Optical Memory and Neural Networks最新文献

Microstrip Patch Antennas (MPA) are being used more and more in current communication systems because of their advantages such as being Lightweight, easy to construct, and low cost. However, MPA operational bandwidth and power handling capabilities are restricted. In this research, a novel unit cell MPA is designed and optimized using a Radial Basis Function Neural network (RBFNN). Flame-retardant (FR4) metamaterial is used in the fabrication of the envisioned antenna and the device. The High-Frequency Structure Simulator (HFSS) version 15 software is used for the design and simulation of the model. The design is simulated at a frequency range of 2 to 6 Hertz. Finally, the implementation of the antenna is performed using Complementary Split Ring Resonator (CSRR) technique. The proposed structure produces an excellent reflection coefficient, and Voltage Standing Wave Ratio (VSWR), which are –15.12 at 1.5 GHz, –55.41 at 2.5 GHz, and –25.63 dB at 3.5 GHz and 2.0 dB respectively. Simulation results show an excellent outcome, as return losses are 23.18, 38.67, and 44.12 dB at 0.6, 1.7, and 3.5 GHz respectively, and the gain is 8.5 dB at 6 GHz, which are quite similar to the actual values. The proposed unit cell antenna outperformed the other previously designed microstrip antenna and is suitable for wireless communication systems.

The research concerns the use of machine learning algorithms for multiscale coding of digital video sequences. Based on machine learning, the digital image coder is generalized to the coding of video sequences. To this end, we offer an algorithm that allows for videoframes interdependency by using linear regression. The generalized image coder uses multiscale representation of videoframes, neural network three-dimensional interpolation of multiscale videoframe interpretation levels and generative-adversarial neural net replacement of homogeneous portions of a videoframe by synthetic video data. The method of coding the entire video and method of coding videoframes are exemplified by block diagrams. Formalized description of how videoframe correlation is taken into account is given. Real video sequences are used to carry out numerical experiments. The experimental data allow us to make a conclusion about the promise of using the algorithm in video coding and processing.

Attention-based neural machine translation (attentional NMT), which jointly aligns and translates, has got much popularity in recent years. Besides, a language model needs an accurate and larger bilingual dataset_ from the source to the target, to boost translation performance. There are many such datasets publicly available for well-developed languages for model training. However, currently, there is no such dataset available for the English-Afaan Oromo pair to build NMT language models. To alleviate this problem, we manually prepared a 25K English-Afaan Oromo new dataset for our model. Language experts evaluate the prepared corpus for translation accuracy. We also used the publicly available English-French, and English-German datasets to see the translation performances among the three pairs. Further, we propose a deep attentional NMT model to train our models. Experimental results over the three language pairs demonstrate that the proposed system and our new dataset yield a significant gain. The result from the English-Afaan Oromo model achieved 1.19 BLEU points over the previous English-Afaan Oromo Machine Translation (MT) models. The result also indicated that the model could perform as closely as the other developed language pairs if supplied with a larger dataset. Our 25K new dataset also set a baseline for future researchers who have curiosity about English-Afaan Oromo machine translation.

In these modern days, cancers like Skin cancers is the general type of cancer that alters the life style of millions of citizens in each time. Around three million people are identified with it in each and every year in the US alone. Skin cancer related to the irregular enlargement of cells. On account of malignancy characteristic skin type cancer is termed as melanoma. Melanoma seems on skins because of the contact to ultraviolet emission and hereditary reasons. Thus melanoma seems like brown and black colour, but also occurs anyplace of the patient. Mostly the skin type cancers could be treatable at the earliest phases of beginning. So a fast recognition of skin cancer could rescue the life of patient. However, identifying skin cancer in its starting phases may be difficult and moreover it is expensive. Thus in the paper, they try to cope with such types of problems by making a wise decision scheme for skin lesion identification like the starting phase, that should be set into a smart robot for physical condition monitoring in our present surroundings to support early on detection. The scheme is enhanced to classify benign and malignant skin lesions with different procedures, comprising of pre-processing for instance noise elimination, segmentation, and feature extraction from lesion sections, feature collection and labelling. Following the separation of lots of raw images, colour and texture characteristics from the lesion regions, is employed to categorize the largely prejudiced noteworthy subsets for fit and cancerous circumstances. In it SVM has been applied to carry out benign and malignant lesion detection.

Winters-springs 2019–2020 and 2010–2011 became the periods of the severest ozone events in the Arctic throughout the satellite era. They stemmed from extremely cold and persistent polar stratospheric cloud (PSC) seasons, conducive to record strong chemical ozone destruction. TEMIS observations indicate that the total ozone (TO) column diverged from long-term norm by 45 to 55% in 2020 and by 37 to 44% in 2011 at Arctic sites; and by 27 to 32% in 2020 and by 27 to 36% in 2011 at midlatitudes. Aura MLS profiles showed that the minimum temperature was 8–13% lower than norm over the Arctic in 2020 and 8–12% lower in 2011. The ozone mixing ratios were 4% of the long-term mean at height of 20 km on March 27, 2020 and 25% at height of 21 km on March 20, 2011 for Eureka; and 7% at 19 km on April 19, 2020 and 24% at 20 km on March 20, 2011 for Ny-Ålesund. The divergences in water vapor and ozone mixing ratios, water vapor mixing ratio and temperature, and ozone mixing ratio and temperature show stronger correlations in 2020 than 2011. The correlations weaken equatorward, until becoming almost insignificant at extra-vortex latitudes.

