, occlusion, and face spoofing. The face detection system has become increasingly vulnerable to occlusion. Occlusion refers to the deliberate shielding of one's face with a helmet, sunglasses, scarves, or other items in order to avoid being caught. These issues have a significant impact on the development of image processing techniques and system's performance. In this paper the Haar Cascade Classifier (HCC) scheme is projected for face detection where precision as well as minimal processing time are important factors for ATM. The proposed scheme uses deep learning models such as Convolutional Neural Networks to enhance the reliability in feature extraction plus classification of images. Face biometric access control devices are becoming more common in everyday lives, but they remain vulnerable to spoofing attacks. This paper also proposes face spoofing identification using Local Binary Pattern (LBP) that has useful features for face detection. The proposed spoofing attack detection technique has yielded encouraging results. [ABSTRACT FROM AUTHOR]

The rapid developments of deep learning techniques are the process that can easily create and use a fake media known as Deepfakes. Deep learning techniques can generate facial motion transferring, gender-changing, and face swapping with others. Currently, it is possible to create hyper-realistic digital media like images and videos through deep learning techniques that are easily accessible. However, that becomes a threat for everyone, if used for harmful purposes. This work presents how deep learning is used for Deepfake generation and detection. There are two methods that are most popular for Deepfake generation: Generative Adversarial Networks (GANs) and autoencoder techniques that is presented in this work. It also, introduces a survey of some researchers' works and methods that used for Deepfake detection during the recent years. This paper proposed a method to discriminate if that video is real or fake by using a Support Vector Machine (SVM) classifier, the method presented in this work for deepfake detection is based on Gray Level Co-occurrence Matrix (GLCM) as a feature extraction method applying on 4 bands to obtain 13 feature form each one to achieve good accuracy of the classifier with a score 70.47% that batter than previous work that didn't exceed 60% of accuracy. [ABSTRACT FROM AUTHOR]

last years the risk of dying from depression has increased, Depression creates suicidal thoughts for people that cause serious disabilities in daily life. Sentiment analysis (SA) is a topic has been on research for decades, which aim to find the nature of the text and classifies into positive, negative and neutral, for today progression of technology a lot of data is available for SA from text and image. This paper aim to apply pre-processing, feature extraction and classification methods on Twitter tweets for perform SA focusing on depression. Deep Belief Network (DBN) technique used for the classification of the text data obtained from depression dataset, where the tweets are classified as positive or negative. The results are presented using standard evaluation metrics precision, recall, f-score and accuracy, which results are 99.97%, 98.47%, 99.21% and 98.77 respectively. [ABSTRACT FROM AUTHOR]

In recent years face recognition system is emerging technology in research areas. Face recognition is used in many applications such as attendance management system, People tracking system and access control system etc. Major challenges faced in face recognition are detection and recognition because it is not easy to detect multiple faces in one frame and difficult to recognize the faces with poor resolution. Therefore, the main objective of this paper is to obtain better accuracy by using the combination of FaceNet and Deep Neural Network (DNN). In this proposed system, FaceNet is used for feature extraction by embedding 128 dimensions per face and DNN is used to classify the given training data with extracted feature of FaceNet. The outcome of the system is practical, reliable and consumes less time. [ABSTRACT FROM AUTHOR]

Around the world, many different languages are spoken, each with a unique writing style. A challenging problem requiring substantial research and cutting-edge computer technologies in image processing is identifying the author of a handwritten text document. In past years, different dataset for character recognition, word recognition, and recognition of writer were reported in the literature. Keeping this in mind, the study introduces the researcher's contributions to writer identification strategies on diverse datasets and techniques. This study's primary goal is to discuss writer identification systems' work on various Indian and non-Indian languages based on various machine and deep learning techniques. The paper also discusses many stages of the identification process to assist researchers in the field of identification by giving analyses of various preprocessing approaches, feature extraction methods, and classification strategies on both Indic and non-Indic scripts. [ABSTRACT FROM AUTHOR]

Image processing domain involves various steps like preprocessing, augmentation, segmentation and recognition of images. Consistent image segmentation technique plays crucial role in computer vision domain. Using sophisticated segmentation techniques or segmentation models on image datasets resolves recognition, detection, feature extraction and classification issues in image processing and computer vision problems. Several literature surveys have shown improvement in the results after applying these segmentation techniques in the domain of machine learning as well as in deep learning. In this paper, we are analyzing mainly wheat leaf diseases by applying segmentation techniques or segmentation models in both machine learning and deep learning areas. We are also reviewing various segmentation techniques applied on other crops for comparative analysis with the wheat crops. [ABSTRACT FROM AUTHOR]

