Fabrics and garments are some of the most important utilitarian items for human beings. The textile industry is rapidly growing with various fabric models and different attractive designs. To hold up the development of the textile industry, rigid measures have to be taken to check the quality of fabric during manufacturing. Thus, skilled workers are required to screen the quality of the fabric manually. This paper focuses on designing a deep learning framework to detect various fabric types and classify the defects using artificial intelligence. The proposed work has two phases; in phase 1, the input image is preprocessed with a novel Pseudo-Convolutional Neural Network (P-CNN) having zero tunable parameters. In phase 2, a modified Convolutional Neural Network (CNN) is applied to the preprocessed image to detect and classify major fabric defects. The Convolutional Neural Network (CNN) uses appropriate hidden layers to acquire an undeniable degree of accuracy for defect classification using images. The dataset consists of five different fabric defects such as broken pick defects, pattern defects, weft yarn deformity, soiled fabrics, and plain fabric defects that are considered for training and validation of the proposed architecture. The performance of the proposed network is measured using metric parameters such as sensitivity, specificity, and accuracy. The proposed technique has high accuracy for different fabric types used for testing the creation network.

Users can connect things, systems, network, services, and, in particular, control systems using the Internet of Things (IoT). The design and implementation of an IoT-based security surveillance system employing Node MCUs and Wi-Fi network connectivity is described in this study. Adding wireless fidelity to embedded systems offers up a world of possibilities, such as worldwide monitoring and control, secure data storage, and much more. Sensor nodes and a controller part make up this surveillance system. Remote user alerts and mobility are two of the system’s primary features. When a Wi-Fi connected microcontroller is integrated with a PIR sensor, the sensor looks for object movements and sends an alert to the user via an online cloud platform (ThingSpeak). This surveillance system is made up of sensor nodes and a controller. Two of the system’s main characteristics are remote user alerts and mobility.

Recently, deep learning techniques are widely used in various computer vision applications such as pattern recognition, data classification, object detection, image enhancement, etc. Deep learning frameworks outperform classical algorithms due to their flexibility and interoperability. The main aim of this work is to remove Gaussian-Impulse noise with blind (unknown) noise densities in digital images. The proposed work has two phases; in phase 1, the noisy image is preprocessed with a novel Pseudo-Convolutional Neural Network (P-CNN) having no tunable parameters. The Pseudo-CNN is a customized three-layered network inspired by conventional con-volutional neural networks. The first-level feature detector has k number of (2d + 1) x (2d + 1) Weight Initialized Adaptive Window (WIAW) filters. The weights or coefficients of WIAW filters are initialized using the noise probabilistic distribution. The loss function is not estimated in P-CNN because there is only one forward pass and no backpropagation. The PCNN has excellent Impulse noise rejection capability. In phase 2, a modified Convolutional Neural Network (CNN) is applied to the preprocessed image to obtain a latent clean image. To improve the network performance, we have utilized residual learning and batch normalization. The high degree of localized pixel correlation established by P-CNN helps the modified CNN to learn compact representations and salient features of the input image. The proposed method not only denoises images corrupted with blind noise levels of Gaussian-Impulse noise but is also effective for removing Impulse noise in digital images. Our experi-mental results imply that the proposed method gives good qualitative and quantitative results compared to various state-of-the-art techniques. Convolutional neural networks are pertinent for parallel computation using powerful GPUs, which help to improve the denoising performance.

In this paper, we have created a voice synthesizer especially for the people who are partially paralyzed, and our aim is to retain their original voice of the targeted speaker. The model consists of orator encoder, synthesizer, spectrogram generator, and vocoder. The orator encoder is a trained model by using the voices of variety number of speakers including noisy speech and without caption. This is used to generate a stable proportional insertion vector from only a few seconds of a targeted speaker’s source speech. The synthesizer based on Tacotron 2 that is used to condition the generated Mel spectrogram based on the embedded speaker. An auto-regressive vocoder network is used to produce the waveform samples from the Mel spectrogram. Our model is very useful to retain the original voice of the speakers who have been partially paralyzed because they will be unable to raise their volumes of their voices and it is able to produce natural voice of the speaker that is unseen during the training purpose. The proposal is compared with several state-of-art methods and established a mean opinion score is 94%.

Clothing is one of the fundamental requirements for living. The fabric business is a steadily developing industry because the interest in dress will never diminish. To support the develop-ment of the clothing industry, the clothing industry needs to take rigid measures to keep up the quality of the pieces of fabric they produce. The industry needs a worker to screen the quality of the fabric using a manual fabric review framework. The goal of this article is to plan a pro-found deep learning algorithm to recognize the fabric types using computer vision. This article focuses on identification of fabric defects using convolutional neural network with the use of appropriate pooling layer, softmax layer, and rectified linear activation layer to acquire an un-deniable degree of precision. The photographs of garments with various fabric defects like fabric broken pick defect, fabric with pattern, soiled fabric, fabric weft yarn defect, and plain fabric are considered for evaluation of the architecture. The performance of the architecture is measured with various performance measures like sensitivity, specificity, and accuracy. The algorithm produces the highest accuracy of 97.5 and 100 % for the training and testing sam -ples, respectively, for soiled fabric type.

