Federated learning (FL) offers a privacy-preserving paradigm for distributed machine learning, but its application to intrusion detection systems (IDS) in IoT networks is hindered by severe class imbalance, highly non-IID data, and high communication overhead. These challenges severely degrade the performance of conventional FL methods in real-world network traffic classification. To overcome these limitations, we propose Sentinel, a personalized federated IDS (pFed-IDS) framework that incorporates a dual-model architecture on each client, consisting of a high-capacity personalized teacher and a lightweight globally shared student model. This design balances deep local adaptation with efficient global aggregation while preserving privacy and reducing communication overhead by transmitting only the compact student model. Sentinel integrates three key mechanisms to ensure robust performance: bidirectional knowledge distillation with adaptive temperature scheduling, lightweight multi-level feature alignment between teacher and student representations, and a class-balanced loss to handle highly skewed traffic. On the server side, normalized gradient aggregation with equal client weighting mitigates client drift and improves fairness across clients. Extensive experiments on the IoTID20 and 5GNIDD benchmark datasets demonstrate that Sentinel significantly outperforms state-of-the-art federated baselines under extreme data heterogeneity, while lowering communication overhead.

Steering angle prediction plays an important role in autonomous vehicle control. Existing approaches rely on single modality like camera or LiDAR data to predict the steering angle, ignoring the rich information obtained after fusing all sensor modalities. In this work we proposed the multimodal fusion framework to improve the steering angle prediction accuracy. Specifically, we aligned LiDAR features to the FLAVA vision language space and fused these aligned features with image features to predict the steering angle of the vehicle. The proposed method demonstrates the considerable accuracy in steering angle prediction outperforming single modality approaches and highlighting the importance of multimodal fusion to predict the steering angle of the vehicle. Our code is available at: https://github.com/ParvezAlam123/Aligning-LiDAR-to-Vision-Language-Space-A-fusion-approach-for-Steering-Angle-Regression-of-a-Vehicle

Simultaneous Localization and Mapping (SLAM) is fundamental to autonomous navigation, relying on multimodal dataset to generate high-definition (HD) maps for accurate localization and path planning. However, most existing SLAM datasets primarily feature structured environments with abundant landmarks, such as urban areas with buildings and well-defined road infrastructures. In contrast, highways in semi-urban areas pose significant challenges due to the limited availability of prominent features. To bridge this gap, we introduce TIAND-SLAM (TiHAN-IIT Hyderabad Autonomous Navigation Dataset), a novel multi-modal dataset collected from highway roads, both under and above a flyover in Hyderabad, India, as well as the campus data within the IIT Hyderabad (IITH) campus and TiHAN testbed. TIAND-SLAM is designed to facilitate research on SLAM generalization in feature-scarce environments too. The dataset includes 30 trajectories ranging from 50 meters to 2.5 kilometers, recorded using a sensor suite comprising a LiDAR, six cameras, radar, RTK-GNSS, and an Inertial Measurement Unit (IMU) sensor. Ground truth localization is obtained using RTK-GNSS ensuring precise benchmarking. Additionally, we evaluate SLAM performance by generating maps using LiDAR data. TIAND-SLAM serves as a valuable resource for advancing SLAM research in highway and unstructured terrains, promoting robustness and adaptability in real-world autonomous navigation scenarios.

This paper introduces TiHAN-V2X, a comprehensive dataset capturing dynamic Cellular Vehicle-to-Everything (C-V2X) communication under real-world conditions in India. The dataset comprises four sub-datasets corresponding to Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), Infrastructure-to-Vehicle (I2V), and Vehicle-to-Cloud (V2C) communication modes. Each sub-dataset includes time-synchronized spatial (e.g., location, speed, heading) and Quality-of-Service (QoS) metrics (e.g., RSSI, SNR, latency, packet error rate, throughput). We detail scenario-wise measurements-covering congested urban, moderate suburban, highway, and cloud offloading conditions-and present a novel Data-Driven Markov Chain and Markov Decision Process Framework (MDP) for CV2X Communication Optimization. TiHAN-V2X provides highresolution, real-world data that reflects the unique challenges of the Indian environment.

This paper presents an innovative method in autonomous drone technology, addressing the critical challenges of achieving precise positional accuracy and safe landings in environments where GPS reliability is compromised. Traditional methods have struggled with accurate marker detection for precision landing due to GPS limitations and environmental variations and costly and used heavy high-resolution cameras. Our innovative method leverages vision based ArUco marker detection, by adopting deep learning techniques and utilizing low-resolution, cost-effective cameras, this method offers an efficient alternative for marker detection, overcoming the drawbacks associated with high-resolution imaging. A pivotal aspect of our research is the development of the TiHAN Dataset for Marker- Based Precision Landing in Drones (TDMBPLD) which contains the high- and low-resolution data, which has played a crucial role in achieving remarkable detection accuracy rates 99% in post-processing and 88% in real-time scenarios and precision landing 5 cm. This paper also proposes and tests a system for continuous remote monitoring and marker detection using edge cloud computing. In this system tests two unique datasets showed that proposed method significantly improves marker detection and reconstruction, enhancing real-time UAV landing capabilities and making drone technology more efficient and practical.

