The rapid increase in transportation vehicles in urban areas has led to several challenges, such as long traffic queues, increased carbon emissions, and other environmental and infrastructural issues. To address these problems, governments encourage citizens to use public transportation services instead of private vehicles. However, to rely effectively on public transport, individuals need to be aware of transit schedules as well as potential delays caused by the dynamic nature of smart cities and changing weather conditions. Conventional transit tracking systems mainly emphasize real-time monitoring and offer limited predictive insights into service reliability. To overcome this limitation, the proposed intelligent public transportation analytics system features a Flask-based backend integrated with machine learning models that analyze historical data, traffic density (using the TomTom API), and weather patterns to predict potential delays. The processed data is stored and displayed through a React-based frontend dashboard, allowing users to interactively view performance metrics, real-time updates, and predictive analytics. The system provides insights for commuters and authorities by identifying factors like weather, peak-hour congestion, and special events. By integrating multiple data sources and predictive models, it enhances transit reliability, optimizes route planning, and supports data-driven urban traffic management.

Content-Based Video Retrieval (CBVR) systems identify videos similar to a query by directly analyzing visual content, avoiding dependence on textual descriptions. In this paper, we propose a CBVR system that introduces a novel adaptive keyframe selection strategy aimed to preserve the most informative visual content. Unlike existing methods that employ fixed similarity thresholds and a uniform number of keyframes across all videos, the proposed approach dynamically determines both the similarity threshold and the number of keyframes for each video based on its content, enabling video-specific adaptability. First, a novel keyframe selection pipeline based on ResNet50 is used to extract deep features from individual video frames. Redundant frames are removed by retaining only those with minimal feature similarity, where the similarity threshold is automatically determined for each video using the Elbow method. Second, an optical flow–based frame selection strategy is applied to refine the selected keyframes by incorporating motion information. Deep spatiotemporal features are then extracted from both using a 3D deep learning model, and their fusion into a unified video-level representation constitutes an additional novel contribution, as it jointly exploits complementary content and motion cues rather than relying on a single information source. The proposed system achieved an 18.5% F1-score gain on the SumME dataset for video summarization and improved video retrieval performance on UCF101 and HMDB51, with Top-10 accuracy gains of 20.4% and 34.2%, Top-20 mAP gains of 46.7% and 68%, respectively, and a 1.6% Top-10 precision on UCF101 over the closest competing methods.

The journal retracts the article “Improving Efficiency of Large RFID Networks Using a Clustered Method: A Comparative Analysis” [...]

Imagine searching for a specific scene in a movie, not by remembering its title or actors, but by describing the action itself. This is the essence of content-based video retrieval (CBVR), a technique that searches for a video based on what’s inside it, rather than relying solely on manually assigned labels. Unlike traditional methods, which can be time-consuming, error-prone, and struggle with vast datasets, CBVR offers a more efficient and accurate approach.Our proposed system leverages the strong capability of deep learning, a subset of artificial intelligence, to analyze videos and extract their key characteristics. This process occurs in two stages: offline and online. Through the first stage, important features are extracted from all videos in the dataset and stored for future use. When a user submits a query video, its features are extracted in real-time (online) and compared to the stored features of all videos. The videos with features most similar to the query, essentially those with the “closest match,” are then presented to the user.To capture the full essence of a video, our system employs a two-stream neural network architecture. This innovative approach allows us to extract both temporal features, which capture the changes and motion patterns within the video (think: someone running or jumping), and spatial features, which pivot about the static visual content of each individual frame (think: the objects and scene depicted).By utilizing a pre-trained neural network called ResNet-60, our system benefits from existing knowledge and can efficiently extract meaningful features from videos. To evaluate its effectiveness, we tested our system on the UCF101 dataset, a widely used benchmark consisting of 101 categorized videos. Our approach obtained accuracy 93,7% for top 5 retrieval and 95.95% for top 10 retrieval. The outcomes illustrate that our approach obtains superior accuracy compared to other state-of-the-art video retrieval methods.

