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.

The early detection of colorectal cancer (CRC) is critical and yet persistently challenging objective for clinical practice. This paper introduces a robust transformer based framework that integrates Deep learning (DL) models (Convolutional Neural Networks: ADaRDEV2I-22 and ResNet-152), Vision Transformers (ViTs), and XGBoost to enhance CRC classification from medical images. The DL model performs high-quality initial feature extraction, while ViTs effectively model long-range spatial dependencies often missed by traditional architectures. XGBoost is employed for final structured data classification. This helps in boosting diagnostic precision and robustness. In order to address the imbalance nature of the dataset, an improved generalization, synthetic data generation has been employed using Generative Adversarial Networks (GANs) and standard augmentation techniques. Under 5-fold stratified cross-validation (the primary evaluation protocol), the framework achieves an accuracy of 95.8 ± 0.7%, precision of 94.8 ± 0.9%, recall of 95.1 ± 0.8%, and F1 score of 94.9 ± 0.7%, surpassing standalone DL models and ViT models by 4.1% - 5.2 percentage points over the best baseline (90.6%). The data augmentation techniques led to a 7.3% gain in recall for underrepresented classes. Furthermore, model interpretability was increased through SHAP, and LIME. Despite its complexity, the framework maintains computational efficiency, with only a 12% increase in training time compared to single-model systems. The proposed methodology offers an interpretable, scalable and high-performance solution for DL based transformer based colorectal cancer detection, supporting its integration into real-world clinical decision support systems.

Large language models (LLMs) have been enormously successful in natural language understanding and generation yet their resistance to adversarial manipulation is a significant area of concern. This article is comprised of the methodological empirical study of the LLM strength in the context of the contextual adversarial attacks, which exploit the semantic dependencies and promote framing without altering the syntactic structure. There were four contextual adversarial methods that were created and applied to a range of language understanding and reasoning tasks, which included misleading instructions, semantic obfuscation, contextual distraction and role manipulation. The performance degradation, consistency of responses and the qualitative error severity were the measures of the robustness. The results depicted that there was a significant loss of accuracy and instability of answers in the adversarial environments, and there was the gravest impairment in the attacks of the misleading guidance and role manipulation. Hallucinations and logical inconsistencies were determined to be the significant failure modes, a fact that suggested an inherent defect in contextual reasoning. The findings reveal that the high baseline performance is not an indicator of robustness and introduces the significance of the robustness-aware assessment systems and specially designed defense strategies to ensure the consistent functioning of the LLMs in the real-life setting.

This study presents a comprehensive comparative research of the significant blockchain consensus algorithms, such as Proof of Work, Proof of Stake, Delegated Proof of Stake, Practical Byzantine Fault Tolerance, Proof of Authority, and hybrid mechanisms, and the purpose of it is to assess the performance, safety and applicability of blockchain to the modern use of blockchain. The studies analyze the key metrics, including throughput, latency, resource consumption, finality behavior, and fault tolerance in various network conditions based on a quantitative, simulation-based methodology with the help of secondary datasets. The results indicate that there are high levels of performance differences among consensus algorithms, as permissioned and delegated algorithms show better efficiency, and low latency, whereas public mechanisms put more emphasis on decentralization, disregarding speed and energy efficiency. The paper identifies the trade-offs inherent to consensus design and points out that no one mechanism is optimal, instead it needs to be chosen based on applicationspecific factors to do with scalability, trust, security, and decentralization. These insights help gain better insight into the issue of consensus behavior and make future choices regarding building the blockchain system.

Feature descriptors play a crucial role in image retrieval by representing images in a compact and discriminative manner. This survey paper explores various techniques for preparing feature descriptors used in image retrieval systems. We categorize the approaches into traditional handcrafted descriptors and deep learning-based descriptors. For each category, we discuss popular techniques, their underlying principles, and their advantages and limitations. We also provide insights into recent advancements and benchmark datasets used for evaluation. In order to help academics and practitioners choose the best strategies for their applications, this survey attempts to give them a thorough grasp of feature descriptor preparation methods in image retrieval.

Online Signature Verification (OSV) systems are crucial for secure authentication in digital environments, where handwritten signatures are electronically captured and verified. These systems use machine learning and deep learning methods to analyze unique dynamic features of a signature, such as pen pressure, speed, and stroke order, in addition to its static shape. Recent approaches leverage techniques such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Generative Adversarial Networks (GANs) and hybrid models, to analyze the unique dynamic and static features of signatures. However, there exists a problem of designing real-time systems that are responsive, and have reliable performance using the mentioned deep learning techniques because these models have to be trained each time a new user is added to the database. The objective of our experiment is to build an OSV system that integrates a CNN-based Siamese Network and a webpage created using ReactJS to allow the user to upload their signature or store it in the database. The model extracts spatial features from the signatures and makes decisions based on the similarity of the uploaded signature with the original signature of the user. The system is designed to handle real and forged signatures through continuous model training. Using a combination of signature preprocessing techniques, feature extraction, and classification models, this system aims to ensure a robust, reliable, and secure method for identity verification in online transactions and sensitive digital applications,

Document Clustering is the process of collecting similar kind of documents into one group based on any particular similarity function. Document clustering is also referred as text clustering. Informative features like phrases and their weights are considered to be more important to perform efficient document clustering. This paper mainly deals on two key parts for achieving efficient document clustering. The first part is a phrase based document model named as the Document Adjacency List, it explains about the construction of a phrase based model of the document set. It produces efficient phrase matching which is useful to decide the similarity among the documents. The second part is the document clustering algorithm that is proposed to enhance the Document Adjacency List for clustering based on the similarity measure. The combination of the above two parts leads to better calculation of similarity among documents and similarity further helps to calculate document clustering.