Black carbon (BC), a potent component of air pollution, is particularly concerning due to its multifaceted role in contributing to climate change and health. Thus, accurate BC mass concentration prediction is crucial. Recently, machine learning (ML) methods for predicting BC concentrations are gathering momentum due to the challenges of employing physics-based models, which necessitate thorough process comprehension. The main objective of this paper is to develop a grey wolf optimization (GWO)-based support vector regression (SVR) model that utilizes the GWO’s optimization capabilities to enhance the SVR to predict near-surface BC mass concentrations. The study utilizes the primary data of nearsurface BC gathered at the national network of aerosol observatories, named the Aerosol Radiative Forcing over India NETwork (ARFINET) under ISRO from 2002 to 2020 across the Trivandrum region of India. The proposed model’s performance, evaluated against the seasonal autoregressive integrated moving average (SARIMA) and conventional SVR models through metrics such as root mean square error (RMSE), mean absolute error percentage (MAPE), and the coefficient of determination $\left(R^{2}\right)$, demonstrates superior predictive accuracy. It achieves the lowest RMSE and MAPE and the highest $R^{2}$ during both training and testing phases, indicating its effectiveness over existing methods. Furthermore, the model forecasts BC concentrations for the next three years. The outcomes highlight the proposed model’s potential for effectively predicting BC mass concentrations and serve as a viable tool for future air quality control efforts.

Accurately predicting aerosol optical depth (AOD), a key parameter for characterizing atmospheric aerosols, is essential due to the increasing prevalence of air pollution and its detrimental effects. From existing literature on AOD time-series prediction, linear models like seasonal autoregressive integrated moving average (SARIMA) are commonly used, while nonlinear models such as machine learning (ML) and deep learning (DL) have gained popularity recently for their ability to handle nonlinear patterns. This study introduces a hybrid model, particle swarm optimization-seasonal autoregressive integrated moving average-support vector regression (PSOSARIMA-PSOSVR), which integrates linear and nonlinear modeling through residual modeling to significantly enhance AOD prediction accuracy. By combining these approaches, we aim to better handle the nonlinearities and overall variability in AOD data, leading to more accurate predictions. To address hyperparameter tuning challenges, including the risks of model misspecification and overfitting or underfitting, PSO is utilized for optimization. PSO’s evolutionary optimization ensures efficient tuning for optimal model performance. Monthly AOD data sourced from the moderate resolution imaging spectroradiometer (MODIS) satellite covering the northern Indian region from 2001 to 2019 is used for experiments. Performance metrics such as root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination ( R^2 ), Nash-Sutcliffe efficiency (NSE), and root mean square error ratio (RSR) are employed to evaluate model accuracy and dependability. The results of this study illustrate that the proposed PSOSARIMA-PSOSVR model outperforms both standalone PSOSARIMA and PSOSVR models. Moreover, it consistently surpasses SARIMA, SVR, and long short-term memory (LSTM) in AOD prediction, effectively addressing non-stationarity and variability challenges in AOD data. This study suggests the hybrid model as a promising approach for improving the accuracy of AOD prediction.

Intruder detection and alert systems are an essen- tial component for both residential and commercial premises. Most surveillance systems in remote locations depend on the availability of the internet on-site to transmit alert signals. This paper presents a novel experimental intruder alert system that doesn't use the internet on-site for the alarm mechanism. using an FMCW Radar and other peripherals such as camera, RaspberryPi and LoRa. The system leverages the long-range and low power consumption capabilities of LoRaWAN to transmit signals, while the FMCW radar and camera work together to detect and track intruders. To further enhance the accuracy of the system, the Yolov4 model is used for detection. Overall, this system has potential to significantly enhance security measures in remote locations.