Data is the new oil, and in today’s data-driven world, the ability to generate, store, retrieve, and derive meaningful insights from it is crucial. This data often includes unstructured multimedia content, which poses unique challenges for effective management and utilization. Videos are crucial as they contain data such as object audio motion text. Content-based video retrieval (CBVR) is critical for managing and searching large multimedia databases, as it enables retrieval based on the actual content of videos rather than metadata or textual descriptions. Traditional retrieval methods often fail to achieve accuracy due to limited content representation capabilities. In response, the paper proposes a CBVR system that utilizes a deep learning based video representation and a classification approach utilizing a pre-trained ResNet-152V2 model to learn feature representation and classification scores. Experiments were conducted on the DGX-1 server using the UCF-101 Human Action Recognition Dataset, which comprises 13,300 videos across 101 categories. The proposed method outperformed the most competitive existing approaches, achieving a 30% higher precision and recall, a 20% improvement in F1-score, and a 40% increase in AUC, demonstrating its superior effectiveness in CBVR.

The exponential growth of information and communication technologies has led to an unprecedented surge in digital data, with a significant portion comprising unstructured visual media, such as images and videos that often lack metadata. This absence of structured metadata presents a major challenge in efficiently managing and extracting value from these vast repositories, rendering traditional search and retrieval methods ineffective. Content-based video retrieval (CBVR) has emerged as a crucial solution, transforming fields such as traffic analysis, video surveillance, medicine, and sports. Unlike static images, videos consist of objects that continuously move and undergo appearance changes over time, making it challenging to effectively capture both spatial details and temporal dynamics. This paper proposes a method that leverages the strengths of DenseNet-151, a densely connected convolutional neural network designed for extracting rich spatial features, and Long Short-Term Memory (LSTM), a recurrent neural network specialized in capturing temporal dependencies. Additionally, a modified distance function incorporating fuzzy class membership is employed to enhance the retrieval of similar videos from the database, leading to improved retrieval performance. The effectiveness of the proposed method is evaluated on the UCF101 Human Action Recognition dataset, demonstrating a 6

The exponential growth of visual data, driven by advancements in computing, storage, and networking technologies, underscores the need for more precise methods and techniques for efficient retrieval and management. A Content-Based Retrieval (CBR) system facilitates the retrieval of media, including images, videos, and audio, from a store by analyzing their content rather than relying on metadata such as descriptions, titles, or textual annotations. Forged videos are rapidly increasing, undermining trust in video content and posing significant security risks. Various performance metrics are employed to assess the confidence level of CBR systems. Furthermore, to effectively retrieve forged videos corresponding to a given original video, it is essential to assess the relevance of the evaluation metric. In this paper, we introduce a novel evaluation metric for CBR systems, termed the Weighted Distance-Based Score and Normal Precision Pair (WDBSP). This metric leverages distance values as weights to assess retrieval performance while simultaneously measuring precision. WDBSP is particularly valuable in scenarios where multiple CBR systems exhibit identical precision yet differ in actual performance. The proposed method quantifies the retrieval strength, while normal precision captures the density of relevant retrieved materials. To validate the effectiveness and significance of the proposed metric, multiple use cases are designed, enabling a comprehensive comparison with benchmark metric commonly employed at this scale. As a case study, we utilize the REWIND forgery video dataset to retrieve manipulated videos corresponding to an original query. Experimental results demonstrate that the proposed metric provides a more accurate and reliable evaluation of CBR systems, particularly when considering the top-k retrieved media.

With the widespread adoption of high-speed networks such as 4G and 5G, along with the explosive growth of social media platforms, video content is now frequently captured and shared online without accompanying metadata such as tags or descriptions. This absence of textual annotations presents a significant challenge for indexing and retrieving relevant video content. Content-Based Video Retrieval systems address this issue by analyzing the visual content of videos rather than relying on external metadata. However, only limited efforts in the literature have jointly explored both the spatial and temporal context of video data for retrieval. To address this gap, we propose a Content-Based Video Retrieval framework that leverages a 3D Convolutional Neural Network, specifically the R(2+1)D architecture enhanced with transfer learning. This model decomposes spatiotemporal convolutions to more effectively capture both spatial and temporal video features. In addition, we introduce a novel classification-similarity-based weighted distance approach, which overcomes the limitations of traditional distance-based and classifier-based retrieval methods. Experimental evaluation on the UCF101 dataset demonstrates that the proposed system achieves a significant improvement in retrieval performance, with over a 20% increase in AUC compared to baseline techniques.

