Local Patch Network With Global Attention for Infrared Small Target Detection

Infrared small target detection plays an important role in the infrared search and tracking applications. In recent years, deep learning techniques have been introduced to this task achieved noteworthy effects. Following general object segmentation methods, existing methods usually process image from global view. However, locality of targets extreme class-imbalance between background pixels are not well-considered by these which causes low-efficiency on training high-dependence numerous data. A local patch network (LPNet) with attention is proposed article detect jointly considering properties images. From view, a supervised module trained spread map suppress most irrelevant features. patches split features share same convolution weights each other LPNet. By leveraging both properties, data-driven framework has ability fusing multiscale for detection. Extensive experiments synthetic real datasets show that method achieves state-of-the-art performance comparison traditional methods.