Authors: Sangani Harshil, Kalariya Meet, Baraiya Ravi, Vasani Bhumil, Dr. Vikram B.Kaushik

Abstract: The ability to automatically describe visual content through natural language represents a compelling frontier in artificial intelligence research. Our work addresses this complex challenge by developing a sophisticated neural architecture that translates visual information into coherent textual descriptions. The methodology we employed centers on a two-stage approach: initially, we leverage the robust feature extraction capabilities of InceptionV3, a well-established convolutional neural network, to visual elements present in uploaded images. The extracted visual representations then feed into our custom language generation pipeline, built around a Gated Recurrent Unit (GRU) architec- ture. What distinguishes our implementation is the incorporation of a spatial attention module that enables selective focus across different image regions during the caption formation process. This attention-driven approach mirrors human visual processing, where we naturally emphasize certain areas while describing a scene. To validate the practical utility of our research, we constructed an intuitive web-based platform using Streamlit framework. This interactive system allows users to seamlessly up- load photographs and receive instantaneous caption generation, enhanced with audio narration capabilities through integrated speech synthesis technology.

Published by: vikaspatanker

Authors: Praveen Lokanath

Abstract: The rise of artificial intelligence (AI) is reshaping the nature of work in ways that echo past industrial revolutions, yet with unprecedented speed and complexity. This paper explores how previous waves of technological transformation — from mechanization in the 18th century to the digital revolution of the late 20th century — can inform our understanding of AI's current and future impact on employment, labor markets, and workforce dynamics. Drawing lessons from history, the study highlights patterns of job displacement, creation, and evolution, emphasizing the critical roles of policy, education, and social adaptation. It also examines the unique characteristics of AI that distinguish it from earlier innovations, particularly its capacity to automate cognitive tasks and decision-making processes. By synthesizing historical insights and contemporary developments, the paper offers a framework for anticipating the challenges and opportunities AI presents, aiming to guide stakeholders in shaping a more equitable and resilient future of work

Published by: vikaspatanker

Authors: Atharva Daga, Viraj Laddha, Prathmesh Jain, Tanmay Sharma

Abstract: Brain tumor detection is important in neuroimaging, affecting patient outcomes and prognosis. To improve detection capabilities, this study uses MRI & CT Scan Image to classify brain tumor while employing deep learning techniques. We test how well pre-trained models like VGG-19, DenseNet-121, and ResNet-50 perform by using detailed information from MRI and CT scans to improve the accuracy of detecting brain tumors and help identify them more clearly and precisely, facilitating swift diagnosis and informed treatment planning. This research utilizes image fusion and prediction algorithms to address challenges such as limited data diversity and difficulties in differentiating tumor boundaries from surrounding tissues, thereby improving model performance. By evaluating the results, we identified the most accurate model for brain tumor diagnosis and provided insights into its use and impact on diagnosis. This research advances technology and improves patient outcomes through more accurate and timely diagnoses. Analysis shows Resnet-50 achieving the highest accuracy among all other models is effective for tumor detection.

 

Published by: vikaspatanker

Authors: M Devendar Reddy, S Akhil Reddy, Anand Jawdekar, N Saiprem,, B UdayKiran Reddy