Maintaining optimal tip clearance or tip gap is challenging in the Gas Turbine Engine (GTE). Meanwhile, the rotor blades should not rub the casing. When the capacitive sensor is used to measure the tip clearance in the form of a single peak signal for every blade pass, often the signal will be affected by stationary and non-stationary noises during engine running. This leads to distorted multiple peaks for every blade pass. In this work, the wavelet denoising technique removes the noise, and then the peak frequency in each blade pass is detected through a short-time Fourier transform (STFT). Finally, the cubic spline interpolation technique is employed to obtain the continuous time domain blade pass signal. This work uses the compressor stage of GTE data collected from the Gas Turbine Research Establishment (GTRE), DRDO, Bangalore. From the experimental analysis, this paper observes that the proposed methodology produces substantial results compared to the expected results.

This study focuses on character recognition techniques for vehicle number plate recognition. It employs the K Nearest Neighbor (KNN) algorithm to detect the characters in the vehicle's number plate. In addition, it successfully identifies the state of origin of vehicles from the vehicle number thereby easing documentation of vehicles. Moreover, the algorithm marks the time of entry of vehicles in a systematic manner. The algorithm provides a methodological solution to keep track of the numerous vehicles entering a premise in an orderly fashion.

Millimeter wave radar (mm-Wave radar) technology has paved the way for non-contact human-machine interaction using gestures. The proposed work uses a 60 Giga Hertz Frequency Modulated Continuous Wave Multiple Input Multiple Output Radar (FMCW MIMO radar) to perform hand gesture-based applications. In the paper, 2 different gestures namely, forward and backward are classified and used in the application. Data for the aforementioned gestures was extracted using Texas Instrument's 60 GHz FMCW MIMO radar from a group of 10 people. For data classification, the Micro-Doppler signatures of the gesture data are extracted. The classification of these micro-Doppler features is done by using a vanilla Convolutional Neural Network (CNN) algorithm.

Automatic speech recognition is being used everywhere these days. An essential part of this is language identification. Our goal here is to identify the language of the media title, such as song names and movie titles, to help in speech recognition. The focus here is to classify solely using the title of the media without any additional data in their transliterated form to classify them into their original native language using natural language processing, machine learning, and deep learning techniques. Transliterated titles of the song and movie names are being used. This work explores and implements various natural language processing and machine learning methods such as N-grams, SVMs, LSTMs, and MuRIL to classify the text titles according to their language. The results of various implementations are compared and contrasted as an approach of its own to classify the data.

Accurate aerosol optical depth (AOD) forecasting aids in the air quality analysis as it is an important parameter that provides a quantitative measure of aerosols existing in the atmosphere. To an extent, to date, AOD forecasting has been limited to conventional statistical methods such as simple linear regression, multiple linear regression, and Seasonal Auto-Regressive Integrated Moving Average (SARIMA). Prophet models have gained popularity recently due to their simple interpretable parameters and efficiency in dealing with strong seasonal time-series data. This paper aims at the development and analysis of Prophet Model for the AOD forecasting across Delhi, Mumbai, Kolkata and Trivandrum in India based on the monthly mean AOD(550nm), data obtained from Moderate Resolution Imaging Spectroradiometer (MODIS) over 19 years (2001 to 2019). Conventional SARIMA model is used to validate the performances of the Prophet model. The forecasts for the upcoming 5 years were implemented at a 95 percent confidence interval. The model evaluation was carried out based on the performance metrics and the execution time taken. The performance metrics considered are the correlation coefficient (R), root mean square error (RMSE) and mean absolute percentage error (MAPE). The Prophet Model performed reasonably better than the SARIMA model with the lowest MAPE and RMSE in all four regions. The correlation obtained by the Prophet Model are 0.83, 0.79, 0.522, and 0.55 and by SARIMA are 0.81, 0.76, 0.520, and 0.51 across Delhi, Mumbai, Kolkata and Trivandrum respectively. Prophet based models took only a few seconds to execute, unlike SARIMA which took a minimum of half an hour.

