Gujarat Matikam Kalakari & Rural Technology Institute

Abstract— This paper presents a machine learning-based system for offline signature verification, aimed at distinguishing between genuine and forged handwritten signatures. The proposed system utilizes a Siamese neural network architecture to extract and compare feature embeddings from two input signature images: a reference (genuine) signature and a query (test) signature. By computing the Euclidean distance between the feature vectors, the model determines the degree of similarity between the two signatures. A predefined threshold value of 0.2 is employed to classify the query signature as either genuine or forged. The system is developed using Python, leveraging TensorFlow for model training and inference, and Streamlit to provide a user-friendly web-based interface for real-time signature verification. Experimental results demonstrate the model's effectiveness in achieving reliable verification performance, highlighting its potential applications in domains such as document authentication, financial verification. transactions, and identity Varification. Keywords : Signature verification, Siamese neural network, Machine learning, Offline handwriting recognition, Biometric authentication, Euclidean distance, Deep learning, TensorFlow, Image similarity, Streamlit.

Fabric plays a significant role in our daily life. The physical characteristics and look of the cloth are determined by its weave in addition to its color and texture. Sometimes it is important to identify the fabric’s weave kind. Additionally, it could be necessary for the end user to verify the type of fabric weave. To identify the fabric’s weave type, a manual invasive technique is typically employed, which also necessitates a thorough understanding of the fabric weaving process. This article proposes an automatic recognition method that uses pixel-level information to identify color, weave, and density patterns which is a low-cost, non-invasive, image-processing-based method. Good quality camera is used to take the pictures of the fabric. First, the skew of the warp and weft threads is corrected using a technique based on the Hough transform. Second, projection of a brightness technique is used to identify the yarns and nodes in the improved images. The results of proposed method is used to compute the density. Proposed technique applied for three different yard dyed woven fabrics. In this method, warp density error is around 1%. Next, using the boundary information, each node’s type is determined. Knowing the kind of each node allows us to obtain the weaving pattern.

Large Language Models (LLMs) have demonstrated remarkable capabilities and are increasingly integrated into real-world applications. Yet, because they are trained on vast amounts of raw data, these models often inherit and amplify social biases, which can influence downstream tasks in problematic ways. Such unfair behavior raises ethical concerns and risks harmful social consequences. This paper offers a thorough review of research addressing fairness in LLMs. To reflect how model scale and training approaches shape research directions, we categorize prior work into two streams: studies on medium-sized LLMs within pre-training and fine-tuning settings, and studies on large-scale LLMs within prompting-based settings. For medium-sized models, we summarize evaluation metrics and debiasing strategies by distinguishing between intrinsic and extrinsic bias. For large-scale models, we highlight recent advances in fairness assessment, explore the underlying causes of bias, and examine approaches to mitigation. We conclude by discussing the key challenges that remain and propose potential directions for advancing fairness in the development and use of LLMs.

Malware analysis is crucial part of cybersecurity that is essential to researchers in order to completely understand malicious software behaviour and come up with methods that can discover it. Despite file structures and API integrations being checked thoroughly through static analysis without execution, it is always missing dynamic actions. Despite its monitoring real operations, dynamic analysis is both resource intensive and susceptible to analysis prevention techniques. This study thoroughly uses both methods on eight real samples from six malware families to discover malicious patterns in how they act. All our finding underline why hybrid analysis should be done for powerful malware detection. Malware familial classification also requires hybrid analysis. Our analysis establishes significant differences between static signatures and dynamic behavior and demonstrates that hybrid approaches are required for accurate malware classification.

Abstract—Alcohol-impaired driving and neglect of helmet usage are major contributors to road accidents and fatalities. To solve these problems, this study presents the creation of a Smart Helmet system integrating alcohol detection, helmet- wear detection, and an LED display, utilizing Arduino technol- ogy. An MQ-3 sensor installed within the helmet monitors the rider’s breath for alcohol content. If the alcohol concentration exceeds a predefined threshold, the system displays immediate warning messages on the LED screen. Additionally, a helmet- wear detection mechanism confirms that the system only activates fully when the s a f e t y g e a r is properly worn, alerting the rider if the helmet is not in place. This dual-safety approach aims to promote responsible behavior without requiring modifications to the E n g i n e s t a r t system. The design emphasizes real- time monitoring, cost-effectiveness, and user- friendly operation. Experimental results validate the reliability and responsiveness of the system, demonstrating its potential to enhance rider safety and significantly reduce alcohol-related and helmet-related road accidents. Index Terms—Smart Helmet, Alcohol Detection, Helmet-Wear Detection, Arduino, Road Safety, LED Display.

The current study assesses several water-based PVT system thermal absorber configurations. The thermal absorber in PVT system plays a vital role in efficiency evaluation as it lowers PV temperature and collects heat energy. The current study aims to discover and analyze advanced thermal absorber design by comparing well-received spiral circular absorbers and non-cooled PV with proposed semi-circular thermal absorbers with varying flow configurations. The proposed thermal absorber maintains surface contact with PV panels and improves heat transfer thereby yielding better thermal and electrical efficiency. Simulated PVT systems have a constant water flow rate and solar radiation. The CFD-FLUENT software was preferred to evaluate the PVT system in steady-state conditions for the investigation. Under constant ambient and inlet water temperatures of 299 K, the PV temperatures at the surface, water discharge temperature, and pressure drop were measured. It was discovered that a thermal absorber could effectively lower PV surface temperature by cooling. A zigzag thermal absorber was the most efficient since it produced the highest water outlet temperature and lowest PV surface temperature while also slightly raising the pressure drop. In comparison with a non-cooled PV system, a zigzag thermal absorber PVT system yields 11.97% more electrical efficiency, with an addition of 76.75% thermal efficiency. It was also noticed that a conventional spiral circular PVT system provides 13.5% electrical efficiency and 54.8% thermal efficiency while an electrical efficiency of 13.61% and thermal efficiency is 76.75% was obtained from a zigzag thermal absorber PVT system. The zigzag thermal absorber PVT system had a high initial investment of INR 38809.00. It showed a simple payback of 4.63 years, a 28% return on investment with a promising 2.1 Debt Service Coverage Ratio. It is advisable to consider incorporating zigzag semi-circular PVT in the prospective improvements of the PVT system.