Small target detection with decoupling and decorrelation of confusion features

Abstract

Small target detection has important application value in military, cigarette defect detection, monitoring and remote sensing fields. However, due to the complex background interference of infrared images, the background and target features are confused with each other, which makes it difficult for existing detection methods to effectively distinguish the features of the background and the target, resulting in poor detection effect. Therefore, to solve this problem, this paper proposes a small target detection method based on confusion feature decoupling and decorrelation, aiming to achieve accurate detection of small targets in complex environments by extracting robust small target features. Specifically, in confusion feature decoupling and decorrelation, we propose a confusion feature decoupling and decorrelation module. By introducing a feature decoupling mechanism, the features of the input image are decomposed into independent background features and target features, and feature decorrelation is used to achieve independence between the target and the background. The target features after decoupling and decorrelation are purer, which helps to reduce background interference and thus improve detection performance. Systematic experimental results show that the detection performance of the proposed method on public small target detection datasets is much better than that of existing advanced detection methods.