Due to the increasing complexity of modern multi-functional radars in the electromagnetic environment, it is a challenging task to classify and identify the presence of different radar emitters. The presence of multiple number of active transmitters in the multistatic radar system makes radar emitter identification a big data problem as all are emitting dense complex signals in the electronic reconnaissance field. In order to classify and identify the radar emitters accurately and rapidly many researchers proposed numerous algorithms. To determine the radar emitter identification methods developed, this paper reviews various methods and techniques through a literature survey and classification of articles (collected from the online database) has been made from various algorithms and methods point of view.

One of the major problems in target tracking is to identify the measurement of the interested target among the bulk of data. Many algorithms based on identifying single measurement among the received data were presented. However with increase in false alarm rate these models failed. Instead of identifying single measurement among the received signals and discarding others, alternative approach is to associate each measurement with different weights and this approach is called Probabilistic Data association (PDA). However performance of PDA algorithm is visibly degraded for multiple targets. Hence this paper presents the integration of PDA with K-nearest neighbors’ algorithm (KNN) to effectively track multiple targets and reduce the complexity in multiple target tracking. KNN classifies the measurements among targets and then PDA algorithm is applied to each target. Hence the multiple targets tracking problem is split into multiple- single target tracking problem with the help of KNN. Simulations were done using MATLAB and their performances are presented in this paper.

The Automatic Identification System (AIS) is one of the most successful maritime technologies which allows ships to autonomously communicate their positions and movements with one another, thereby playing a crucial role in collision avoidance. AIS is essentially a communication system that identifies and exchanges information such as the MMSI (Maritime Mobile Service Identities) number, navigation status, rate of turn, speed over ground, course, position, and etc with other maritime vessels and base stations. This paper aims at the development and implementation of a simulator that simulates AIS communication based on AIS standards, i.e., capturing the necessary AIS messages and displaying them on the chart plotters or live maps.

Automotive accidents are a leading cause of severe injuries and loss of life. The chances of survival and recovery for the victims depend greatly on the time taken for them to receive medical attention. In this paper, the development of a vehicular accident detection and alert system has been discussed. The main purpose of the device is to help reduce Emergency Response Services (EMS) response time by eliminating the delay in alerting emergency services of an accident. The system consists of a standalone device that can be installed into any vehicle coupled with a mobile application running on one of the passengers’ smartphones. Henceforth referred to as the ‘Black Box’, the device detects accidents under various scenarios, such as a head-on collision, or toppling of the vehicle. In case an accident is detected, an alert SMS is automatically sent to the nearest EMS through the custom developed mobile application, with the coordinates of the site. An abort function has also been incorporated for cases of false-positive decisions.

This paper focuses on identification of helicopter by exploiting the concept of micro-Doppler effect which is prominent in targets containing rotating, oscillating or vibrating parts in it. Radar received signal is analyzed by Short Time Fourier Transform (STFT) to extract the micro Doppler (mD) signature. From the mD signature, the helicopter parameters are estimated. In a multiple helicopters scenario, estimated parameters will be a mixure, pertaining to the multiple helicopters. These parameters are classified further using a machine learning algorithm, namely k-means clustering to classify the helicopters. Simulated results for the synthesized received signal shows the betted estimates of the helicopter parameter through mD signature. Dataset containing basic parameters like number of blades, blade length and rotational rates of the UN-1N helicopter (rotor with 2 blades), the SH-3H helicopter (rotor with 5 blades) and the CH-54B helicopter (rotor with 6 blades) are considered for the classification. Results show a good classification. When analysed with different SNR level in dataset, at lower SNR, observed some ovelapping in the classification.

Over the last decade, research on multistatic radar waveform design has attracted significant consideration among the radar system designers.Among the waveform design techniques, linear frequency modulated waveforms are widely used, but the problem associated with it is range Doppler coupling and complexity in achieving adequate pulse-to-pulse diversity. In this letter, an orthogonal linear frequency modulated (OLFM) waveform has been used as the transmitted waveforms, which has good correlation properties and large time bandwidth product. The ambiguity function (AF), which is the output of the matched filter, is used to analyze the target resolution capabilities of the OLFM waveforms. The mathematical expressions of the AF with OLFM waveforms for the multistatic architecture are represented based on the target fluctuations observed at the receiver. These expressions allow the ambiguity in resolving target position and velocity vectors for any transmitted waveform in the multistatic topology. The normalized cross-correlation value of the LFM waveforms is compared with the OLFM waveforms, which signify the isolation factor between two transmitted waveforms at the receiver. A performance comparison has been made based on the probability of target detection for various multistatic scenarios under numerous signal-to-noise ratio (SNR) conditions.