Abstract: The results of the experiment indicate that combining reinforcement learning with vision-based techniques can offer signifi- cant improvements in autonomous naviga- tion [2],[5]. Scale-Invariant Feature Trans- form (SIFT) was particularly effective in recognizing both the delivery target and potholes with a high degree of accuracy [7],[10], ensuring reliable performance under varying conditions. Canny edge detection and the Hough Line Transform proved to be highly efficient tools for lane identification [4],[6], allowing the robot to maintain pre- cise lane alignment during movement. Fur- thermore, IMU-based orientation correction provided additional robustness, preventing errors caused by yaw drift and other orien- tation issues [7]. Collectively, these meth- ods enabled the robot to adapt dynami- cally to its environment and demonstrate consistent success across repeated trials [2]. These findings suggest that the proposed framework not only addresses the imme- diate problem of pothole detection [9],[10] but also enhances the overall safety and reliability of autonomous vehicles. Look- ing ahead, the study shows strong poten- tial for real-world applications, as it pro- vides a scalable and practical solution that can be integrated into future self-driving systems to improve passenger safety, vehi- cle durability, and overall traffic efficiency [5]. Autonomous driving continues to be one of the most promising innova- tions in intelligent transportation sys- tems, but real-world challenges such as potholes still pose serious risks to safety and efficiency [2],[5]. This study explores the application of rein- forcement learning for addressing the issue of pothole detection and avoid- ance in self-driving cars [2]. To evalu- ate the framework, a detailed robot simulation was built in the Webots environment, making use of Python programming and OpenCV for vision processing [8]. Within this setup, the robot was designed to complete three key tasks: it first identifies a delivery target symbolized by a gnome placed in the environment, then transitions into lane-following mode to maintain safe navigation, and finally responds appropriately by halting when a pot- hole is detected on its path [8]. Each of these components plays a crucial role in ensuring safe and reliable op- eration. The framework integrates several technologies, including real- time computer vision for object detec- tion, IMU sensor feedback for orien- tation correction, and motor control for smooth navigation [7]. These el- ements work together to enable the robot to perceive its surroundings, adapt to hazards, and make sequen- tial decisions that reduce the risk of accidents [2]. The results of the experiment in- dicate that combining reinforcement learning with vision-based techniques can offer significant improvements in autonomous navigation [2],[5]. Scale- Invariant Feature Transform (SIFT) was particularly effective in recogniz- ing both the delivery target and pot- holes with a high degree of accuracy [7],[10], ensuring reliable performance under varying conditions. Canny edge detection and the Hough Line Trans- form proved to be highly efficient tools for lane identification [4],[6], al- lowing the robot to maintain pre- cise lane alignment during movement. Furthermore, IMU-based orientation correction provided additional robust- ness, preventing errors caused by yaw drift and other orientation issues [7]. Collectively, these methods enabled the robot to adapt dynamically to its environment and demonstrate consis- tent success across repeated trials [2]. These findings suggest that the pro- posed framework not only addresses the immediate problem of pothole de- tection [9],[10] but also enhances the overall safety and reliability of au- tonomous vehicles. Looking ahead, the study shows strong potential for real-world applications, as it provides a scalable and practical solution that can be integrated into future self- driving systems to improve passenger safety, vehicle durability, and overall traffic efficiency [5].

DOI: https://doi.org/10.5281/zenodo.17174389

 

Published by: vikaspatanker

Authors: Ryan Gautam

Abstract: This secondary research review evaluates the extent to which current autonomous vehicle (AV) pedestrian detection datasets and validation protocols represent and protect disabled road users—including wheelchair users, white cane users, guide dog handlers, and mobility scooter users—across lighting and weather conditions. Synthesizing peer reviewed studies, standards analyses, government reports, and advocacy documents from 2015–2025, the review finds systematic underrepresentation of disability categories and accessibility infrastructure in widely used datasets, alongside documented detection biases that elevate risk for vulnerable pedestrians under low light and non standard movement scenarios. Current validation frameworks (e.g., functional safety and SOTIF) and regulatory pathways provide limited, non specific guidance on disability inclusive testing, allowing deployments that lack demonstrable parity performance for disabled pedestrians. The paper proposes a minimum pre deployment standard requiring disability inclusive dataset composition, category specific performance thresholds (with edge case coverage), and independent third party audits, with ongoing post deployment monitoring. This framework is feasible within established safety and regulatory processes and is necessary to align AV deployment with equity and safety obligations for all road users.

DOI: https://doi.org/10.5281/zenodo.17173964

 

Published by: vikaspatanker

Authors: Sandip Bhaskar Patil

Abstract: Hydrogen has been promoted as a versatile clean energy vector capable of decarbonizing hard-to-abate industrial sectors. This paper critically reviews the technical, economic, and infrastructural factors that determine whether hydrogen can move from todays too much "hype" to a reality of fit-for-purpose energy solution for industry and the energy sector. We review production pathways (gray/blue/green hydrogen), electrolyser technologies, storage & transport challenges, and industrial use-cases (steel, chemicals, refineries, and high-temperature heat). Key findings: (1) green hydrogen currently remains significantly more expensive than fossil-derived hydrogen, though projections indicate cost declines with scale and deployment; (2) hydrogen is likely to be most viable where direct electrification is infeasible (high-temperature heat, feedstock); (3) infrastructure and implementation gaps are significant and require policy, supply-chain, and finance coordination. The paper concludes with fit-for-purpose deployment pathways and policy recommendations to enable industrial adoption.

DOI: https://doi.org/10.5281/zenodo.17173815

 

Published by: vikaspatanker

Authors: Neelima Kasarla, Praneeth Kumar Kandukuri, TakeYadav Yerragolla, Srikanth Pandula

Abstract: The demand for digital content is growing quickly. Finding skilled creators is now a challenge. SnapHire is a platform that simplifies the hiring process for photographers, video editors, and reel makers. It has a simple interface. Clients can view portfolios, check skills, see availability in real time, and book services right away.This reduces delays compared to traditional hiring methods. Creators gain better visibility and can handle client requests more easily. Businesses and individ- uals can rely on skilled professionals. SnapHire is built to grow with tools that improve workflow and user experience. The platform emphasizes speed, clarity, and reliability. It connects clients and creators effectively. SnapHire offers high-quality content for marketing, branding, social media, and personal events like weddings and parties. It makes booking simple, increases creator visibility, and improves communication. This method offers a smooth and dependable experience for clients and professionals.

