The article has been submitted to IEEE Transactions on Geoscience and Remote Sending

we propose in this paper a robust infrared small target detection method jointing multiple information and noise prediction, named MINP-Net.

We contribute an infrared small target segmentation dataset, called NCHU-Seg. The presented NCHU-Seg dataset consists of 590 infrared images which are almost selected from the real-world infrared images photographed. The infrared small targets mainly include aircraft, birds and ships, and the target sizes are distributed between 3×3 pixels and 9×9 pixels within a 256×256 infrared image, which are strictly meet the SPIE definition for infrared small target.

• Tested on Ubuntu 20.04, with Python 3.9, PyTorch 1.11, Torchvision 0.12.0, CUDA 11.3.1, and 1x NVIDIA 2080Ti
• The NUAA-SIRST download dir [ACM]

python train.py

python test.py 

python visulization.py

1. Li, Boyang and Xiao, Chao and Wang, Longguang et al., Dense Nested Attention Network for Infrared Small Target Detection.//arXiv preprint arXiv:2106.00487. 2021. [code]