Improved Image Classification in Rapid Detection Tasks Using Super Resolution

This paper explores the enhancement of classification accuracy for blurred images in rapid detection scenarios using Super Resolution Generative Adversarial Networks (SRGAN). Blurred images, common in real-time applications such as security surveillance and medical diagnostics, pose significant challenges for accurate image classification. While traditional methods like bilinear interpolation offer some improvements, they fall short in significantly enhancing image quality and classification performance. This paper proposes a novel detection system integrating SRGAN with MobileNetV2 to address these challenges. Through a series of controlled experiments, we demonstrate that SRGAN effectively reconstructs high resolution images from low resolution inputs, leading to a substantial improvement in classification accuracy. Specifically, SRGAN enhanced images achieved a classification accuracy of 96.35%, outperforming both the original blurred images (90.62%) and those processed with bilinear interpolation (90.26%). Additionally, SRGAN shows superior performance in image quality metrics, with higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) scores compared to bilinear interpolation. These results demonstrate the effectiveness of SRGAN in real-time applications that require both precise and rapid image analysis, indicating its advantages over traditional image enhancement techniques.