Machine learning (ML), a subset of artificial intelligence (AI), is revolutionizing industries by leveraging statistical algorithms that learn from data and experiences. Unlike traditional programs following predetermined sequences, ML algorithms discern patterns and predict outcomes through extensive datasets. This transformative technology has profoundly impacted diverse sectors, including manufacturing, finance, retail, transportation, entertainment, and healthcare. The influence of ML is amplified by the accessibility of extensive datasets and the escalating computational prowess of modern systems. As ML algorithms progress, they are fundamentally reshaping business operations, streamlining processes, enhancing decision-making, and fuelling innovation across sectors. The impact of machine learning algorithms on healthcare applications and the usage of diverse data sources, such as electronic health records, medical imaging, wearable devices, and genomic data, is discussed in this chapter.

In the digital era, low-cost hardware like sensors and cameras has led to the creation of numerous image databases for various applications. This has led to the need for retrieval systems that rely on visual content, and these types of systems are called content-based image retrieval (CBIR) systems. It’s a method utilized to locate and extract digital images from extensive databases by considering their visual attributes, as opposed to relying exclusively on metadata or written descriptions. In order to obtain appropriate images from the database, features including colour histograms, texture patterns, and shape descriptors are being used to determine similarities between the images. Over the course of the last twenty years, efforts have been directed towards creating hand-crafted features tailored for CBIR systems. However, depending solely on distance-based retrieval methods is a formidable task. Hence, this study strives to leverage the capabilities of classifiers as well for the purpose of retrieval. So, the proposed CBIR paradigm uses not only the hand-crafted features but also the strength of the classifier with weighted distance metricTherefore, the proposed CBIR paradigm is designed in a way that it uses the strength of the NaiveBayes classifier to compute weighted distance using hand-crafted wavelet features to get similar images from the database. The performance of the proposed method is evaluated on three most popular texture datasets and found to be better among all the methods reported in this work.

With rapid growth of social media platforms and widespread use of handheld devices such as mobile phones and video cameras, the number of videos being captured and shared over the internet has increased significantly. However, due to the lack of organization, most of these videos lack semantic context. Traditional methods of video retrieval involve searching for relevant videos using attached semantics. which has led to the need for content-based video retrieval, where video contents are utilized for searching, whether by video or text queries.The primary goal of our system is to provide relevant videos from a database. Our proposed approach in this paper employs Pearson’s coefficient of correlation (PCC) for key frame extraction from videos, subsequently building a feature vector that represents the video’s content. We have also experimented with linear binary pattern (LBP) and Colour moments (CM). We have used precision metric for evaluating performance. For conducting experiments, we utilized the UCF101 dataset, comprising 13,320 videos across 101 categories.

During the COVID-19 pandemic time, educational institutions have really played a good role in imparting online education to students. Their career and academic tenure were not affected as contrary to the past pandemics throughout world history. All this has been possible through long sessions of classes, quizzes, assignments, discussions, chat interactions, and examinations through online video-based learning using computer interactive measures. The students were privileged to utilize digital technologies for longer durations for learning purposes. However, these long stretches have adversely affected their body postures, and physical and mental health as they majorly remain confined to chairs with restricted levels of physical activities. Thus, there is a need to have a model which can act as an insight for parents, doctors (pediatricians), and academic policymakers to decide on maximum hours for online teaching and related activities during future pandemics. The novel model proposed in this work helps to predict the impact of enhanced students’ computer interactions on their physical and mental health. The method proposed uses a novel model which is advanced and computationally strong. The model follows a two-step methodology, where at the first level, a variant of already existing machine learning algorithm is proposed and at the next level, it is optimized further using a hybrid bio-inspired optimization algorithm. The model consists of proposing a variant of XGBoost model (step1 optimization) followed by a hybrid bio-inspired algorithm (step2 optimization). The work considers a humongous dataset with varied age groups of students with more than 10 attributes. The proposed model is highly efficient in making predictions with 98.07% accuracy level and 98.43% F1-score. The time complexity of the model obtained is also of order of “n” where “n” depicts the number of input variables. Strong empirical results for other parameters also like specificity (95.63%) and sensitivity (96.74%) ascertain the enhanced predictive power generated using the proposed model. An extensive comparative study with other machine learning models ascertains the elevated accuracy and predictive power using the proposed model. Till now none of the researchers have proposed any such pioneering tool for parents, doctors, and academicians using advanced machine learning algorithms.

