Receptive-Field and Direction Induced Attention Network for Infrared Dim Small Target Detection With a Large-Scale Dataset IRDST

Abstract

Infrared small target detection plays an important role in military and civilian fields while it is difficult to be solved by deep learning (DL) technologies due to scarcity of data and strong interclass imbalance. To relieve scarcity of data, we build a massive dataset IRDST, which contains 142 727 frames. Also, we propose a receptive-field and direction-induced attention network (RDIAN), which is designed using the characteristics of target size and grayscale to solve the interclass imbalance between targets and background. Using convolutional layers with different receptive fields in feature extraction, target features in different local regions are captured, which enhances the diversity of target features. Using multidirection guided attention mechanism, targets are enhanced in low-level feature maps. Experimental results with comparison methods and ablation study demonstrate effective detection performance of our model. Dataset and code will be available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://xzbai.buaa.edu.cn/datasets.html</uri> .