Paddy is a critical global food crop, and its productivity is directly dependent on the number of panicles. Accurate panicle counting is essential for effective crop management. Existing localization-based methods rely on intermediate representations like density maps, which are often counter-intuitive and prone to errors. To quantify and locate panicles directly rather than sequentially, we propose P2PNet-Paddy, a novel approach based on the Point-to-Point Network (P2PNet) architecture, which was initially developed for crowd counting and localization. While the original P2PNet uses the Hungarian algorithm for point matching, it is found to result in uncertain matching results without context, reducing the counting efficiency. It is found that P2PNet encounters significant performance limitations when applied to agricultural datasets characterized by highly imbalanced panicle density distributions. We suggest integrating a density-based sampler to mitigate the impact of panicle density imbalance in the dataset and the k-nearest neighbors Matching Objective (KMO) Hungarian matcher to enhance the matching process by incorporating additional contextual information. Our proposed approach led to notable improvements, reducing MAE by 36% and MAPE by 48.3% compared to the baseline P2PNet model. The proposed method achieved an MAE of 7.31, MAPE of 39.18, andR2score of 0.87 on the test data. The dataset and code will be made publicly available at https://github.com/tejasri19/P2PNet-Paddy.

Automation of crop yield estimation is crucial to cultivate efficient breeding techniques to fulfill the increasing population demands and adapt to climate change. The panicle count of paddy is directly associated with the yield of crop. In this work, we implement computer vision-based image segmentation methods for segmenting panicles by utilizing Unmanned Aerial Vehicle (UAV) captured multispectral imagery. The developed algorithms in this study could be used as a reference for panicle counting models during the vegetative and heading stages of the paddy crop. In addition, we introduce a semi-automatic image annotation method for effortless creation of labeled paddy crop datasets. With this method, our developed tool collaborates with an annotation tool to offer preliminary image annotations, which users can subsequently refine the annotations.

Object detection and subsequent perception of the environment surrounding a vehicle play a very important role in autonomous driving applications. Existing perception algorithms do not generalize well since most algorithms are trained in well-structured driving environment datasets. To deploy self-driving cars on the road, they should have a reliable and robust perception system to handle all corner cases. This paper introduces a multimodal dataset, TIAND*(TiHAN-IITH Autonomous Navigation Dataset), collected from structured and unstructured environments seen in and around the city of Hyderabad, India, as an aid to further research in the generalization of object detection algorithms. The sensor suite contains four cameras, six radars, one Lidar, and GPS and IMU. TIAND comprises 150 scenes, each spanning a duration ranging from 2 minutes to 4 minutes. Subsequently, we present the object detection model’s performance using camera, radar, and Lidar data. Additionally, we offer insights into projecting data from Lidar to camera and from radar to camera.

LiDAR odometry is an important problem for autonomous vehicles, robotics, drones, etc. This paper proposes a data-driven deep learning-based LiDAR odometry network LiDAR-OdomNet (LiDAR Odometry Network). The network has been trained on the KITTI odometry benchmark. It predicts translation parameters of the pose matrix with 0.0919 RMSE value, which is the minimum error obtained compared to current methods. An ablation study has been done using experiments to determine the significance of the proposed approach. We have analyzed every parameter of the pose matrix and plotted the results. To check LiDAR-OdomNet generalization, we have looked at some samples from the TIAND dataset and analyzed the results. Our code is available at: https://github.com/ParvezAlam123/LiDAR-OdomNet

Object detection serves as an important perception task for both autonomous vehicles and Advanced Driver Assistance Systems (ADAS). While object detection in camera images has been extensively studied, tackling this task with Light Detection and Ranging (LiDAR) data presents unique challenges due to its inherent sparsity. This study introduces a pioneering approach for LiDAR-based object detection, wherein LiDAR point cloud data is ingeniously transformed into a pseudo-RGB image format, subsequently fed into the YOLOv8 network originally designed for camera-based object detection. Trained and rigorously evaluated on the KITTI dataset, our method demonstrates outstanding performance, achieving an impressive mean Average Precision (mAP) of more than 86%. The model was also tested on some point clouds from TiHAN IITH Autonomous Navigation Dataset (TIAND). This remarkable result highlights the efficacy of the proposed approach in harnessing LiDAR data for robust object detection, thus contributing to the advancement of perception capabilities in autonomous driving and ADAS applications.

This paper introduces an innovative approach for cooperative vehicle platooning, integrating LiDAR-guided navigation, shared waypoints, and adaptive cruise control to optimize trajectory following and return maneuvers. Utilizing a pre-defined high-definition (HD) LiDAR map of the route, each vehicle in the platoon receives shared waypoints and utilizes onboard LiDAR sensors to localize itself relative to these waypoints. An adaptive cruise control system, guided by localized position and desired speed, regulates each vehicle's movement to ensure safe following distances and adherence to the planned trajectory. The results demonstrate the effectiveness of this approach in achieving precise trajectory following, maintaining optimal inter-vehicle spacing, and adapting safe velocity and heading according to the trajectory, thereby enabling coordinated platooning. This paper presents an autonomous platooning system for cooperative driving, showcasing the integration of advanced LiDAR-based technologies and control mechanisms.

LiDAR map-based localization and navigation play a crucial role in autonomous navigation, especially in GNSS-denied areas, by matching LiDAR data with a 3D point cloud map in real time. However, initiating localization from intermediate points on the map, distant from the origin, presents challenges. Typically, localization begins from the map’s origin, where algorithms can easily match current points to those near the origin. Challenges arise when starting from intermediate points, where the algorithm tries to match the current points’ features with those near the origin of the point cloud map. This mismatch can cause localization failure, leading to what is popularly known as the robot kidnapping problem. To tackle this challenge, we propose a solution involving the creation of unique nodes on the point cloud map by fusing LiDAR and GNSS data. Subsequently, live GNSS data is utilized to identify the nearest node, and the corresponding initial pose is published to initiate the localization problem at intermediate points on the map. Extensive real-time testing of this algorithm has been conducted at the IIT Hyderabad campus. The code for the same will be released at: https://github.com/Rakshith-Ram/Localize_Anywhere