AbstractTime delay estimation (TDE) is an active research area due to its importance in a wide range of applications, such as target detection and transmitter identification. At present, the majority of the radar systems are using multistatic architecture because of its improved detection and tracking performance. In multistatic radar systems, where more than one transmitter and receivers are deployed, it is a challenging task to separate multiple echoes at the receiver to estimate the required parameters and identify the number of transmitters. The delay estimation process becomes more complicated when closely spaced multiple targets are involved in the detection process. In this study, various issues associated with conventional matched filtering‐based target detection techniques are investigated in terms of nonlinear time‐frequency representation with ambiguity function and cross‐ambiguity functions. The problems of delay estimation and cross‐range resolution are analyzed through simulations in a multistatic radar system.

Classification of target from a mixture of multiple target information is quite challenging. In This paper we have used supervised Machine learning algorithm namely Linear Regression to classify the received data which is a mixture of target-return with the noise and clutter. Target state is estimated from the classified data using Kalman filter. Linear Kalman filter with constant velocity model is used in this paper. Minimum Mean Square Error (MMSE) analysis is used to measure the performance of the estimated track at various Signal to Noise Ratio (SNR) levels. The results state that the error is high for Low SNR, for High SNR the error is Low.

The implementation of multistatic radar architecture for the detection and tracking of targets engendered the development of numerous sensor data fusion models. This article presents a sensor model based on a probabilistic approach for the data fusion in the multistatic radar. The received polar coordinate measurements are converted to the Cartesian coordinates for the state estimation of the target using a Kalman filter. Then a distributed probabilistic maximum likelihood estimation based distributed data fusion method with five sensors is implemented for the data fusion. The simulation results validate the effectiveness of the data fusion algorithm under various measurement noise conditions. It is observed that the performance of the probabilistic data fusion algorithm performs better as compared with the traditional data fusion algorithm.

In signal processing, the noise is one of the factors which make the extraction of an original signal complex. There are so many techniques to discard noise and to get back original information. In general, to discard noise Filtering is a technique used, Wiener filtering technique is one of the filtering techniques which filter a noise-corrupted signal by linear-time-invariant-filtering (LTI), and Wiener filter reduces the mean square error value [MSE] between the observed signal and original signal. in general all male voices are base-band signals, if these signals are mixed up with a random noise then the noise corrupted data and original information are not independent to each other so the wiener output gives us the combination of both signals one as a major component and other as a minor component, so here we are making output signals of Wiener filter orthogonal using some mathematical relations such that these output male signals (base-band signals) become independent and we can use wiener filter for these independent signals to get the desired output. The minimum mean square error analysis results were observed for initial wiener outputs and orthogonal signals wiener output at different SNR levels.

Data fusion technique combines data from two similar sensors placed in different location in order to reduce the error in filtered state estimate. In this paper, state vector fusion (SVF) and measurement fusion (MF) are used to fuse the bearing measurement and also to fuse the derived heading measurement. The derived heading from bearing measurement increases the accuracy of target state estimate. Here, the Lagrange three point difference (LTPD) method has been proposed to derive heading from the set of bearing measurements. Two sensors with single target scenario are considered and the heading parameters are derived for each sensor. The bearing and derived heading measurements from two different sensors are fused using SVF or MF and then nonlinear Extended Kalman filter (EKF) is used to obtain the optimized state estimate. Simulations have been carried out in order to compare the SVF and MF fusion techniques for the bearing measurements as well as the derived heading parameters using existing centered difference (CD) and proposed LTPD.