A Comparative Review of Deep Learning Methods for Landmine Detection from Vision Transformers to Ground-Penetrating Radar
Landmines present a real problem around the world, causing injuries and a real threat to deminers’ lives. Due to that, the search for safe and efficient demining methods has been a humanitarian priority for a long time. Traditional methods rely on physical probing and manual demining. However, these methods give high false-alarm rates, risks, costs, and are time-consuming. Thus, there is an urgent need to find innovative technologies and study their potential to give 100% efficient solution. This paper provides a comparative review of recent research in landmine detection and classification, focusing on the application of Artificial Intelligence and Deep Learning. We will evaluate the number of recently used deep learning methodologies, datasets and achieved performance results. This will help identify how Artificial Intelligence can succeed in solving the problems of demining operations. In this study, we will analyse the use of different algorithms with the overall goal of specifying the best findings to ensure high territory clearance, as well as specifying the challenges. This review provides important insights into the current state of the field, highlighting solutions that can enhance demining operations and improve detection accuracy.
Published by: Leidi M.Saleh Aouto
Author: Leidi M.Saleh Aouto
Paper ID: V12I3-1180
Paper Status: published
Published: May 21, 2026
Comparative Evaluation of Machine Learning Approaches for Physiological Stress Detection Using ECG and EEG Signal Modalities
Stress is a complicated physiological and psychological phenomenon that has a huge impact on the health and performance of people. Conventional methods of measuring stress are dependent upon self-reports, making it difficult to determine if they are truthful and cannot be noted in real-time. In this study, the use of machine learning techniques to identify stress objectively based on physiological signals taken from ECG and EEG was examined. While other studies have only examined one signal type (either ECG or EEG), this study compared the two modalities side-by-side under the same experimental conditions, using equal amounts of data from publicly available datasets. Multiple machine learning models (Logistic Regression, Support Vector Machine, Random Forest, XGBoost, LSTM) were compared and contrasted using available datasets. Physiological signal features (heart rate variability, electrodermal activity) were taken into account and analysed to understand both autonomic and neural activation due to stress. The results demonstrated that the Random Forest model yielded the highest level of performance (F1-score=.86, AUC=.90), indicating that Random Forest is better suited to handling complex physiological signals than any other machine learning model. An analysis of the physiological signals indicated that stress causes a decrease in heart rate variability, an increase in skin conductance, and an increase in cardiovascular activity (during the physical response) due to increased sympathetic nervous system stimulation. In summary, this research points to machine learning-based techniques yielding dependable, non-invasive means for detecting stress. This research indicates that there is an opportunity to use physiologic signal analysis in combination with AI to provide future possibilities for monitoring mental health and wearable technology.
Published by: Kevin Shah
Author: Kevin Shah
Paper ID: V12I3-1185
Paper Status: published
Published: May 21, 2026
IoT-Enabled Smart Parking and Blind Spot Detection
Accelerating urbanisation, combined with a sharp rise in privately owned automobiles, has placed considerable pressure on city infrastructures, particularly with respect to parking management and vehicular safety. Locating an available parking slot consumes significant driver time, indirectly amplifying fuel usage, traffic density, and urban air pollution. Simultaneously, limited driver visibility around a vehicle—especially in blind-spot regions—continues to be a leading contributor to low-speed accidents during parking manoeuvres. This paper offers a thorough review of prevailing technological frameworks, with particular emphasis on deploying ESP32 microcontrollers alongside arrays of heterogeneous sensors to achieve real-time proximity monitoring and obstacle awareness. The study traces the evolution from conventional camera-only solutions toward affordable, omnidirectional IoT architectures designed to strengthen situational awareness comprehensively. Key findings underscore the value of ultrasonic sensing paired with cloud-connected dashboards and smartphone interfaces in building a safer, fully networked vehicular ecosystem.
