Authors - Nandinee Mudegol, Abhijeet Urunkar, Vedika Dhende, Vaishnavi Katkar, Chetna Ghengare, Samiksha Harer Abstract - Agriculture is a very important sector, but the farmers are facing problems in early identification of plant diseases and monitoring crop health. Manual checking of crops is time-consuming and can lead to late detection of diseases, which lowers the yield of the crop. To address this problem, authors have pro-posed a system called Agridiagnosis. The proposed system combines two major features: plant disease detection using image processing and crop health monitoring using NDVI. Farmers can upload images of leaves to be detected and suggested treatment. At the same time, the system provides a color-coded map of crop health with respect to NDVI values. The combination of both features in a single platform allows the system to help farmers easily comprehend the crop conditions and take timely measures to increase productivity.
Authors - Romildo Silva, Maria Tavares, Filipa Silva, Carlos Lopes Abstract - This paper investigates the use of AI agents as consumers of tokenized real-world asset (RWA) data in financial environments. A Python-based agent was developed to automatically retrieve, process, and analyze financial information from publicly accessible APIs for selected traditional and tokenized assets, including SPY, QQQ, PAXG, and ONDO. The proposed framework evaluates data quality through quantitative metrics such as latency, completeness, null rate, and data volume, complemented by descriptive statistical analysis and Shapiro-Wilk normality testing. The results indicate that traditional financial assets exhibit higher informational stability, while tokenized assets present greater variability and non-normal behavior. The study demonstrates that autonomous AI systems can effectively consume heterogeneous financial data sources and highlights the growing importance of information quality and consistency in AIdriven financial ecosystems.
Authors - Anupama Y K., G M Trupti, Arun Kumar N Abstract - Due to the rapid evolution of cyber threats with the growth of the internet and cyber threats, we now live within the cyber domain, which is under great pressure and strain from cyber threats. The rise in popularity of Cloud Infrastructure, which offers customers scalable data storage, has led to development of new vulnerabilities to business owners, as well as a growing shift in the way hackers operate. Traditional methods of detecting cyber attacks, such as IDSs, typically encounter issues when dealing with a large amount of class imbalance and have difficulty adapting to newer attack vectors. This results in an increase in false positives that companies receive when monitoring their systems for cyber attacks, as well as a decreasing ability to detect less frequent, but very high-impact, types of cybersecurity threats. The solution involves developing an AI-based intrusion detection system that combines the use of Borderline SMOTE to balance the classes incorrectly identified, with an ensemble method called maximum vote that combines three classifications methods: Decision Trees, XGBoost and tuned AdaBoost. The system is evaluated using the KDD Cup 1999 benchmark dataset which contains regular traffic as well as different attack types including DoS/DDoS (Neptune, Smurf, Teardrop), Probe (Nmap, Ipsweep, Portsweep, Satan), R2L (Guess password, Back) and U2R (Buffer Overflow, Rootkit, Land). Experimental results show that the max-voting ensemble out performs the individual base models (Decision Tree, AdaBoost, and XGBoost) and standard single classifier IDS methods along with baseline algorithms. This leads to more reliable detection of both minor and major attacks in cloud security scenarios. These findings highlight the effectiveness of combining Borderline-SMOTE with ensemble learning to build a scalable and robust IDS suitable for real-time cloud security monitoring.
Authors - Y. Zamarripa-Rivera, S. Villagrana-Barraza, D.I. Ortiz-Esquivel, L.E. Banuelos-Garcia, M. Molina-Almaraz, G. Díaz-Florez Abstract - Plant propagation by cuttings requires controlled microclimatic conditions to promote rooting, reduce water stress, and improve process reproducibility. This paper presents the design, implementation, and functional validation of an ESP32based monitoring and control platform for a plant cutting propagation chamber. The system integrates multi-zone sensing, ON/OFF-based control with PWMassisted thermal actuation, local CSV data logging, Wi-Fi communication, and web-based supervision. Temperature and relative humidity were monitored in the external environment, stem zone, and root zone, while light intensity, water flow, actuator states, and PWM commands were recorded as operational variables. Validation was conducted through two 15-day experimental campaigns, generating more than 40,000 time-stamped environmental and operational records. The platform maintained differentiated microclimatic conditions, with average rootzone temperatures between 23.79 and 24.31 °C and root-zone relative humidity between 75.74 and 80.32%. Biological validation with six basil cuttings achieved an overall rooting success rate of 83.3%, reaching 100% in the second campaign. These results demonstrate the potential of low-cost ESP32-based systems for controlled-environment agriculture and smart propagation applications.
Authors - Olivier ZONGO, SOMDA Dekpeltakié Augustin METOUALE, Mamadou DIARRA, Abdoulaye SERE Abstract - Automatic assessment of obstetric ultrasound remains a challenge due to its operator-dependent nature and the dynamic context of fetal labor. This study proposes a Temporal Quality Gate Framework to standardize diagnostic plane validation using a novel Temporal Attention-Gated LSTM (TA-LSTM) architecture. We formulate the task as a binary classification problem to distinguish standard diagnostic planes from non-diagnostic sequences, using the IUGC 2024 dataset of 434 transperineal ultrasound videos (266 positive, 168 negative). The TA-LSTM extracts spatial features via a ResNet-18 backbone and dynamically weights temporal dependencies using an attention mechanism. Under 5-fold cross-validation with strict patient-level splitting, the TA-LSTM achieves a mean AUC-ROC of 0.989 ± 0.008 and a mean test accuracy of 95.63% ± 2.85% (peak accuracy of 98.85%) with an inference latency of 14.2 ms on GPU. Our framework acts as a robust Quality Gate, ensuring that subsequent automated measurements, like the Angle of Progression (AoP), are performed on high-quality validated inputs, making it highly suitable for resource-limited clinical environments.
Authors - S. Zouini, A. Meddaoui, A. Jrifi Abstract - Statistical Process Control (SPC) is a well-established methodology for industrial quality management. The growing complexity of modern manufacturing environments — driven by Industry 4.0, high-dimensional sensor data, and nonlinear process dynamics — exposes the limits of classical monitoring approaches based on fixed thresholds and Gaussian assumptions. This paper proposes an AI-Driven Statistical Process Control (AI-SPC) framework that integrates PCA-based Hotelling’s T2 monitoring with a Random Forest classifier within a closed-loop architecture. The framework is evaluated on five fault scenarios from the Tennessee Eastman Process (TEP) benchmark. Results show that the hybrid AND-logic strategy achieves a false alarm rate of 0.071 — a 45% reduction relative to PCA-T2 alone (0.130) — while maintaining a detection rate of 96.1% and a detection delay of 5.4 samples. A variable contribution analysis further supports fault diagnosis by identifying the most deviant process variables at the moment of detection. These results confirm that combining statistical rigor with data-driven flexibility produces a more reliable and interpretable monitoring system than either approach deployed independently.
