Forest fires are now increasing in recent years with the impact of climate change. Early detection of the smoke is of prime importance for preventing the fast growth of the fire. New advancements in the YOLO-based models for the object detection of YOLOv7 to YOLOv12 models have significantly enhanced the real-time detection of the objects. However, YOLOv12 exhibits limited sensitivity to remote and small-scale fire and smoke regions, as it does not fully capture high-resolution spatial details. To resolve this problem this paper introduces F2-YOLO signifies Fire and Fume-detection, a lightweight and version of YOLOv12, specifically designed for early detection of forest fire and smoke. In this improved network, F2-YOLO integrates three key parts: a Lightweight Deep Adaptive Module (LDAM) responsible for efficient feature sharpening, a Multi-Level Feature Fusion (MFF) module for enhancing features related to small objects, and a Small-Object Detection Head (SODH) for superior localization accuracy. When tested using the ADSF and Wildfire datasets, F2-YOLO is shown to have a mAP@50 of 98.5% and mAP@50-95 of 73.92%, which is greater than other YOLO models.

Smart agriculture relies heavily on visual data captured using low-cost mobile devices in rural environments. Farmers often face challenges in identifying crop diseases, pest attacks, and soil conditions due to the lack of expert assistance and limited access to reliable information. Image retrieval systems can support decision-making by retrieving visually and semantically similar agricultural images from a reference database. However, traditional classification-based approaches are not well suited for retrieval tasks and often require large computational resources. This paper introduces a deep metric learning-based image retrieval framework for smart agriculture applications, optimized for resource-constrained rural environments. A light convolutional neural network is used for obtaining discriminative embeddings, and a triplet loss function is used to learn an embedding space that is aware of similarities. The proposed approach is validated on agricultural image datasets, with comparisons to existing CNN models for image retrieval tasks in the literature. The results have shown that the proposed approach can highly enhance the retrieval accuracy with lower computational complexity, meeting the requirements for practical rural applications.

Early prediction of the Alzheimer disease (AD) is important because it prevents detection of the disease in its early stages that are best addressed by clinical intervention and management of the disease in its early stages, and since the cognitive symptoms overlap and are non-linear in nature. The current research indicates a hybrid classifier paradigm of earlystage prediction of Alzheimer disease by secondary data of cognitive assessment by voting. The methodology is quantitative and experimental in nature and includes data preprocessing, feature selection and combination of several base classifiers, these include Decision Tree, Support Vector Machine, k-Nearest Neighbour and Logistic Regression. Voting strategies used to combine predictions of individual classifiers include majority, weighted and soft voting to increase the accuracy of the classification. The findings of the experiment show that the voting-based hybrid model and weighted voting strategy, in particular, are more accurate, precise, and recall better and might be more robust than individual classifiers. The analysis of feature contribution leads to a focus on memory and executive functioning as the most critical areas of cognition, and the analysis of confusion matrixes proves that there is no significant misclassification of memory and cognitive stages of close closeness. Generally, the results confirm the usefulness of ensemble learning in the initial stage of the Alzheimer disease and indicate that the suggested hybrid model can be consistent and clinically useful decision-making model in cognitive assessment-based diagnosis.

The blistering development of the concept of smart cities has escalated the use of Internet of Things (IoT) gadgets to monitor the urban facilities in real time, and this produces massive amounts of data that compromise the conventional cloud-based structures in terms of high latency, high bandwidth occupancy, and high energy consumption. This paper presents and analytically judges an edge-enhanced IoT architecture that will be used to improve energy efficiency and real-time performance of smart city monitoring systems. The study adopts a quantitative, model-based study that is based on all secondary data and analyzes a three-layer IoT-edge-cloud system and evaluates system performance based on end-to-end latency, energy usage, bandwidth usage, computational load allocation, and data reduction efficiency. The comparative assessment and analytical modelling show that edge-level data processing is a significant way to cut down processing and transmission time, decrease overall energy usage despite extra use of edge-layers, reduce bandwidth requirements by data reduction, and push out the computation load off the centralized cloud servers. The results show that the edge assisted IoT architectures are more scalable, responsive and energy-efficient than cloud-centric systems and therefore provide a robust and sustainable solution to large-scale real-time urban city monitoring and future urban digital transformation.

