Rose Color Detection for Blind People's using Deep Learning

Abstract

This thesis presents the development of a mobile app-based solution for rose color detection designed specifically for visually impaired individuals. The solution leverages advanced Vision Transformer (ViT) architectures, particularly ViT-B16 and ViT-B32, to enable real-time, accurate, and accessible color recognition. Addressing the limitations of traditional color identification methods, the proposed solution empowers users to independently experience and identify rose colors, fostering inclusivity and autonomy. The study incorporates synthetic data generation techniques to overcome the challenges of limited labeled datasets, enhancing model generalization across diverse environmental conditions. The lightweight nature of ViT-B16 and ViT-B32 ensures compatibility with standard mobile devices, optimizing computational efficiency while maintaining high accuracy. Intuitive feedback tailored to the needs of visually impaired users provides actionable and descriptive insights into detected colors. The research methodology involves data collection, model training using ViT-B16 and ViT-B32 architectures, and iterative app design, followed by rigorous testing under varied real-world conditions to evaluate performance. The results demonstrate the app’s effectiveness, achieving 99.81% accuracy, and potential as a practical assistive tool. This study contributes to the growing field of AI-driven assistive technologies by addressing critical gaps in accessibility, dataset diversity, and real-world adaptability. Beyond its immediate application in rose color detection, the findings have broader implications for developing inclusive technologies that enhance the quality of life for individuals with visual impairments. The thesis concludes with recommendations for future improvements and scalability of the proposed solution.