Object Detection and Localization Optimization Algorithm Based on Attention Mechanism and Variational Convolution

To address the challenges of small and medium-sized target recognition, as well as inconspicuous foreground features in complex backgrounds, this paper proposes a novel visual object detection algorithm. The algorithm is built upon an improved attention mechanism and variability convolution to enhance detection accuracy. First, the Convolutional Block Attention Module (CBAM) is integrated into the YOLOv8 algorithm to boost the model's performance and robustness. Next, the Deformable Convolutional Network (DCN) is incorporated into the backbone network of YOLOv8 to improve the model's adaptability to irregularly shaped targets. By strengthening the model's attention to key features and facilitating more diverse feature extraction, target detection performance is effectively optimized. Finally, the advantages of these improvements in detection accuracy and generalization ability are validated through ablation experiments and data visualization comparisons.