In recent years, medical image transmission using a multi-cloud system has played a significant role in e-Healthcare infrastructure. It allows medical practitioners to easily store, retrieve, and share patients’ medical information across multiple stakeholders. However, multi-cloud image transmission may be vulnerable to multiple security breaches, such as authentication, confidentiality, and security issues. Motivated by these issues, this paper proposes a data-hiding scheme for secure medical image transmission in a multi-cloud environment. The proposed scheme ensures imperceptible robustness and watermark security at a low computational cost. Here, the medical image is divided into a number of shares using Neighbor Mean Interpolation (NMI). To achieve confidentiality, Electronic Patient Healthcare Record (EPHR) is encrypted using Double Scan Pixel Position Shuffling (DSPPS) approach. Then, the encrypted EPHR is divided into shares and embedded in the cover medical image shares. Finally, a minimum of 50% of watermarked image shares are utilized to retrieve the original medical image and encrypted EPHR, consequently reducing multi-cloud latency and computational burden. Experimental results show that the proposed scheme shows high imperceptibility, robustness, and watermark security at a low computational cost. Comparative analysis with some of the recent popular data hiding schemes shows that the proposed scheme has improved imperceptibility and robustness by 10%–15% (approximately) with higher watermark security at a low computational cost.

Image classification stands as a fundamental task in computer vision, and Convolutional Neural Networks (CNNs) have emerged as highly proficient tools, demonstrating remarkable accuracy and performance. However, with the increasing complexity and diversity of image datasets, there is a growing need to improve the robustness and generalization of CNN-based classifiers. One promising approach to address this challenge is the ensembling of CNNs. Ensembling involves combining the outputs of multiple CNNs to enhance classification performance. This technique leverages the strength and diversity of individual models to achieve superior results compared to using a single model alone. Therefore, GLS-NET, an ensemble framework is proposed which uses three parallel ResNet50 CNNs and takes different features as input so as to induce the diversity in data which in turn can learn discriminative features to produce high accuracy. The proposed framework is evaluated on the most popular dataset, EMNIST, and achieved good performance improvement in accuracy. EMNIST is the most popular dataset used extensively in evaluating the performance of many deep learning techniques.

Due to the availability of various photo editing tools, intruders can tamper with an image very easily. So, various watermarking and tamper detection approaches have been proposed by researchers. Basically, tamper detection techniques focus on embedding the watermark, extracting the water mark, and identifying the tampered regions. But it is very important that the tampered pixels should also be corrected. We bring forward an image watermarking technique for tamper detection and correction using remainder value differencing (RVD) and extended Hamming code (EHC). It operates on a pixel group of size 2 x 2. Watermark bits (WBs) are generated from four most significant bits of the pixels in a pixel group by EHC and concealed in four lower bit planes by the principle of RVD. The WBs are extracted at the receiver along with the identification of tampered pixels. The tampered pixels are corrected by the developed correction logic. As the principle of RVD is used, precautions are taken to avoid the fall-off boundary problem. The efficacy of this technique is accessed through various quality metrics. It is noted that it performs better than the existing techniques. The recorded peak signal-to-noise ratio value is 45.49 dB with structural similarity value 0.9889. The tampered pixels are identified and corrected.