Accurate and timely information about crop types and their distribution is crucial for effective agricultural management, resource allocation, and policy-making. Remote sensing (RS) imaging has the potential to automate the identification and mapping of crops over large areas. The low spatial resolutions of satellite-based platforms and the limited spectral capability of RGB/multispectral cameras are the bottlenecks in efficient crop categorization. Unmanned aerial vehicle (UAV)-based hyperspectral imaging (HSI) technology has the potential to classify/map agricultural landscapes efficiently due to its extensive coverages, rich spectral information, and high spatial and temporal resolutions. However, very few crop hyperspectral (HS) datasets are publicly available, on which the research community relies on for developing and testing algorithms, which are created from either in situ spectral measurements or low-resolution satellite images. This letter presents UAV-borne Crop HyperSpectral Image (UC-HSI) datasets in the 385–1021-nm spectral range. The detailed steps for creating clean HSI datasets from UAV-based HS images are described. We also proposed a novel convolutional transformer fusion architecture to classify the crop HSI datasets efficiently and compared its performance with machine learning (ML) benchmarks; it obtained the best accuracy of 95.26% in categorizing ten crop varieties. The datasets and codes will be made publicly available athttps://github.com/sankaraug/CrHyperSto benefit the RS community, allowing them to develop and test algorithms on UAV-based HSI data.

The proliferation of billions of heterogeneous Internet of Things (IoT) devices at a rapid pace has resulted in a marked expansion of attack surfaces. Numerous new attacks are constantly emerging to undermine the network’s availability, data confidentiality, and systems’ integrity due to inadequate security measures and resource limitations. Intrusion detection systems (IDSs) are used as the first line of defense to identify early instances of cyber-attacks targeting critical points. However, Next-Generation Networks (NGNs) with dense connectivity pose a challenge for traditional IDS approaches, as they raise concerns about users’ data privacy. Federated learning-based IDSs (Fed-IDSs) are an emerging and promising solution, as they permit the training of machine learning models on decentralized data stored on devices without compromising privacy. However, Fed-IDSs also have some unique issues. We identified that the existing Fed-IDSs have poor performance since the datasets used for evaluation, or the data in the real world, are highly imbalanced, and classes are not uniformly distributed. Motivated by this, we developed a novel IDS to effectively address the problem of class imbalance in federated learning at both the local and global levels. Following this, we evaluated the performance of our Fed-IDS under both independent and identically distributed (IID) and non-IID data settings and observed its generalizability to detect various attacks improved greatly. Extensive experiments are conducted to illustrate the effectiveness and benefits of this proposal.

Light Detection and Ranging (LiDAR) sensor play a vital role in the fields like environment perception for an autonomous vehicle, surveying and many other fields. LiDAR emits an enormous amount of data, which is difficult to transmit over a wireless medium. The need for LiDAR data transmission arises specifically in the Intelligent Transportation System (ITS) and surveying. It is advantageous to transmit LiDAR (mounted on a car) data over a wireless medium and receive the same by another vehicle or Road Side Units (RSUs) or control stations of a tesbed setup. Similarly, visualizing LiDAR data remotely while data acquisition by LiDAR mounted on Unmanned Aerial Vehicles (UAVs) could simplify the process of surveying. This paper presents a client-server-based LiDAR data streaming system using socket programming. In this system, the server transmits LiDAR point-cloud percept data, and the client captures and visualizes the streaming data over a WiFi medium. The server and the client are both equipped with the Robot Operating System(ROS). An extensive analysis of data transfer on two popular protocols, TCP and UDP, has also been presented. The data transmission and reception on ROS platform was successfully performed with a tolerable latency.

End to end control using steering angle prediction of autonomous vehicle is the research problem which has been solved using camera based sensor. In our work we have developed deep learning architecture for solving steering angle prediction using real time LiDAR data. Because LiDAR sensor gives 360 degree view of autonomous vehicle. Camera is not useful for various lighting condition specially in night time while LiDAR sensor is independent of various light scenarios. We have recorded LiDAR data and corresponding steering angle and proposed small benchmark dataset for research. We have performed our experiment on recorded dataset.

Autonomous vehicles and the Advance Drive Assistance System (ADAS) use a number of sensors for the perception of surroundings. One of the frequently used sensors is Light Detection and Ranging (LiDAR). LiDAR gives accurate depth information of the objects around an autonomous vehicle. One of the vital steps of LiDAR-based perception is real-time object detection. Object detection must be performed reasonably well for safety-critical applications with reasonably good accuracy and low latency. Another vital parameter to consider is the detection at a larger distance, the area where LiDAR data becomes sparser. LiDAR-based object detection becomes challenging at more considerable distances as the point cloud becomes sparser. To mitigate this problem, a novel method for point density normalization has been adopted to make the detection less affected by distance. In this paper, an efficient LiDAR point cloud object detection, based on the YoloV4 model, has been proposed, which gives high detection accuracy with the ability to process more than 27 frames per second. For the evaluation of the proposed method of detection, the Kitti dataset was used.

In this paper we propose a lane keep assist and warning system using automotive radars in different real time scenarios. Owing to radar’s immunity to weather and lighting conditions, it is possible to detect objects with high accuracy. We use Bayesian filtering technique to accurately detect and track vehicles moving towards the ego vehicle. A warning signal is generated if there is a vehicle in the proximity zone. In this work we explain the architecture of lane keep assist and warning (LKAW) systems using radar and also compare it with other lane keep assist systems which use camera as their primary sensors.

