Portable Network Research Articles

Tensor neural networks via circulant convolution

Enough computation and storage resources have made Convolutional Neural Networks (CNNs) prosperous in many fields, but it is still challenging to deploy these models on mobile or embedded devices. Designing effective and efficient convolution modules can vastly reduce the number of parameters and computational complexity of CNNs to favor the deployment. This paper proposes a Circulant Convolution (CC) established on a tensor algebra operator of t-product. Specifically, a feature map can be parameterized by multilinear operation to make the channels positively coupled, and then a more compact parameter space and stronger parameterization capability can be simultaneously obtained. Replacing standard spatial convolution with CC can effectively reduce almost 9 times the number of network parameters and FLOPs, and the impact on accuracy is much smaller than other well-designed convolution modules for lightweight networks. CC can be embedded into existing network architectures as a plug-and-play module, and its topology structure can be easily extended to high-dimensional data. To favor the use of CC in CNNs, a circulant convolution module (CCM), also known as the bottleneck of CC, is also designed by combining CC and pointwise convolution. In further, a lightweight network CCMNet is constructed based on incorporating CC and CCM into an existing lightweight backbone. Extensive experiments on the benchmark dataset CIFAR10/100, ImageNet, Lung-CXR, and MS COCO, demonstrate that our proposed CC could significantly improve the performance of several advanced lightweight networks, and CCMNet is competitive with the start-of-the-art portable neural networks.

Conditional Generative Data-Free Knowledge Distillation

Knowledge distillation has made remarkable achievements in model compression. However, most existing methods demand original training data, while real data in practice are often unavailable due to privacy, security and transmission limitation. To address this problem, we propose a conditional generative data-free knowledge distillation (CGDD) framework to train efficient portable network without any real data. In this framework, except using the knowledge extracted from teacher model, we introduce preset labels as additional auxiliary information to train the generator. Then, the trained generator can produce meaningful training samples of specified category as required. In order to promote distillation process, except using conventional distillation loss, we treat preset label as ground truth label so that student network is directly supervised by the category of synthetic training sample. Moreover, we force student network to mimic the attention maps of teacher model and further improve its performance. To verify the superiority of our method, we design a new evaluation metric is called as relative accuracy to directly compare the effectiveness of different distillation methods. Trained portable network learned with proposed data-free distillation method obtains 99.63%, 99.07% and 99.84% relative accuracy on CIFAR10, CIFAR100 and Caltech101, respectively. The experimental results demonstrate the superiority of proposed method.

Detection and Localization of Buried Pipelines Using a 3-D Multistatic Imaging Radar

Subsurface pipeline detection and localization is an important problem in gas and oil transport as well as in construction and excavation missions. A 3-D subsurface multistatic imaging radar with a novel focusing algorithm for pipelines is demonstrated in this article. The traditional back-projection (BP) algorithm, which is suitable for discrete scatterers, cannot be used for extended scatterers such as pipelines whose scattering phase centers are dependent on the relative positions of the transmitter and receiver. To circumvent this problem, a new imaging algorithm that tracks the scattering phase centers in a way that is specific to pipelines is presented and shown to provide significantly better imaging performance. An analytical solution for pipeline scattering is used to derive the algorithm, and justifications for the simplifying assumptions made are provided. The algorithm is tested by applying it to realistic lossy sand simulations data as well as experimental measurements data obtained by a portable vector network analyzer (VNA). Both simulation and measurement results demonstrate the ability of the algorithm to detect and localize metallic as well as dielectric pipes.

Development of a Simulated Portable Mesh Network via ESP8266-Based Devices with Utility Application

Mesh network is very effective for node-to-node communications. In networks that require and demand devices for home monitoring and control, uncommon topology like Mesh is viewed to have a more effective advantage. However, when it comes to networks, Mesh Network is almost left untouched in most studies simply due to the existing network solutions such as WLAN or Bluetooth. The existence of low-powered WPAN devices fill the demand in integrating devices into mesh networks. However, constant configuration could be very complex and difficult for a mesh using existing WPAN devices since nodes constantly move in and out dynamically in the mesh network. This study focused on the development of a simulated mesh network through ESP8266 enabled devices to deliver a straightforward setup and portable connectivity. The study also developed a utility application that integrates an algorithm for node network operations, and a facility that keeps track of mesh network information including delays, node connections, and data transmission. Three(3) portable ESP8266 devices were used and configured in the study. The devices were integrated in a simulated mesh network and subjected to network processes including identity tagging, dynamic connection, routing, One-way Delay and Payload Size Tests. Results of One-Way Delay and Payload Size Tests indicate consistency of transmission and receiving of data of nodes connecting and disconnecting to the simulated network. This can be used as basis in extending further into more high-end nodes like handheld devices and even computers. Similarly, the utilization of dedicated algorithm for Mesh in this study proved that portability can be achieved to avoid loss of connection when nodes abruptly fail in the network.