Published by: Rahul G P, Prithviraj R, KS Maneesh Kumar, Manvith DO, Raghuveer C M
Author: Rahul G P
Paper ID: V12I3-1182
Paper Status: published
Published: May 20, 2026
Review of Heart-Risk Monitoring System
Cardiovascular diseases continue to be among the major causes of death globally, hence making early diagnosis and monitoring crucial to enhance the quality of care. Systems that utilise electrocardiogram (ECG) signals together with artificial intelligence,e such as machine learning, deep learning, and the Internet of Things (IoT), together with wearable health devices, have revolutionised cardiac diagnostics in the contemporary age. In this literature review, there will be an extensive evaluation of new developments in ECG signal processing and arrhythmia detection techniques, wearable ECG monitors, and intelligent health applications. This work assesses different machine learning algorithms that include SVM, CNN, LSTM, MLP and hybrid deep learning algorithms that can be applied to classify ECG signals. Other areas that are covered include remote IoT healthcare systems, cloud computing based on ECG monitoring, explainable artificial intelligence models, FHIR interoperability standards and others. The strengths, limitations, data sets, pre-processing techniques, and results achieved by recent studies are reviewed.
Published by: D R Vishal, Harshith Kumar K M, Jashwanth S R, Bipin Babu R, Harshada J Patil
Author: D R Vishal
Paper ID: V12I3-1177
Paper Status: published
Published: May 18, 2026
GovAI – Smart Government Scheme & Exam Finder Using Intelligent Eligibility Filtering
The increasing number of government welfare schemes and competitive examinations in India has made it difficult for citizens to identify opportunities suitable for their eligibility. Most users face challenges due to scattered information sources, a lack of awareness, and complex eligibility conditions. To solve this problem, the proposed project “GovAI – Smart Government Scheme & Exam Finder” provides a web-based recommendation platform that suggests suitable government schemes and competitive examinations based on user details. The system collects information such as age, income, gender, occupation, educational qualification, and state from users. Using eligibility-based filtering logic, the application recommends relevant schemes and examinations along with application links. The system is developed using Python, Flask, HTML, and CSS, and deployed online using GitHub and Render. The proposed platform reduces manual searching effort, improves accessibility, and provides a centralised solution for personalised recommendations. The project also demonstrates the practical use of intelligent filtering systems and modern web technologies in improving public service accessibility.
Published by: Mohammed Zakhwan, Dr. G. Sharmila Sujatha
Author: Mohammed Zakhwan
Paper ID: V12I3-1167
Paper Status: published
Published: May 18, 2026
Scalable Quality-Aware Depth Map Generation Using Edge-Conditioned Deep Learning Priors
While monocular depth estimation remains a primary hurdle in computer vision, this research presents a sophisticated hybrid framework designed to extract high-fidelity depth information from static 2D images. The core of this methodology lies in its dual-stream architecture: it synchronizes a global depth hypothesis generated via Deep Learning with a localized, edge-sensitive segmentation strategy. To ensure the system remains versatile across a spectrum of hardware from high-performance servers to resource-constrained mobile devices, this work implements a quality-scalable block partitioning scheme. By discretizing the image into adjustable blocks, the system can dynamically balance computational overhead against spatial precision. This process is deeply informed by the luminance channel's edge probability, which acts as a structural guide to ensure that depth transitions are mathematically anchored to the actual physical boundaries of objects. To bridge the gap between discrete block processing and a continuous, natural depth field, a guided bilateral filter is employed in the final stage. This specific refinement serves two purposes: it effectively dissolves 'staircase' or blocky artifacts resulting from the segmentation, while simultaneously acting as a 'boundary-lock' to preserve the crispness of foreground silhouettes. The resulting depth maps exhibit a granular level of detail, particularly at high-resolution block settings—providing the necessary structural accuracy for seamless 3D conversion, cinematic depth-of-field effects, and high-immersion Augmented Reality (AR) environments. GENERATION USING EDGE-CONDITIONED DEEP LEARNING PRIORS
Published by: Ramola Joy P, Remya Madhavan U
Author: Ramola Joy P
Paper ID: V12I3-1168
Paper Status: published
Published: May 18, 2026
