Enhanced Deeplab Network for Infrared Small Target Detection

With the development of deep-learning methods, convolutional neural networks are widely used to detect infrared small targets. But as the pooling layers get deeper, detailed information about infrared small targets will be lost since the infrared targets are dim and small. Thus we proposed an infrared small target detection network based on the deeplabv3+ model. In the proposed model, high-level semantic information and low-level high-resolution information are fused to extract the feature of infrared targets. Besides, SA attention module is adopted to extract the feature of infrared targets more precisely and suppress the noise of the background by using an attention mechanism before fusion. Experiments on open-source dataset sirstaug prove that our method EDeeplab is superior to other state-of-the-art methods.