Detection and Localization of Pipeline Leaks Using 3-D Bistatic Subsurface Imaging Radars

Oil pipeline leak detection and localization is an important problem with regard to environmental issues as well as loss of resources in oil transport systems. This article shows that a 3-D subsurface multistatic imaging radar can better detect water or oil leak when more scattered signals from the leaked region are collected from multiple directions and added coherently. However, a primal drawback in achieving detection is the dominance of the signal from the pipeline itself compared to the leak signal making direct radar-based leak detection quite challenging. By utilizing the differences in the scattering mechanisms of the pipeline and the leak, this article proposes a technique that identifies and eliminates the pipeline signal from the overall radar response, which improves leak detectability. Permittivity of oil–sand mixture is determined experimentally and fit to a semiempirical mixing formula. A realistic physics-based model is used to determine the 3-D volumetric shape of oil leak. Then, the pipe with leak is simulated in a full-wave simulator with the permittivity of the leak assigned using the mixing formula. Comprehensive simulations are carried out for oil leaks in various soil mixtures and pipe materials to test the effectiveness of the proposed approach. In addition, the proposed approach is investigated experimentally using a portable vector network analyzer (VNA) where the response of a small water leak out of metallic as well as PVC pipes is measured in bistatic settings. Both simulation and measurement results demonstrate the effectiveness of the proposed approach in detecting and localizing pipeline leaks.

Portable Microwave Reflectometry System for Skin Sensing

In spite of the technological advancement of the healthcare system, monitoring skin hydration still remains a challenging task. In particular, the state-of-the-art solutions are inadequate to obtain continuous real-time monitoring, especially in a wearable perspective. Starting from these considerations, this article addresses the implementation of an innovative wearable device that can monitor skin hydration through microwave reflectometry technique. To this purpose, a new flexible wearable sensing element (SE) was designed, manufactured, and validated to sense variation of skin hydration of the body, ensuring simultaneously a better wearability, and surface adherence to the skin. The system was preliminarily validated through a numerical analysis and then experimental tests were carried out through a low-cost portable vector network analyzer (VNA), which is particularly convenient in view of fully-wearable applications and comparable, from the metrological point of view to more expensive VNAs operating in the frequency range of interest. Results showed that the proposed SE configuration and the low-cost VNA hold potential for achieving a fully-wearable system for monitoring skin hydration.

Software defined radio for vector network analysis: Configuration, characterization and calibration

A portable vector network analyzer operating from 30 MHz to 850 MHz, using a commercial low-cost software-defined radio is described. The system can be used for transmission (S21) or, using a directional bridge, reflection measurements (S11). A complete software package is developed in order to define and operate the system as an automatic analyzer including the correction of systematic errors. The analyzer has heterodyne receivers with an specific intermediate frequency, digitizer sampling rate and block size chosen to correctly process the signals. Measurements to characterize the transmitters and receivers on the board are shown. System errors are determined by comparison with laboratory equipment measurements and manufacturer data.

A portable low-cost reflectometric setup for moisture measurement in cultural heritage masonry unit

A low-cost and portable reflectometric system was described to correlate the water content inside building material with the resonance frequency of a planar probe. An inexpensive planar patch resonator was realized, and a high sensitivity load cell was developed to investigate the external force effect on the resonance frequency shift. Regarding the material under test, four stones commonly employed in Italy Cultural Heritage buildings were used: Pietra Gentile, Leccese, Carparo, and Tuff. Measurements were carried out at different water content values and were performed using a portable vector network analyzer. Two different trials were carried out ten times for each moisture level: with an unknown handheld force exerted on the probe and using the load cell to ensure a repeatable force. Results showed that a repeatable applied force guaranteed more reproducible measurements. The achieved humidity/frequency calibration curves can be used, in practice, for non-invasive on-site water content monitoring of historic structures.

A Novel Method for the Accurate Measurement of Soil Infiltration Line by Portable Vector Network Analyzer.