This paper presents the early identification of water stress in groundnut canopy using unmanned aerial vehicle (UAV) based hyperspectral imaging (in 385-1020 nm) and machine learning (ML) techniques. An efficient hyperspectral imaging (HSI) data analysis pipeline was presented which includes image quality assessment, denoising, band selection, and classification. A novel ML-based ensemble feature selection (FS) algorithm has been proposed for optimal water stress sensitive waveband selection. The data analysis pipeline and the selected bands were validated on HSI data acquired at two different water stress levels. Wavelengths 515.05, 552.16, 711.92, 724.75, and 931.92 nm were identified as optimal water stress sensitive bands in groundnut canopy, using which we could identify early stress with 96.46% accuracy. The proposed data analysis pipeline and ensemble FS algorithm will benefit crop phenotyping applications such as early abiotic stress detection.

Weed detection and removal is an essential task for a better crop yield as they compete for resources along with crops. Manual weed control methods are time-consuming and labour-intensive tasks with room for errors. The critical challenge is to reliably and accurately detect weed from the field. To achieve this, UAV and ground-based sensing along with deep learning, especially CNN models are used to automate crop management. CNNs based on encoder-decoder models are preferred for semantic segmentation. However, they perform poorly for low-level features such as noisy boundaries. This paper presents our work on paddy rice data collection for crop-weed segmentation and a network model based on an overcomplete representation (Kite-Net) augmented on a transfer learning-based encoder-decoder model (TernausNet) for segmenting the RGB image into three classes: crop, weed and background. The network results show convincing seg-mentation mask output on overlapping crop-weed images.

Water is essential for any crop production. Lack of sufficient supply of water supply causes abiotic stress in crops. Accurate identification of the crops affected by drought is required for achieving sustainable agricultural yield. The image data plays a crucial role in studying the crop's response. Recent developments in aerial-based imaging methods allow us to capture RGB maize data by integrating an RGB camera with the drone. In this work, we propose a pipeline to collect data rapidly, pre-process the data and apply deep learning based models to segment drought affected/stressed and unaffected/healthy RGB maize crop grown in controlled water conditions. We develop an ensemble-based framework based on U-Net and U-Net++ architectures for the drought stress segmentation task. The ensemble framework is based on the stacking approach by averaging the predictions of fine-tuned U-Net and U-Net++ models to generate the output mask. The experimental results showed that the ensemble framework performed better than individual U-Net and U-Net++ models on the test set with a mean IoU of 0.71 and a dice coefficient of 0.74.

The Light detection and ranging (LiDAR) sensor is used for perceiving the environment of an autonomous vehicle. LiDAR data or point cloud is processed to get the obstacles and their speed around an autonomous vehicle. Based on the information retrieved from LiDAR data and the data from other sensors, real-time decisions are taken for the proper navigation. Hence, the time taken in LiDAR data processing should be minimized. One of the important steps for LiDAR data processing is the segmentation of the obstacles. In this paper, we present a quantitative comparison between two different approaches for point cloud segmentation, Euclidean distance-based Cluster Extraction and Cylindrical range image-based method. Based on the simulation performed on ROS (Robot Operating System) platform, we found that the second method is much faster as compared to the first method. In addition to that, the second method can perform proper segmentation at a larger distance from the sensor.

The monitoring of growth stages of a crop is vital for farmers to optimize the use of agronomic inputs and crop-management. Manual observation of the growth stages of a crop in large fields is a time consuming and labor-intensive task. To reduce human efforts in this tedious work, Unmanned Aerial Vehicle (UAV)-based remote sensing with the emergence of different technologies like deep learning is helping in monitoring the health of a crop. However, Convolutional Neural Network (CNN) based models need a lot of labeled data and computations to get trained to automate the process. In this paper, a pixel-based segmentation method has been proposed for tassel detection, and estimation of growth stages like tasseling, and day to 50% tasseling of maize crop. The performance analysis shows that the proposed method reduces time in developing the dataset for the training of CNN models. It also gives an advantage over training-time and computational complexity when compared to CNN models like YOLO and Faster-RCNN.

Real-time data transmission is one of the challenging tasks for LiDAR (Light Detection and Ranging) based applications. These applications are becoming more popular in the field of Surveying and Intelligent Transportation System (ITS). The size of Point cloud data (pcd files) generated by LiDAR is generally quite large. In this paper, a real-time Point cloud transmission over Wi-Fi is suggested. Also in order to avoid the transmission of huge data, an Octree-based Point cloud Compression technique is used. This Compression technique is provided by Point cloud Library (PCL). We encode the Point clouds into text files of much lower size as compared to pcd file size. Instead of sending the Point cloud data we send the text files which is further decoded on the receiver side. The preliminary experimental results show that this method has the potential to be used for an exchange of information (3-D view or Point cloud) between two vehicles for Intelligent Transportation System.

The Internet of Things (IoT) with its plethora of applications brings up new challenges in optimizing power consumption, error performance, latency, and throughput in its communication devices. Recently, there has been increasing necessity of reconfigurable and multistandard transceivers in order to bring adaptability in the IoT devices to suit the nature of the application and satisfy the aforementioned performance metrics as well. In this paper, we focus on the efficient design of receivers compliant with IEEE 802.15.4 which is a prominent protocol for low power IoT applications. We propose a robust phase noise compensation method which efficiently removes the residual phase noise remaining after the coarse frequency offset compensation due to direct-sequence spread spectrum operation on large packets. We also propose a dual-mode receiver using the proposed compensation method and showcase its ability to cater to the diverse nature of IoT applications. The detailed architecture of proposed dual-mode receiver is presented along with its FPGA prototyping and ASIC implementation. We have analyzed overall power consumption by the proposed dual-mode receiver considering the packet error rate and retransmission scenario. The results show that the proposed receiver saves significant energy consumption by changing its mode in favorable channel environments.