Multiple input multiple output-orthogonal frequency division multiplexing (MIMO-OFDM) communication, allowing high data rates and low latency, is a crucial enabler for 5G and beyond wireless networks. However, its sensitivity to eavesdropping and physical-layer attacks poses serious security challenges. This research proposes a novel framework that enhances physical layer security in MIMO by integrating deep learning–based key generation and federated learning–assisted secure transmission. Capitalizing on the spatial uniqueness of mmWave propagation, we employ a cross-layer mobilenetV2 pyramid mutual attention network to extract rich physical-layer features—such as angle of arrival and angle of departure, for robust key generation without requiring prior key exchange. To enhance the security of the transmission process, we introduce a multichannel squeeze-and-excitation combined network trained in a federated learning setting, allowing distributed devices to jointly learn a secure transmission model without compromising data privacy. Experimental results demonstrate that the proposed method significantly enhances resistance to passive eavesdropping and ensures low bit error rates, even under dynamic channel conditions and high secrecy rate, key generation rate of about 1.63 at 40 dB signal-to-noise ratio and bit error rate of 0.1 at 25 dB. The proposed approach enables lightweight, scalable, and privacy-preserving security in mmWave-enabled communication systems.

The importance of obtaining details regarding transportation vehicles has been increasing in developing countries such as India. Object detection plays an important role in the automatic identification of the content of an image or video without human intervention. Various deep learning models have been developed for object detection using CNNs (convolution neural networks). This paper proposes a method for vehicle detection from still images using an optimized YOLO-NAS (You Only Look Once-Neural Architecture Search) frame work. This model is verified with earlier YOLO models for improved accuracy and optimization. The experiments were conducted on two publicly available datasets. Indian Driving Dataset (IDD) and DATS_2022 having exclusively images of various traffic scenes on Indian roads. The proposed method out performs the existing object detection models in terms of detection accuracy. Results show that the proposed method is good at detection accuracy measured in Average Precision and Recall.

For a precise diagnosis and treatment plan to be planned in a clinical setting, brain tumour categorization from magnetic resonance imaging (MRI) data is essential. Clinicians can efficiently monitor the course of the disease, select the best course of treatment, and gauge the effectiveness of that course of treatment with the help of fast and accurate tumour classification. Additionally, machine learning models can help radiologists make better diagnoses by lowering interpretation errors and enhancing the quality of MRI image interpretation. Classifying brain tumours also aids in research endeavours to discover biomarkers, comprehend the biology of tumours, and create tailored treatments for various tumour subtypes. Current techniques for classifying tumours in the brain from MRI data frequently rely on labour-intensive, inconsistent manual segmentation and feature extraction. These techniques might not be able to accurately classify tumours due to minute variations in their morphology or texture. Furthermore, the accessibility and scalability of traditional procedures in clinical settings may be limited due to the need for expertise in radiology and medical imaging. Furthermore, manual feature engineering could miss significant tumour traits or not fully utilize MRI data for categorization. To overcome the shortcomings of current approaches, the suggested system makes use of machine learning techniques to improve and automate the classification of brain tumours using MRI image data. In order to extract discriminative features directly from MRI scans, this work uses machine learning algorithms. The proposed models are capable of reliably classifying brain cancers into important categories and effectively differentiating between different types of tumours by training them on large-scale MRI datasets labelled with tumour labels.

Sentiment analysis is an area of study that is extensively researched. It is concerned with the process of recognizing and classifying opinions that are conveyed in text into one of three categories: positive, negative, or neutral polarities. One of the most important steps in improving the accuracy of sentiment classification is feature selection, which is a process that helps reduce dimensionality, eliminate features that are irrelevant, and improve the efficiency of the model. This study explores the performance of a variety of feature selection methods for sentiment analysis, with a special emphasis on hybrid wrapper-based approaches. In order to create a feature vocabulary that is as efficient as possible, SVM- Recursive Feature Elimination (RFE) is followed by Sequential Feature Selection (SFS/SBS) in this particular case. The efficacy of these selected characteristics is evaluated with the use of a number of machine learning classifiers, such as Support Vector Machines (SVM), Naive Bayes (NB), K-Nearest Neighbors (KNN), Logistic Regression (LR), and Decision Tree (DT). In addition, in order to further improve the classification performance, ensemble learning approaches, including as Random Forests, Extra-Trees, XGBoost, AdaBoost, and stacking, are used. The findings of the experiment show that the hybrid feature selection strategy, when used in conjunction with ensemble methods, leads to a considerable improvement in sentiment classification accuracy. As a result, a solid foundation for opinion mining in text-based data is provided.

Automatic number plate recognition (ANPR) plays a major role in real life applications and several techniques have been proposed. The localization or detection of the number plate of the vehicle images is the basis for any ANPR system. This paper proposes a robust method for localization of number plates in different conditions. There are two stages; first the preprocessing of the input image is performed and then localization is done. After preprocessing the statistical measures such as root mean square error and peak signal to noise ratio are calculated. Next the localization is done using symmetric wavelets and mathematical morphology. Experimental results show that this method gives dominant values of RMSE and PSNR. Experiments were performed on a database and also on a sample of 280 images of different countries taken from various scenes and conditions; results show that success rate of 77.14% on database and 92.14% on sample images achieved.