Accurate measurement of soil infiltration lines is very important for agricultural irrigation systems. It can help monitor the irrigation of soil to control irrigation amounts and promote crop growth. The soil infiltration line is a complex dynamic boundary and is difficult to model accurately, leading to estimation deviation. A traditional TDR (time domain reflectometry) method is used in soil infiltration line measurement, but it lacks good applicability and accuracy. In this paper, we proposed a method—VFTT (The vector network analyzer’s frequency domain signals are converted to the time domain)—by the time domain to frequency domain conversion principle to improve the accuracy of soil infiltration line measurement. The experiment results show that the measurement method of soil infiltration line based on VFTT has high accuracy and robustness. After fitting the measured value with the actual one, R2 reaching more than 0.98 can effectively measure the position of the soil infiltration line.

Dual discriminator adversarial distillation for data-free model compression

Knowledge distillation has been widely used to produce portable and efficient neural networks which can be well applied on edge devices for computer vision tasks. However, almost all top-performing knowledge distillation methods need to access the original training data, which usually has a huge size and is often unavailable. To tackle this problem, we propose a novel data-free approach in this paper, named Dual Discriminator Adversarial Distillation (DDAD) to distill a neural network without the need of any training data or meta-data. To be specific, we use a generator to create samples through dual discriminator adversarial distillation, which mimics the original training data. The generator not only uses the pre-trained teacher’s intrinsic statistics in existing batch normalization layers but also obtains the maximum discrepancy from the student model. Then the generated samples are used to train the compact student network under the supervision of the teacher. The proposed method obtains an efficient student network which closely approximates its teacher network, without using the original training data. Extensive experiments are conducted to demonstrate the effectiveness of the proposed approach on CIFAR, Caltech101 and ImageNet datasets for classification tasks. Moreover, we extend our method to semantic segmentation tasks on several public datasets such as CamVid, NYUv2, Cityscapes and VOC 2012. To the best of our knowledge, this is the first work on generative model based data-free knowledge distillation on large-scale datasets such as ImageNet, Cityscapes and VOC 2012. Experiments show that our method outperforms all baselines for data-free knowledge distillation.

Secure Data Sharing Platform for Portable Social Networks with Power Saving Operation

Several subscribing and content sharing services are largely personalized with the growing use of mobile social media technology. The end user privacy in terms of social relationships, interests and identities as well as shared content confidentiality are some of the privacy concerns in such services. The content is provided with fine-grained access control with the help of attribute-based encryption (ABE) in existing work. Decryption of privacy preserving content suffers high consumption of energy and data leakage to unauthorized people is faced when mobile social networks share privacy preserving data. In the mobile social networks, a secure proxy decryption model with enhanced publishing and subscribing scheme is presented in this paper as a solution to the aforementioned issues. The user credentials and data confidentiality are protected by access control techniques that work on privacy preserving in a self-contained manner. Keyword search based public-key encryption with ciphertext policy attribute-based encryption is used in this model. At the end users, ciphertext decryption is performed to reduce the energy consumption by the secure proxy decryption scheme. The effectiveness and efficiency of the privacy preservation model is observed from the experimental results.

Deep learning with convolutional neural network for estimation of the characterisation of coronary plaques: Validation using IB-IVUS

IntroductionDeep learning approaches have shown high diagnostic performance in image classifications, such as differentiation of malignant tumors and calcified coronary plaque. However, it is unknown whether deep learning is useful for characterizing coronary plaques without the presence of calcification using coronary computed tomography angiography (CCTA). The purpose of this study was to compare the diagnostic performance of deep learning with a convolutional neural network (CNN) with that of radiologists in the estimation of coronary plaques. MethodsWe retrospectively enrolled 178 patients (191 coronary plaques) who had undergone CCTA and integrated backscatter intravascular ultrasonography (IB-IVUS) studies. IB-IVUS diagnosed 81 fibrous and 110 fatty or fibro-fatty plaques. We manually captured vascular short-axis images of the coronary plaques as Portable Network Graphics (PNG) images (150 × 150 pixels). The display window level and width were 100 and 700 Hounsfield units (HU), respectively. The deep-learning system (CNN; GoogleNet Inception v3) was trained on 153 plaques; its performance was tested on 38 plaques. The area under the curve (AUC) obtained by receiver operating characteristic analysis of the deep learning system and by two board-certified radiologists was compared. ResultsWith the CNN, the AUC and the 95% confidence interval were 0.83 and 0.69–0.96, respectively; for radiologist 1 they were 0.61 and 0.42–0.80; for radiologist 2 they were 0.68 and 0.51–0.86, respectively. The AUC for CNN was significantly higher than for radiologists 1 (p = 0.04); for radiologist 2 it was not significantly different (p = 0.22). ConclusionDL-CNN performed comparably to radiologists for discrimination between fatty and fibro-fatty plaque on CCTA images. Implications for practiceThe diagnostic performance of the CNN and of two radiologists in the assessment of 191 ROIs on CT images of coronary plaques whose type corresponded with their IB-IVUS characterization was comparable.