In this letter, we propose two novel methods for four-class motor imagery (MI) classification using electroencephalography (EEG). Also, we developed a real-time health 4.0 (H4.0) architecture for brain-controlled internet of things (IoT) enabled environments (BCE), which uses the classified MI task to assist disabled persons in controlling IoT-enabled environments such as lighting and heating, ventilation, and air-conditioning (HVAC). The first method for classification involves a simple and low-complex classification framework using a combination of regularized Riemannian mean (RRM) and linear SVM. Although this method performs better compared to state-of-the-art techniques, it still suffers from a nonnegligible misclassification rate. Hence, to overcome this, the second method offers a persistent decision engine (PDE) for the MI classification, which improves classification accuracy (CA) significantly. The proposed methods are validated using an in-house recorded four-class MI data set (data set I, collected over 14 subjects), and a four-class MI data set 2a of BCI competition IV (data set II, collected over 9 subjects). The proposed RRM architecture obtained average CAs of 74.30% and 67.60% when validated using datasets I and II, respectively. When analyzed along with the proposed PDE classification framework, an average CA of 92.25% on 12 subjects of data set I and 82.54% on 7 subjects of data set II is obtained. The results show that the PDE algorithm is more reliable for the classification of four-class MI and is also feasible for BCE applications. The proposed low-complex BCE architecture is implemented in real time using Raspberry Pi 3 Model B+ along with the Virgo EEG data acquisition system. The hardware implementation results show that the proposed system architecture is well suited for body-wearable devices in the scenario of Health 4.0. We strongly feel that this study can aid in driving the future scope of BCE research.

Recent advances in the Brain-Computer Interface (BCI) systems state that the accurate Motor Imagery (MI) classification using Electroencephalogram (EEG) plays a vital role. In this paper, we propose a novel real-time feature extraction and classification architecture for four class MI using a combination of Kullback-Leibler Regularized Riemannian Mean (KLRRM) and Linear SVM. By using the KL regularization, the robustness of the features extracted to the noise and outliers is improved. The performance of the proposed architecture is analyzed on the four class MI dataset 2a from the BCI Competition IV. The performance analysis shows that the proposed architecture achieves an average classification accuracy of 74.43% and 51.53% for both the good and noisy subjects respectively. Also, the emphasis is laid on understanding the performance of regularization, and the improvement of robustness to the noise and outliers is demonstrated using the noisy subjects.

Fatty Liver Disease (FLD), if left untreated can progress into fatal chronic diseases (Eg. fibrosis, cirrhosis, liver cancer, etc.) leading to permanent liver failure. Doctors usually use ultrasound scanning as the primary modality for quantifying the amount of fat deposition in the liver tissues, to categorize the FLD into normal and abnormal. However, this quantification or diagnostic accuracy depends on the expertise and skill of the radiologist. With the advent of Health 4.0 and the Computer Aided Diagnosis (CAD) techniques, the accuracy in detection of FLD using the ultrasound by the sonographers and clinicians can be improved. Along with an accurate diagnosis, the CAD techniques will help radiologists to diagnose more patients in less time. Hence, to improve the classification accuracy of FLD using ultrasound images, we propose a novel CAD framework using convolution neural networks and transfer learning (pre-trained VGG-16 model). Performance analysis shows that the proposed framework offers an FLD classification accuracy of 90.6% in classifying normal and fatty liver images.

Fatty liver disease is the major cause for the liver dysfunction and is highly prevalent in developed and developing nations. Fatty liver can progress into chronic diseases if left untreated. Depending on the density of accumulated fat in liver, the liver is diagnosed into three classes/grades namely Grade 1, Grade 2, and Grade 3 apart from being Normal. For fatty liver diagnosis, doctors use the texture properties of liver parenchyma to quantify the fat in liver. The texture properties of ultrasonic liver parenchyma changes with the proportion of fat and acts as a useful feature for the doctors in doing the diagnosis. The diagnostic accuracy of fatty liver is less due to minute variations observed in the texture properties. Hence to improve the classification accuracy in diagnosing the fatty liver, we propose a convolution neural network based computer-aided diagnosis algorithm for categorizing the ultrasound liver parenchyma texture into four classes. The proposed algorithm is analyzed using 1000 texture images comprising of 250 images belonging to each class. Performance analysis shows that the proposed framework classifies the texture with an accuracy of 93.5% when 80% and 20% of data used for training and testing respectively.

Recently proposed IEEE 802.15.4-2015 MAC introduced a new prioritized contention access (PCA) for transfer of time-critical packets with lower channel access latency compared to carrier sense multiple access-collision avoidance (CSMA/CA). In this paper, we first propose a novel Markov-chain-based analytical model for unslotted CSMA/CA and PCA for industrial applications. The unslotted model is further extended to derive the analytical model for slotted CSMA/CA and PCA. Primary emphasis is laid on understanding the performance of PCA compared to CSMA/CA for different traffic classes in industrial applications. The performance analysis shows that the slotted PCA achieves a reduction of 63.3% and 97% in delay and power consumption respectively compared to slotted CSMA/CA, whereas unslotted PCA achieves a delay reduction of 53.3% and reduction of power consumption by 96% compared to unslotted CSMA/CA without any significant loss of reliability. The proposed analytical models for both slotted and unslotted IEEE 802.15.4-2015 MAC offer satisfactory performance with less than 5% error when validated using Monte Carlo simulations. Also, the performance is verified using real-time testbed.