We developed an algorithm of software product for processing the data from lidar sensing at the wavelengths of 299/341 nm for a vertical path of atmospheric sensing with the spatial resolution from 1.5 to 150 m. The main options of the software include: recording the atmospheric lidar sensing data, conversion of DAT to TXT file format, and retrieval of ozone concentration profiles. The software complex, developed on the basis of our algorithm to process the lidar sensing data, makes it possible to obtain the ozone concentration profiles from 4 to 20 km. The blocks of recording the data from atmospheric lidar sensing and retrieving the ozone concentration profiles allow for a visual control of the recorded lidar returns and retrieved ozone concentration profiles. We present an example of retrieving the ozone concentration profile from lidar data, which was obtained in 2022.

Semantic segmentation is crucial for autonomous driving as the pixel-wise classification of the surrounding scene images is the main input in the scene understanding stage. With the development of deep learning technology and the impressive hardware capabilities, semantic segmentation has seen an important improvement towards higher segmentation accuracy. However, an efficient sematic segmentation model is needed for real-time applications such as autonomous driving. In this paper, we discover the potential of employing the design principles of two deep learning models, namely PSPNet and EfficientNet to produce a high accurate and efficient convolutional autoencoder model for semantic segmentation. Also, we benefit from data augmentation for better model training. Our experiment on CamVid dataset produces optimistic results and the comparison with other mainstream semantic segmentation models justifies the used approach.

One of the current challenges in deep learning is semantic segmentation. Moreover, human part segmentation is a sub-task in image segmentation, which differs from traditional segmentation to understand the human body’s intrinsic connections. Convolutional Neural Network (CNN) has always been a standard feature extraction network in human part segmentation. Recently, the proposed Swin Transformer surpasses CNN for many image applications. However, few articles have explored the performance of Swin Transformer in human part segmentation compared to CNN. In this paper, we make a comparison experiment on this issue, and the experimental results prove that even in the area of human part segmentation and without any additional trick, the Swin Transformer has good results compared with CNN. At the same time, this paper also combines the Edge Perceiving Module (EPM) currently commonly used in CNN with Swin Transformer to prove that Swin Transformer can see the intrinsic connection of segmented parts. This research demonstrates the feasibility of applying Swin Transformer to the part segmentation of images, which is conducive to advancing image segmentation technology in the future.

One of the most frequent causes of death for women worldwide is breast cancer. In most cases, breast cancer can be quickly identified if certain symptoms emerge. But many women with breast cancer don’t show any symptoms. So, it is very critical to detect this disease in early stage also numerous radiologists are needed to diagnose this disease which is quite expensive for the majority of cancer hospitals. To address these concerns, the proposed methodology creates a Faster-Regional Convolutional Neural Network (Faster-RCNN) for recognizing breast cancer. Ultrasound images are collected and pre-processed utilizing resizing, adaptive median filter, histogram global contrast enhancement and high boost filtering. Image resizing is utilized to change the image size without cutting anything out. Adaptive median filter is utilized to remove unwanted noise present in the resized image. Histogram global contrast enhancement is used to enhancing the contrast level of the image. High boost filtering is utilized to sharpening the edges present in the image. After that, pre-processed images are fetched as an input to Faster R-CNN, which extract the features and segment the accurate region of the tumour. These segmented regions are classified using Multilayer Perceptron’s for detecting whether the patients are affected by breast cancer or not. According to the experimental study, the proposed approach achieves 97.1% correctness, 0.03% error, 91% precision and 93% specificity. Therefore, the developed approach attains better performance compared to other existing approaches. This prediction model helps to detect breast cancer at early stage and improve patient’s living standard.

We present a review of Neural Machine Translation (NMT), which has got much popularity in recent decades. Machine translation eased the way we do massive language translation in the new digital era. Otherwise, language translation would have been manually done by human experts. However, manual translation is very costly, time-consuming, and prominently inefficient. So far, three main Machine Translation (MT) techniques have been developed over the past few decades. Viz rule-based, statistical, and neural machine translations. We have presented the merits and demerits of each of these methods and discussed a more detailed review of articles under each category. In the present survey, we conducted an in-depth review of existing approaches, basic architecture, and models for MT systems. Our effort is to shed light on the existing MT systems and assist potential researchers, in revealing related works in the literature. In the process, critical research gaps have been identified. This review intrinsically helps researchers who are interested in the study of MT.