PAG-YOLO: A Portable Attention-Guided YOLO Network for Small Ship Detection

The YOLO network has been extensively employed in the field of ship detection in optical images. However, the YOLO model rarely considers the global and local relationships in the input image, which limits the final target prediction performance to a certain extent, especially for small ship targets. To address this problem, we propose a novel small ship detection method, which improves the detection accuracy compared with the YOLO-based network architecture and does not increase the amount of computation significantly. Specifically, attention mechanisms in spatial and channel dimensions are proposed to adaptively assign the importance of features in different scales. Moreover, in order to improve the training efficiency and detection accuracy, a new loss function is employed to constrain the detection step, which enables the detector to learn the shape of the ship target more efficiently. The experimental results on a public and high-quality ship dataset indicate that our method realizes state-of-the-art performance in comparison with several widely used advanced approaches.

Green Communication in Wireless Power Consumption and Energy Efficient Trade-offs

Energy efficiency is a significant issue in portable wireless networks since the battery life of versatile terminals is restricted. Protection of battery power has been tended to utilizing numerous procedures. Wireless sensor networks (WSNs), framed by various little gadgets fit for detecting, processing, and wireless correspondence are arising as a progressive innovation, with applications in different territories. The novel highlights of wireless sensor networks have carried new difficulties and issues to the field of conveyed and communitarian data preparing. In the light of the importance of reducing operating consumpt and maintaining cellular network profitability, energy efficiency in cell networks has received a crucial consideration from both scholars and the business, despite the fact that these networks are “green communication.” Since the base station is the most important energy buyer in the business, efforts have been undertaken to review the use of the base station and to identify ways to energy efficiency improvements. The trade-offs between energy utilization and throughput, under nearby just as under helpful detecting, are portrayed. The Energy efficient tradeoffs have been arranged dependent on every convention layer and examined its effect in the organization energy efficiency.

Image compression methods for efficient storage and transmission

This paper presents a detailed and comprehensive review of image compression methods, emphasizing their role in optimizing both storage and transmission efficiency across various domains, from everyday use in social media to specialized applications like medical imaging and satellite data processing. We systematically explore both traditional and contemporary image compression techniques, categorizing them into lossless and lossy methods, transform-based approaches, and the latest advancements in machine learning-based compression. Lossless compression techniques, including Run-Length Encoding (RLE), Huffman Coding, Lempel-Ziv-Welch (LZW), and the Portable Network Graphics (PNG) format, are discussed for their ability to preserve image quality perfectly, albeit at the cost of relatively lower compression ratios. Conversely, lossy compression methods, such as JPEG and fractal compression, offer significant file size reduction by discarding non-essential data, while still maintaining acceptable visual quality for many practical applications. We further delve into transform-based approaches like Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT), which form the backbone of popular standards such as JPEG and JPEG 2000, enabling more efficient data representation in the frequency domain. Additionally, the study highlights emerging machine learning and deep learning techniques, such as autoencoders and Generative Adversarial Networks (GANs), that are pushing the boundaries of image compression by achieving unprecedented compression ratios while minimizing perceptual loss in image quality. Through a comparative analysis, we evaluate these methods based on multiple performance metrics, including compression ratio, computational complexity, image fidelity (measured via Peak Signal-to-Noise Ratio, PSNR, and Structural Similarity Index, SSIM), and their practical applications across different industries. Our findings suggest that while traditional methods such as JPEG, PNG, and JPEG 2000 remain widely adopted due to their simplicity and efficiency, emerging techniques driven by deep learning show great potential in adapting to specific image characteristics, achieving higher compression ratios, and better preserving image quality under extreme compression. Finally, this paper identifies key challenges and trends in the field, such as the increasing computational demands of advanced techniques, the need for adaptive compression strategies, and the importance of standardization for broad industry adoption. We conclude that while traditional methods will continue to play a significant role, the future of image compression lies in the integration of machine learning and content-aware technologies that dynamically optimize compression performance across diverse image types and application contexts.