In this paper, we propose a novel IEEE 802.15.4 media access control (MAC) power management scheme that achieves the user specified reliability with minimal power consumption at the node. Also, we develop an accurate mathematical model to analyze the effects of constrained on-node memory for sensed data storage on the MAC layer performance. We use three-dimensional Markov chain and M/G/1/K queue to model the IEEE 802.15.4 MAC and on-node packet queue, respectively. By formulating the precise packet service time, the reliability, packet queue overflow losses, delay, and power consumption of the node are analyzed. When compared with simulations and the real-time test bed, the proposed model achieves an accuracy of 97% and 94%, respectively. Also, the performance analysis shows that the proposed power management scheme provides energy savings of up to 74.82%.

Brain Computer Interface (BCI) has recently gained much popularity due to plethora of its applications. In this paper, we propose a novel system architecture to utilize brain signals for controlling Internet of Things enabled environments. The proposed architecture aids in translating brain signals to commands that interact with or control the environment using IoT actuation networks thus executing user desired actions. It comprises of novel, low complex and low power intelligent signal processing architecture for detection of voluntary eye blinks by isolating involuntary eye blinks and IoT enabled wireless actuation network for controlling the environment using commands generated from EEG signal. For the real time performance analysis of the proposed architecture, we developed a wearable device which acquires dual channel EEG using electrodes at Fp1 and Fp2 locations. From the acquired EEG data, the device detects the voluntary eye blinks of the patient and use this information in controlling the environment such as switching HVAC system, lighting or electric fan etc. Performance analysis shows that the proposed intelligent signal processing architecture detects the voluntary eye blinks with 95.2% accuracy when tested on 10 subjects with a low power consumption of 165 mW.

In conventional telesonography, the ultrasound video is encoded at a constant bit rate and transmitted to the expert side for diagnosis, this is highly bandwidth demanding and less adaptable to the varying network conditions. In contrary, Internet based videophone applications adapt to the network conditions with variable bit rate encoding and error concealment algorithms. Since portable ultrasound scanners are coming with Internet connectivity, this motivated us to evaluate the viability of using existing Internet based videophone services for real-time telesonography. Popular videophone services like Skype, Facebook and WebRTC is chosen for this purpose. The performance of these services is analyzed by varying network parameters like bandwidth, packet loss, delay and packet error in a controlled set up. Liver ultrasound video is considered for the study. A total of 273 different network settings is analyzed for evaluating the performance of each videophone service, constituting a total of 819 instances for evaluating three services. Three medical experts were participated in the subjective study. From an extensive analysis, an inference is made that the Internet based videophone services are reliable enough for telesonography provided a network with minimum bandwidth of 1600 kHz, delay up to 200 ms, packet loss up to 3% and packet error up to 3%.

With the proliferation of Internet of Things (IoT), the IEEE 802.15.4 physical layer is becoming increasingly popular due to its low power consumption. However, secure data communication over the network is a challenging issue because vulnerabilities in the existing security primitives lead to several attacks. The mitigation of these attacks separately adds significant computing burden on the legitimate node. In this paper, we propose a secure IEEE 802.15.4 transceiver design that mitigates multiple attacks simultaneously by using a physical layer encryption approach that reduces the computations at the upper layers. In addition to providing confidentiality and integrity services, the proposed transceiver provides sufficient complexity to various attacks, such as cryptanalysis and traffic analysis attacks. It also significantly improves the lifetime of the node in the presence of a ghost attacker by preventing the legitimate node from processing the bogus messages and hence combats against energy depletion attacks. The simulation results show that a high symbol error rate at the adversary can be achieved using the proposed transceiver without affecting the throughput at the legitimate node. In this paper, we also analyze the hardware complexity by developing an FPGA and ASIC prototype of the proposed transceiver.

Medical ultrasound scanning is a widely used diagnostic imaging modality in health-care. Speckle is inherent noise present in ultrasound images reducing the diagnostic accuracy of ultrasound scanning. Speckle noise contributes to high variance between pixels and delineates boundaries of the organs. Effective despeckling involves reducing the variance between pixels corresponding to homogeneous region and to preserve anatomical details simultaneously. Non-Local Means filters are highly successful and produced state of the art results in despeckling ultrasound images. In this paper, we show the effectiveness of Non-Local Means filter with polynomial regression kernel in despeckling ultrasound images. The proposed algorithm is evaluated on software simulated and real time ultrasound images and proved very effective in both despeckling and edge preservation.

Accurate model-based methods in Photo-Acoustic Tomography (PAT) can reconstruct the image from insufficient and inaccurate measurements. Most of the models either make the simplified assumption of spherical averaging or use accurate models that have computationally burdensome implementations. We present a simple and accurate measurement matrix that is derived from the pseudo-spectral PAT model. The accuracy of the measurement matrix is first validated against the experimental PAT signal. We also compare the model against the standard k-wave measurement model and the spherical averaging model. We then highlight several reconstruction strategies based on the nature of the region of interest to further demonstrate the accuracy of the proposed measurement matrix.

Telesonography suffers from inherent limitations due to the need of all time availability of experts in cloud and data connectivity to the device. Computer-aided diagnosis (CAD) used for automatic detection of abnormalities without manual intervention can overcome these limitations. Commercially available ultrasound scanners restrict the installation of new softwares and hence CAD algorithms cannot be integrated into the existing ultrasound scanners. There is a need for an external computing device, which can acquire image data from ultrasound scanners, perform CAD and generate result. Smart-phones are now widely used in personalized healthcare due to its ubiquitous computing capability. Smartphones with embedded CAD can be used as a computing device for automated diagnosis. In this paper, we have developed an Application (APP) for a smartphone to automatically diagnose the kidney in the ultrasound image. With the developed APP, the smartphone can acquire images from any ultrasound scanner, process it and give the diagnostic result. Automatic abnormality detection of kidney is based on Viola Jones algorithm, texture feature extraction followed by SVM classifier. Stones and cysts are the abnormalities detected using the algorithm. The developed APP resulted with an accuracy of 90.91% in detecting the abnormalities.