DOI: https://doi.org/10.5281/zenodo.17173653

 

Published by: vikaspatanker

Authors: Rajendra Singh Kushwah, Satya Prakash Srivastava

Abstract: As the increases the demands of automobiles has cause increases the traffic and increases the number of accidents which shows road safety importance and this is happen because of lack of urgent emergencies facilities in our country , Vehicle accidents and casualty happen on road side should provide on the spot help and rescue as soon as possible . However some problem faces by rescue team to find the accident location. There is no proper communication and retrieve information instantly of the specific accident spot area. This paper proposed a smart and reliable IOT technology provide various features by which we can find instantly the actual accidental location by the help of actual geographic coordinate’s position of vehicle accidents occur. There are various sensors available to check the primary data for detecting accidents. These geographical data can be sent to nearest ambulance for medical assistance, this will save time to reach the rescue team to the accidental location and provide rescue to the injured person. In this paper a real time automobile tracking system via Google earth is introduced and a LifiLink app to find the nearest ambulance location near to vehicle accident spot . The system uses two main components, a transmitting embedded module to interface for vehicle location GPS and GSM devices in order to determine and send vehicle location and status information SMS .

Published by: vikaspatanker

Authors: Sanyam Jain, Amit Kumar, Prakash Mali, Khushal Bharat Shivade, Krishna Gutte

Abstract: Rapid digital change in the fashion and retail world is underway. Virtual try-on technologies are changing how people shop. However, these technologies are also bringing up issues like data privacy and security. This paper introduces Style Secure Zone, an AI-enabled framework for secure, contactless, and personalized shopping for clothing.While physical fitting rooms will still be necessary in traditional department stores, the development of Style Secure Zone will tackle hygiene issues and help retailers improve their operations by reducing wait times. AI-powered and algorithm-supported product recommendations provide thoughtful outfit suggestions based on natural traits, preferences, personal styles, and fashion trends. 3D renders are also created to improve fit and feel.K-Movement’s testing shows that the system boosts consumer confidence, reduces product re- turns, and helps retailers manage inventory challenges related to space availability. Additionally, user assessments showcase great enthusiasm for the integrated, private, and interactive shopping experience across the platform. The overall findings highlight the importance of privacy-oriented digital retail platforms and demonstrate that safe, AI-enabled solutions, like the Style Secure Zone, look to strengthen customer engagement, retailer perfor- mance, and ultimately define the future of intelligent fashion retail.

DOI: https://doi.org/10.5281/zenodo.17164456

 

Published by: vikaspatanker

Authors: Pradeep Kumar Jaiswal, Rakesh Kumar Yadav, Manish Kumar Tiwari

Abstract: The aim of the research was to prepare low-cost adsorbents, including raw date pits and chemically treated date pits, and to apply these materials to investigate the adsorption behaviour of Cr(III) and Fe(III) ions from wastewater. The prepared materials were characterized using SEM, FT-IR and BET surface analysis techniques for investigating the surface morphology, particle size, pore size and surface functionalities of the materials. A series of adsorption processes was conducted in a batch system and optimized by investigating various parameters such as solution pH, contact time, initial metal concentrations and adsorbent dosage. The optimum pH for achieving maximum adsorption capacity was found to be approximately 7.8. The determination of metal ions was conducted using atomic adsorption spectrometry. The experimental results were fitted using isotherm Langmuir and Freundlich equations, and maximum monolayer adsorption capacities for Cr(III) and Fe(III) at 323 K were 1428.5 and 1302.0 mg/g (treated major date pits adsorbent) and 1228.5 and 1182.0 mg/g (treated saga date pits adsorbent), respectively. It was found that the adsorption capacity of H2O2-treated date pits was higher than that of untreated DP. Recovery studies showed maximal metal elution with 0.1 M HCl for all the adsorbents. An 83.3–88.2% and 81.8–86.8% drop in Cr(III) and Fe(III) adsorption, respectively, were found after the five regeneration cycles. The results showed that the Langmuir model gave slightly better results than the Freundlich model for the untreated and treated date pits. Hence, the results demonstrated that the prepared materials could be a low-cost and eco-friendly choice for the remediation of Cr(III) and Fe(III) contaminants from an aqueous solution.

Published by: vikaspatanker