Data Clustering is a primary research focus in large data-driven application domains in the big data cloud as performance of the clustering dynamic data with high dimensionality is highly challenging due to major concern in the effectiveness and efficiency on data representation. Machine learning is a conventional approach to distribute the data into soft partition still it leads to increasing sparsity of data and increasing difficulties in distinguishing distance between data points. In addition, securing the personnel and confidential information of the user is also becoming vital. In order to tackle those issues, a new anonymization based deep privacy preserving learning paradigm has been presented in this paper. The proposed model is represented as deep privacy preserving convolutional auto encoder learning architecture for secure high dimensional data clustering on inferring the distribution of the data over time. Initially dimensionality reduction and feature extraction is carried out and those extracted feature has been taken for clustering on generation of objective function to produce maximum margin cluster. Those clusters are further fine tuned to feature refinement on the hyper parameter of various layers of deep learning model network to establish the minimum reconstruction error by feature refinement. Softmax layer minimizes the intra cluster similarity and inter cluster variation in the feature space for cluster assignment. Hyper parameter tuning using stochastic gradient descent has been enabled in the output layer to make the data instance in the cluster to be close to each other by determining the affinity of the data on new representation. It results significant increase in the clustering performance on the discriminative information's. Detailed experimental analysis has been performed on benchmarks datasets to compute the proposed model performance with conventional approaches. The performance outcome represents that anonymization based deep privacy preserving clustering learning architecture can produce good scalability and effectiveness on high dimensional data. [ABSTRACT FROM AUTHOR]

Most of the peoples are involved in watching a variety of birds to enjoy the beauty of nature and to get relief from daily life struggles. But in some cases they may not be capable of identifying the bird species correctly because of the similarity or features shared among different bird species. To find a better solution for this problem, deep learning technologies can be used. This paper has been discussed to develop a deep learning platform to recognize the bird species by utilizing its captured image. Convolutional neural network is utilized for both feature extraction and classification processes. The experimental results show that this deep learning platform outperforms the other identification techniques and the recognition using images of birds is more effective than the recognition using audio signals of birds. The developed model achieved higher classification accuracy for the training image and this method helps the amateur bird watchers to identify the bird species from the captured bird image easily. [ABSTRACT FROM AUTHOR]

Now a day there is exponential growth in music files, so the need for automatic music genre classification approach is also increasing. Current classification systems of music genres developed using machine learning algorithms and some of the limitations of existing music genre systems are doesn't work with large database, low accuracy etc. In this paper, presented a large music dataset namely GTZAN with 10 distinct music genres like jazz, rock, hip-hop etc. In order train the classification model of music genre system, a deep learning based technique called CNN (Convolution Neural network) is applied. The proposed method performs the classification of music genre into innumerable genre with the help of feature vector extraction, which plays a vital role in training the model. The feature vector called MFCC (Mel Frequency Cepstral Coefficient) considered in classifying the raw files which are converted to respective music genres or types. The performance of various classification techniques are analyzed compared with respect to evaluation metrics. [ABSTRACT FROM AUTHOR]

Implementation of Machine Learning/Deep Learning techniques in Image Processing have paved way to advanced object detection, classification and tracking in a real time environment. Recent advancements are enabling these techniques to be used in the field of real time Infra Red (IR) object detection. IR image processing is being used extensively in the field of military, industrial and medical applications. This paper emphasizes upon the various Deep learning techniques which can be used for automatic target detection/recognition in infra red maritime imagery of sea borne military assets. Our review begins with the introduction to Object Detection, IR imaging, techniques available and various architectures for feature extraction and object detection. Then, we also provide a comparison of performance/detection capabilities on widely used PASCAL VOC 2007 and 2012 datasets. [ABSTRACT FROM AUTHOR]

Heart failure is a diagnostic condition characterized by numerous heart problems at advanced stages, and it is sometimes referred to as chronic heart failure (CHF). Heart sound signals have been extensively researched over the past several decades. In a variety of clinical applications, automated heart sound classification and segmentation approaches offer the ability to screen for diseases in the heart sound. Despite this, the unavailability of a large and open data set of heart sound recordings has made it difficult to do comparative analyses of algorithms in the literature. There are numerous approaches to feature extraction, machine learning (ML), deep learning (DL) classification that are addressed in this paper. Feature extraction can be accomplished using both linear and nonlinear approaches. In linear approaches time and frequency domain features are considered. However, nonlinear methodologies such as correlation dimension, greatest lyapunov exponent, wide multi fractal spectrum, and sample entropy are used to explain the chaos, fractal properties, and complexity of heart sound signals, as opposed to linear approaches. The study's goal is to perform overview on techniques and methodologies used for detection of CHF based on heart sound. In addition, a brief description of a baseline categorization method is provided in this section. [ABSTRACT FROM AUTHOR]