Fast diagnosis plays a crucial role in treating the patients, which can be life saving manytimes. Ultrasound scanning have the capability to image the organs in real time, thus offering quick diagnosis. Recent advancements in computing platforms greatly reduced the size of an ultrasound machine to portable level. Portable ultrasound machines can be used for point of care and remote diagnosis, this comforts the patients by making them not going to the ultrasound scanner for diagnosis. Diagnosing the patients at their bedside reduces the diagnosis and medication time. Lack of sonographers limits the benefits of using portable ultrasound machines in remote diagnosis. Tele-radiology can be effectively used as a solution to provide better health-care to more number of patients with less number of experts. Portable ultrasound in remote diagnosis is concerned with wireless connectivity for web-based diagnosis, security, data aggregation of patients etc., In this paper, we propose a working prototype of portable ultrasound scanning system integrated with wireless connectivity, biometric authentication, Global Positioning System (GPS). Additional interfaces to the portable ultrasound system enables secured cloud based diagnosis, improving the health-care and reducing the geographical separation between patients and doctors.

High sampling rate is necessary for a quality ultrasound image, which demands expensive data-acquisition and computing devices. Compressive sensing(CS) can reconstruct high quality image with less data. It can give optimal solution to high sampling problem in ultrasound imaging. Ultrasound imaging is performed using beamforming of transducer array elements. In this work we present time and frequency domain beamforming matrices and demonstrates how it can be used as CS-matrix to reconstruct ultrasound images. Feasibility of CS with the beamforming matrices are studied using transfer point spread function. Compared to previous work in ultrasound using CS where signal reconstruction is used from undersampled data, we present direct ultrasound image reconstruction from highly undersampled received data. Image reconstruction with time and frequency domain beamformed CS-matrix are showed. Through our results it is clear that compressive sensing in ultrasound imaging can significantly reduce sampling rate by maintaining same image quality as traditional ultrasound imaging.

Classification of medical data is one of the most challenging pattern recognition problems. As stated in literature a single classifier is unable to solve all medical image classification problems due to high sensitivity to noise and other imperfections like data imbalance. So, several individual classifiers have been studied to solve the different types of classification problems arising in medical datasets but all have proven to be useful on some specific datasets. Hence, in this paper, we propose a generic multi-level classification approach for medical datasets using sparsity based dictionary learning and support vector machine approaches. The proposed technique demonstrates the following advantages: 1) gives better performance of classification accuracy over all datasets 2) solves imbalanced data problems 3) needs no fusion and ensemble methods in multi-level classification. The results presented on the 5 standard UCI medical datasets demonstrate that the efficacy of the proposed multi-level classification technique.

Photo-Acoustic Tomography (PAT) combines ultrasound resolution and penetration with endogenous optical contrast of tissue. Real-time PAT imaging is limited by the number of parallel data acquisition channels and pulse repetition rate of the laser. Typical photoacoustic signals afford sparse representation. Additionally, PAT transducer configurations exhibit significant intra- and inter- signal correlation. In this work, we formulate photoacoustic signal recovery in the Distributed Compressed Sensing (DCS) framework to exploit this correlation. Reconstruction using the proposed method achieves better image quality than compressed sensing with significantly fewer samples. Through our results, we demonstrate that DCS has the potential to achieve real-time PAT imaging.

Electrical devices which are controlled remotely by a smartphone have recently been spreading. With some typical existing systems, a user needs to identify a target device to control with only its less user-friendly ID. A system for association between actual position of the device and its position on smartphone display would allow the user to identify it easily. We call such a system “Smart Interaction System”. One of the key technologies for Smart Interaction System is localization of the device. Localization technique using some anchors with the given position to measure received signal strength indication (RSSI) of a device have been developed. They incur user burden to deploy the anchors and measure their positions. Thus alleviating the burden is still a technical challenge. As a solution to the challenge, this paper proposes a new 3D localization technique with a mobile robot including a floor cleaning robot. The mobile robot would reduce the burden by measuring RSSI while cleaning floor instead of the anchors. The proposed localization technique has been implemented with a floor cleaning robot. Experimental results show that the proposed localization technique provides the position of the device with approximately 1000mm of estimation error and is useful for Smart Interaction System.

Smart phones or mobile phones enabled with global positioning system (GPS), different types of sensors, and communication technologies have become ubiquitous application development platform for Internet of Things (IoT) and new sensing technologies. Improving sensing area coverage, reducing overlap of sensing area, and energy consumption are important issues under mobile phone sensing. This paper presents human mobility-based mobile phone sensors sampling algorithm. Human mobility patterns and geographical constraints have an impact on performance of mobile phone sensing applications. The real-outdoor location traces of volunteers, collected using GPS-enabled mobile phones are used for performance analysis of proposed work. The proposed mobile phone sensor sampling algorithm considers velocity of human mobility as an important parameter for improving sensing area coverage and reduction of energy consumption. To an extent overlap between sensing area coverage is allowed to overcome, the reduction of sensor data samples caused by spatial regularities of human mobility. The performance is analyzed and evaluated by considering general regular sampling and proposed sampling method for mobile phone sensing activity. The results show that for normal human walking velocity (<;1.5 m/s) proposed mobile phone sensor sampling algorithm performs better in terms of sensing area coverage and reduction of battery energy consumption for mobile phone sensing activity.