JPG, PNG and BMP image compression using discrete cosine transform

This paper proposes image compression using discrete cosine transform (DCT) for the format of joint photographic expert groups (JPEG) or JPG, portable network graphic (PNG) and bitmap (BMP). These three extensions are the most popular types used in current image processing storage. The purpose of image compression is to produce lower memory usage or to reduce memory file. This process removes redundant information of each pixel. The challenge for image compression process is to maintain the quality of images after the compression process. Hence, this article utilizes the DCT technique to sustain the image quality and at the same time reduces the image storage size. The effectiveness of the DCT technique has been reasonable over some real images and the implementation of the technique has been compared with different types of image extensions. Matlab software is an important platform for this project in order to write a program and perform the progress of project phase by phase to achieve the expected results. Based on the tested images, the DCT technique in image compression is capable to reduce the image storage memory in average about 50% of each image tested.

Piezoelectric Energy Harvesting Methodologies using Ambient Mechanical Vibration: Design Perspective and Challenges

Background and Objective: Piezoelectric energy harvesting is an emergent research area for unobtrusive power supply for fully autonomous micro-scale devices. Method: The method of energy harvesting is to utilize waste ambient mechanical vibrations to generate electrical energy through the piezoelectric effect. Results and Conclusion: The present work highlights the major advances made in the field of micro- electromechanical systems based on piezoelectric energy harvester on extracting ambient vibrations and converting them into usable electric power. The present study explores energy harvesting approaches for portable electronics and self-powered wireless network nodes. The performance matrices like device physics, volume, operating frequencies, design and materials have been thoroughly analyzed in this work. Conventional cantilever fabrication steps have also been discussed. Finally, guidelines for future research and performance enhancements in the field of Piezoelectric Energy Harvesting (PEH) at micro scale have been discussed.

Resourceful Investigation of Malnutrition with Machine Learning Using CNN Procedure

This System used to refer to any condition in which the body does not obtain enough nutrients for proper function in this system Image processing is a technique to convert an image into digital form and perform some action on it, in order to get an enhanced image or to citation some useful information from it. There are some types of image that is Tagged Image Format, Joint Photographic Group, Graphic Interchange Format, Portable Network Graphics, Row Image File. Computer can easily classify more than 16 million color where the eye capability has limitation while identifying color and also same people face the problem such as colors carelessness. So the execution nail color Analysis through computer is a higher procedure as compared with human eyes. Human eyes have limited resolution, finding deviation in nearby pixel intensity is not possible for human eyes, but computer vision can detect every pixel appropriately. So, in this system we used the Convolutional Neural Networks (CNN) algorithm & carried out the result. this system is very helpful for the patient, to find out the prediction or detection of diseases without doctor.

Toward Embedded Sensing Automation and Miniaturization for Portable Smart Cost-Effective Algae Monitor

As an important indicator of water pollution, algae are highly sensitive to changes in their environment and respond to a wide range of pollutants, they provide an early caution signal of worsening ecological condition. In this article, a kind of portable microfluidic lensless depth neural network algae monitor is proposed. The lensless algae image acquisition module, algae segmentation and classification circuit, and touch panel were integrated into the equipment. Therefore, the equipment can collect and analyze algae automatically in wild water body without laboratory. In order to miniaturize the equipment and reduce the cost, lensless microfluidic channel sampling is adopted. In addition, a dual asymmetric quantization algorithm and circuit structure are proposed for the implementation of deep neural network hardware. Finally, the prototype system construction was completed. Compared with the current analysis equipment, the equipment size and hardware cost are greatly reduced, and the accuracy reduction is kept in a small range, which makes a better compromise between the accuracy, hardware cost and circuit power consumption. The performance of the equipment fully meets the needs of the current portable algae monitor equipment, and the cost is greatly reduced compared with the current equipment. The accuracy of the equipment achieves 94.27%, the size achieves 11 * 11 * 17.5cm. These advances in portability and cost are conducive to promoting the transformation of water algae analysis based on artificial intelligence from large servers in the laboratory to portable algae analysis equipment, and promoting the rapid early analysis of water monitoring.

Diversifying Inference Path Selection: Moving-Mobile-Network for Landmark Recognition.

Deep convolutional neural networks have largely benefited computer vision tasks. However, the high computational complexity limits their real-world applications. To this end, many methods have been proposed for efficient network learning, and applications in portable mobile devices. In this paper, we propose a novel Moving-Mobile-Network, named M2Net, for landmark recognition, equipped each landmark image with located geographic information. We intuitively find that M2Net can essentially promote the diversity of the inference path (selected blocks subset) selection, so as to enhance the recognition accuracy. The above intuition is achieved by our proposed reward function with the input of geo-location and landmarks. We also find that the performance of other portable networks can be improved via our architecture. We construct two landmark image datasets, with each landmark associated with geographic information, over which we conduct extensive experiments to demonstrate that M2Net achieves improved recognition accuracy with comparable complexity.