Noise removal is one of the chronic problems while dealing with the images. Such a noise level would be unacceptable in a photograph since it would be impossible even to determine the subject. Denoising plays a major role in retrieving back original signal from noisy observations. In this paper, we propose Futuristic Flask with Convolution Neural Network (FFCNN), a residual learning model in deep convolutional neural network which is trained with large dataset, showing up excellent results for removing Gaussian noise from digital Images. FFCNN is designed to offer presentation metrics for a user in addition to performance measurements for neural model training, using the "Flask" microweb framework to enable interactions. The architecture's feed-forward denoising neural network structure performs discriminative learning for picture denoising, specifically for Additive White Gaussian Noise (AWGN) with a specified noise level and also for blind Gaussian noise. The proposed algorithm is optimized and speeded by layers of batch normalization with GPU computing. The resulting PSNR and SSIM obtained are excellent proving efficiency and effectiveness of the model for several general image denoising task.

Statistical disease modeling, data analysis and planning of resources are important aspects for overcoming the challenges of COVID-19 The proposed pandemic modelling involves pre-screening of virus affected peoples and resource allocation using deep learning frameworks Since most of the virus infected peoples are asymptomatic in nature, hence it is very difficult to recognize and isolate them from the society In this paper a new pre-screening methodology is introduced to classify peoples who are more likely to be virus infected based on image analytics The pre-screening techniques consists of classifying x ray images and coughing sounds by using Convolutional Neural Networks(CNN) The pre-screening results of deep learning frame works are used to prepare a risk score, i e , higher risk score higher probability of infected and vice-versa The proposed method has good classification accuracy for predicting various lung diseases and also can be used in pre-screening covid infected individuals However, results alone cannot be used to confirm the COVID-19 virus, whereas it helps to distinguish people more prone to get virus infected based on a risk score The real time results with an accuracy of 90% indicate the competency of proposed technique © 2021, Universitatea de Vest Vasile Goldis din Arad All rights reserved

Vocal cord paralysis is a common problem faced by individuals, where the vocal cord fails to reverberate to produce sound waves. As a result, they are unable to speak out as they were speaking before. The proposed method is designed for aiding unilateral paralyzed peoples whose vocal cord fails to give the desired reverberations. The proposed system consists of voice-to-text and text-to-voice conversions. The voice of the paralyzed person is artificially reproduced by training a deep neural network with the unaffected voice of the patient. The confidence of the predicted output is improved by introducing voice-to-text conversion block along with the deep neural network. The performance metrics reveals the effectiveness of the proposed algorithm to reproduce natural sound. The similarity index is also high compared to that of other state-of-the-art techniques.

Textile industries are one among important industries that contribute to the GDP of a nation. Century after century there has been advancement in this particular industry. The revenue generated by this sector is decided by the quality of the fabric items. Any fabric that is defect less has a good reception in the market and defective material produces only half of the production cost. Manual defect identification is a tiresome procedure where a individual is assigned to identify the defect in the running fabric. The success rate of such system is only 60%. An automated defect detection technique is the best solution for most of the textile industry as it produces around 96% efficiency in identifying the defect. The proposed algorithm is a new approach to embed with supervised and unsupervised learning classification techniques like support vector machine and K-Means clustering. The technique uses the GLCM texture features for the classification. The output depends on both of the classification results. The accuracy obtained in this embedded technique is 95%.

Color-to-grayscale conversion methods try to identify weights for various color channels for obtaining a grayscale image. These weights can be either fixed globally or computed on a localized basis. This chapter presents an approach for computing the global weights using localized regions chosen using the assistance of Human Vision System. For a given image, the proposed method aims to maximize the required foreground information, which is normally present in the dominant color channel. The proposed method was tested on DIBCO 2013 dataset and qualitatively evaluated using PSNR, MSE, and SSIM. The results obtained have established to be more satisfactory. The experimental results of ours and other color-to-grayscale methods have been tabulated and discussed.

Emotion Recognition (ER) systems is very much important for interpersonal relationship. Emotions are developed by some physiological changes. The straightforward of this effort is to discover the competence of language and facemask elements to deliver the feeling exact information for enhancing the Human-Machine interaction. The techniques and systems used in emotion detection may vary depending on the features inspected. Since both these features complement each other, combining them results in higher performance in terms of accuracy of 94.734%. The proposed system was tested on ENTERFACE’05 database and real time video. For Video, Speeded Up Robust Features (SURF) and Gabor features are used.

In this paper we propose a new methodology for contrast enhancement (CE) and edge preservation of an image using Quaternionic Wavelet Transform (QWT) and Singular Value Decomposition (SVD). Although there are many techniques like Global histogram equalization (GHE), Local histogram equalization (LHE) and other singular value based enhancement techniques, these methods are not robust as it fails to enhance very poor contrast images. By using Quaternionic Wavelet Transform the quality of the image is better enhanced compared to other wavelet transforms. QWT provides very good shift invariance and directionality than DWT. In addition to that the proposed technique maintains the brightness of the image and preserves the edges of the image with relatively fewer artifacts.