In commercial ultrasound systems, the transmit module typically generates the time delayed excitation pulses to steer and focus the acoustic beam. However, the ultrasound transmitter module in these systems has limited access to medical ultrasound researchers. In this paper, we have presented the development of a programmable architecture for 8-channel ultrasound transmitter for medical ultrasound research activities. The proposed architecture consists of 8 transmit channels and Field Programmable Gate Array (FPGA) based configurable delay profile to steer acoustic beam, transmit frequency and pulse pattern length depending on the medical application. Our system operates in pulse-echo mode, with ultrasound transmit frequency up to 20 MHz, excitation voltage up to 100 Vpp, and individual channel control with single high speed Serial Peripheral Interface (SPI). Pre-calculated delay profiles per scanline are generated in Matlab, based on physical parameters of 8 element linear transducer array which are used to steer and focus the ultrasound beam. An experiment is carried with our transmit module to transmit ultrasound into gelatin phantom, acquired echoes and processed for B-mode imaging. The results show that this transmit platform can be used for ultrasound imaging researches and also for medical diagnosis.

Ultrasound imaging has been widely used for preliminary diagnosis as it is non-invasive and has good scope for the doctors to analyze many diseases. Lack of trained sonographers make ultrasound imaging diagnosis time consuming to detect any abnormality. Sometimes the problem cannot exactly be identified which may lead to error in diagnosis. Hence in this paper we present computer aided automatic detection of abnormality in kidney on the ultrasound system itself, to decrease the time for reports and not to depend on the sonographer. We classified the kidney as normal and abnormal case. Segment the kidney region and extract Intensity histogram features and Haralick features from Gray Level Cooccurnace Matrix (GLCM). These features are calculated for a set of large data containing both normal and abnormal cases. Abnormal case includes kidney stone, cyst and bacterial infection. Standard deviation for each parameter is observed, considered only those features with less deviation and implemented on FPGA Kintex board. If the range of mean value is 1.08 to 1.336, skewness is 2.882 to 7.708, Kurtosis is 1.06 to 71.152, Cluster Shade is 72 to 243, Homogeneity is 0.993 to 0.998, the observed kidney image is normal otherwise abnormal.

Carbon Monoxide (CO) induced by traffic pollution is highly dynamic and non-linear. In a pilot research, we collected some fine-grained 1Hz CO pollution data from a residential road and a busy motorway in Hyderabad, India, in preparation of the deployment of a larger scale, longer term wireless sensor monitoring system. Power conservation is an important issue as the sensor nodes are battery operated. We studied the characteristics of the collected data and designed an adaptive sampling algorithm, Bi-Scale temporal sampler, which adapts the sampling frequency to the statistics collected in real time. This design has incorporated practical engineering considerations including minimising electronic noise, sensor warm-up time and data characteristics. Results show that Bi-Scale sampler achieves better energy saving and statistical deviation ratio for our requirements than burst sampling and eSENSE sampling strategies, which are techniques popularly used in environmental monitoring applications.

Standard 12-lead (S12) system and Mason-Likar 12-lead (ML12) system despite of being most acceptable systems for clinical usage are not the preferred lead systems for remote monitoring (RM) applications. Usually RM applications involve wireless transmission of signals and a 2-3 lead system is preferred for bandwidth and storage limitations and data transmission time. Generally, ECG compression techniques are applied for the same, however, compression ratio (CR) depends on the number of channels and decreases with the increase in number of channels. Thus, it facilitates the usage of a 2-3 lead system. However, a reduced lead (RL) system with 2-3 leads may be inadequate for the information desired by the cardiologists who are accustomed to S12 or ML12 system pertaining to its decades old usage. In this paper, we attempt to provide solution to both technical and non-technical limitations of RM applications. We reconstruct S12 and ML12 systems from Reduced 3-lead (R3L) system comprising of basis leads I, II, V2using personalized or patient-specific transformation. Two separate investigations have been carried out for S12 and ML12 with their corresponding R3L systems comprising of their respective basis leads. PhysioNet PTBDB and INCARTDB after wavelet based preprocessing were used in this investigation. R2statistics, correlation (rx) and regression (bx) coefficients were used to evaluate reconstructed signal against the original signal and the mean values obtained were 96.53%, 0.982 and 0.968 (S12) and 96.53%, 0.982 and 0.968 (ML12) respectively. R3L system reduces number of leads and electrodes from 12 and 10 to 3 and 5 respectively, lowers bandwidth and storage requirements, data transmission time and increases CR. The study shows that basis leads obtained from S12 outperforms the basis leads of ML12 for reconstruction of precordial leads.

In this paper, we have proposed an analytical model for optical networks with full wavelength conversion at the nodes. We have derived an analytical expression to compute the carried traffic on links of the network for fixed routing with uniform traffic distribution (UTD). The carried traffic on a link in the network is thinned proportionately based on the blocking probability of the other links on the route. The blocking probability of the network is estimated using Erlang fixed point approximation with the reduced load on the links. The channel utilisation at a particular load is derived using the blocking probability. Thus the analytical model gives an estimate of the blocking probability and the channel utilisation at any given load and is applicable to any network topology. We have computed the carried traffic for a few example networks, such as 14 node NSFNET, 20 node ARPANET and 20 node INDIANET and validated the analytical results with simulations. We show that the analytical method performs well in the desired range of blocking probabilities and it is computationally efficient.

We present an analytical tool to enhance the blocking performance in the circuit-switched wide-area optical wavelength division multiplexed (WDM) networks with no wavelength conversion at the nodes. Given any network of arbitrary topology, the aim is to see if one can achieve the wavelength conversion (optimal) performance by using wavelength reassignment techniques in no conversion networks. If there is significant performance deviation, the tool identifies the critical points in the network which prevents the reassignment to totally removewavelengthcontinuityconstraint(wcc) blocking. Once the critical points are identified, the tool appropriately modifies the routing such that the wavelength reassignment can achieve the optimal performance.