A Hybrid Technique for Robust Object Detection for Identification of Classifying Objects
Object detection in real-world environments faces significant challenges from blur, noise, occlusion, and varying illumination, which degrade detection accuracy and robustness. This paper proposes an enhancement-assisted Faster R-CNN technique that integrates classical image restoration with deep learning to improve detection under degraded conditions. The method applies Wiener filtering to recover lost structural details, followed by adaptive histogram equalization for contrast restoration. Boundary box aware augmentation & optimized anchor estimation to improve localization exactness across diverse object scales during training. At inference, Soft-NMS and TTA enhance prediction stability in crowded scenes. The system is built on a ResNet-50 backbone within the Faster R-CNN architecture and is evaluated under three conditions: original, noisy & blurred images. Results demonstrate mAP improvement from 93.57% to 98.41%, a 3.5% increase in detection rate, and enhanced PSNR and SSIM scores. These findings confirm that integrating image restoration with optimized detection strategies significantly improves robustness without architectural modifications.