Content-Based Remote Sensing Image Retrieval (CBRSIR) is used to find relevant images from large collections of remote sensing images. CBRSIR works by indexing each image in the database with a feature vector. Deep semantic features generated using convolutional neural networks (CNNs) are more powerful than low-level features for CBRSIR tasks because they can comprehend the context and content within an image. However, the major problem with the deep features is its large vector size which in turn can impact the performance of the retrieval system and are more susceptible to noise and outlier data. Therefore, in this work, a modified ResNet50 architecture is proposed that serves as a powerful feature extractor, benefiting from its deep learning capabilities. Specific modifications are introduced to enhance its discriminative power and generalization ability, enabling it to extract more robust deep features for image indexing. The proposed method achieves a mean average precision (mAP) of 0.899 surpassing the popular competing methods ResNet50 and GoogleNet by a substantial margin of 22.02%, 26.79% respectively. Moreover, to address the curse of dimensionality, this study also proposes a novel approach that combines a modified ResNet50 architecture with Linear Discriminant Analysis (LDA) and Maximum Relevance and Minimum Redundancy (MRMR) technique. The proposed approach achieves 85.45% reduction in size of the feature vector using MRMR and 98.19% using LDA, thereby improving retrieval efficiency without impacting the performance.

In this paper, a viable, robust, and highly accurate additive hybrid model employing autoregressive fractionally integrated moving average (ARFIMA) and support vector machine (SVM) with functionally expanded inputs (Additive‐ARFIMA‐SVM) is presented for forecasting the air quality index (AQI). Additionally, thirteen additive and multiplicative hybrid models are introduced. Several alternatives in feature engineering employing functional expansion of inputs are incorporated to boost the performance of hybrid models. Furthermore, a gradient whale optimization algorithm with group best leader strategy (GWOA‐GBL) based meta‐heuristic algorithm is proposed. The missing values are imputed and a variable weight ensemble forecasting model is developed using the proposed GWOA‐GBL algorithm. To evaluate the effectiveness of the proposed Additive‐ARFIMA‐SVM forecasting model with functionally expanded inputs, comparisons are made with sixteen machine learning models, including long short‐term memory (LSTM), five statistical models, seventeen hybrid models, and ten variable weight ensemble models. Extensive statistical analyses are carried out on the obtained results considering four accuracy measures that show the statistical supremacy of the proposed Additive‐ARFIMA‐SVM model and GWOA‐GBL algorithm in predicting the AQI time series. The proposed Additive‐ARFIMA‐SVM model with functionally expanded inputs improves the AQI forecasting performance by 16.34% than autoregressive integrated moving average, 14.47% than ARFIMA, 33.96% than XGBoost, 43.47% than SVM, 49.39% than LSTM, 8.64% than Multiplicative‐ARIMA‐SVM model considering symmetric mean absolute percentage error. The proposed Additive‐ARFIMA‐SVM model is so efficient and reliable that it can be applied to forecast other time series like stock price, electricity load, crude oil price, sunspot number, stream flow, flood, drought etc.

Radio Frequency Identification (RFID) is primarily used to resolve the problems of taking care of the majority of nodes perceived and tracking tags related to the items. Utilizing contactless radio frequency identification data can be communicated distantly using electromagnetic fields. In this paper, the comparison and analysis made between the Clustered RFID with existing protocols Ad hoc On-demand Multicast Distance Vector Secure Adjacent Position Trust Verification (AOMDV_SAPTV) and Optimal Distance-Based Clustering (ODBC) protocols based on the network attributes of accuracy, vulnerability and success rate, delay and throughput while handling the huge nodes of communication. In the RFID Network, the clustering mechanism was implemented to enhance the performance of the network when scaling nodes. Multicast routing was used to handle the large number of nodes involved in the transmission of particular network communication. While scaling up the network, existing methods may be compromised with their efficiency. However, the Clustered RFID method will give better performance without compromising efficiency. Here, Clustered RFID gives 93% performance, AOMDV_SAPTV can achieve 79%, and ODBC can reach 85% of performance. Clustered RFID gives 14% better performance than AOMDV_SAPTV and 8% better performance than ODBC for handling a huge range of nodes.