To detect the objects around an autonomous vehicle is very essential to operate safely. This paper presents to detect and classify the objects for assisting autonomous driving. In autonomous driving systems, the task of object detection itself is one of the most important prerequisites to autonomous navigation. Deep learning one of the computer vision tasks, perform object detection very effectively than compared to earlier methods and this project is to detect the objects like vehicles, persons, traffic lights, etc. In this work, an approach to object detection in deep learning that makes the bounding box for an image to predict is explored. Object detection is the method of detecting the objects present in a given image. Apart from detecting the number of objects present in an image it also specifies in which location that object is present in the image. The objects are detected by means of bounding boxes. In the existing system algorithm like Convolutional Neural Network (CNN) using Resnet-50 were used to detect the objects like vehicles, persons, traffic lights separately. The problem identified here is, in the existing system the camera is fixed in a particular place and it detects objects only if the objects come into the camera frames. It is not detecting both objects and lanes simultaneously when the autonomous vehicle is in motion at any location. To overcome these problems, in the proposed system object detection is performed by mounting camera in front of the moving vehicle. You Only Look Once (YOLO) V3 Algorithm is used for the process of object detection. Compared to earlier detection approaches YOLO V3 shows improvement in detection accuracy. It provides good feature extraction and detection in large-scale. The proposed YOLO al gorithm has better average precision value for detecting all objects than compared to existing CNN using Resnet-50. In addition YOLO V3 algorithm performs lane detection apart from object detection. [ABSTRACT FROM AUTHOR]

Signature forgery detection is the important and most hot topic in forensic department, simultaneously number of cases for forgery signatures are increasing. The spotting of forgery documents is crucial evidence in illegal activities. In this work, convolution neural networks approach is taken for processing the database. The problem of monotonous and low degree of automation of the manual document inspection, put forward to the method for handwritten tampered Signature identification based on the concept of convolutional neural networks. In this type of experiment, we are detecting the difference between the original and false signature/fake signature (forged signature). Here we are extracting the basic features of the real signature by splitting it into 25 layers (can be increased) and crosschecking them with the fake signature from the data set provided. In addition to that, we can add as many signatures as we want to the dataset for identifying the forged signature. [ABSTRACT FROM AUTHOR]

Using computer vision to classify and detect acne lesions, as well as segment wrinkles, is important because of its potential for early skin disease diagnosis and age estimation. The images analyzed in the past mainly included dermoscopy images or face portion images. However, the detection of skin conditions on the face from a face/half body image has never been investigated before. With the success of deep learning in object detection, we propose a deep learning model for acne lesion and wrinkle detection from the face/half body image. The state-of-the-art deep learning networks, Faster Region Convolutional Neural Network, and Residual Network are employed in the study. The features of the image are extracted with 50 layers of Residual Network to produce a conv feature map. From the map, the object proposal is detected with Regional Proposal Network. Finally, the feature map and the proposal are used to classify and localize the region of interest (RoI) in the image. The multiple RoIs in a single image are classified as acne or wrinkle. The detection model achieved a 47.96% mean Average Precision score, therefore, the precision of the proposed method is proven. Instead of using the dermoscopy images to diagnose skin cancer and estimate age, by integrating our proposed model to extract acne or wrinkle from the face/half body image, the diagnosis and estimation works can be analyzed using the face/half body image. [ABSTRACT FROM AUTHOR]

Emotions are elementary for humans, impacting perception and everyday activities like communication, learning and decision-making. Speech emotion Recognition (SER) systems aim to facilitate the natural interaction with machines by direct voice interaction rather than exploitation ancient devices as input to know verbal content and build it straightforward for human listeners to react. During this SER system primarily composed of 2 sections called feature extraction and feature classification phase. SER implements on bots to speak with humans during a non-lexical manner. The speech emotion recognition algorithm here is predicated on the Convolutional Neural Network (CNN) model, that uses varied modules for emotion recognition and classifiers to differentiate feelings like happiness, calm, anger, neutral state, sadness, and fearful. The accomplishment of classification is predicated on extracted features. Finally, the emotion of a speech signal will be determined. [ABSTRACT FROM AUTHOR]

Autonomous cars must take real-time decisions about surroundings to reduce death rates during traffic accidents. Traffic related information's are available in the Traffic sign. It assists to drive better and safer. In the traditional method, the traffic sign is detected by manually or computer vision methods. Those are time consuming processes.No-one will be aware of all traffic signs, so it will let everyone know and learn the signs easily. Traffic sign recognition is just one of the problems that computer vision and deep learning can solve Convolution Neural Network (CNN) architecture.Machinesareable-to identify traffic signs from the German Traffic Signal Recognition Benchmark (GTSRB) dataset that contains forty-three classes. The proposed system has three working stages: image pre-processing, detection, and recognition. Initially, the traffic sign image is pre-processed, and the detailed information present in the traffic sign image is detected by using the histogram equalization method, which improves the contrast of the traffic sign image. After preprocessing, the features of the images are extracted by using CNN architecture with three non-linear activation functions such as Re-Lu, Leaky-Re-Lu and sigmoid. The experimental results compare the results of the above three non-linear activation functions.The activation function Re-Lu and Leaky Re-Lu achieved accuracy above 95%.After feature extraction, the output layer is used to predict the traffic sign images. [ABSTRACT FROM AUTHOR]