Authors: Simhachalam Patnana, S.Sudeer Kumar, Y. Jagadesh Kumar
Abstract: Automatic image recognition has become a fundamental component of modern intelligent systems, finding applications in areas such as food recognition, healthcare imaging, smart surveillance, object detection, and visual analytics. However, traditional image classification techniques often face challenges due to image noise, class imbalance, varying lighting conditions, complex backgrounds, and diverse visual patterns, which reduce classification accuracy and prediction reliability. To address these challenges, this project proposes a Smart Vision Fusion Architecture for Multi-Class Image Recognition (SVFA-MCIR), an intelligent hybrid framework that combines deep learning and machine learning techniques for efficient multi-class image classification.The proposed framework incorporates image preprocessing, enhancement, augmentation, and feature optimization techniques to improve dataset quality and model performance. Existing image recognition models such as CNN, EfficientNet + XGBoost, and DenseNet + XGBoost are initially evaluated to analyze their classification capabilities. To further enhance recognition accuracy and classification stability, the proposed system integrates ResNet50 and XGBoost into a unified hybrid architecture. ResNet50 is utilized to extract high-level visual features and complex image representations, while XGBoost performs optimized multi-class classification using the extracted deep feature vectors.Experimental results demonstrate that the proposed SVFA-MCIR framework achieves superior performance in terms of recognition accuracy, prediction robustness, feature learning capability, and computational efficiency when compared with existing approaches. The framework provides a scalable, adaptive, and intelligent solution for modern image recognition applications and contributes to the advancement of smart vision systems through accurate and reliable multi-class image classification.
DOI: http://doi.org/
