Improvement of Small Target Detection Algorithm Based on SSD

To address the issues of imbalanced target distribution within the dataset and the deficiency of high-level semantic information in the lower-level feature maps for small target detection, this paper proposes a small target detection algorithm grounded in feature fusion. The algorithm builds upon the SSD framework. To enhance the detection background for small targets, a data augmentation technique is employed, bolstering the model's generalization capacity for small and medium-sized targets. To augment the semantic information in the lower-level feature maps, a feature fusion module is incorporated, injecting high-level semantic data into the low-level feature map. Comparative experiments conducted on the dataset demonstrate the algorithm's effectiveness, showing improved overall detection performance and enhanced accuracy for small and medium-sized targets.