IR-TransDet: Infrared Dim and Small Target Detection With IR-Transformer

Abstract

Infrared dim and small target detection is one of the crucial technologies in the military field, but it faces various challenges such as weak features and small target scales. To overcome these challenges, this article proposes IR-TransDet, which integrates the benefits of the convolutional neural network (CNN) and the Transformer, to properly extract global semantic information and features of small targets. First, the efficient feature extraction module (EFEM) is designed, which uses depthwise convolution and pointwise convolution (PW Conv) to effectively capture the features of the target. Then, an improved Residual Sim atrous spatial pyramid pooling (ASPP) module is proposed based on the image characteristics of infrared dim and small targets. The proposed method focuses on enhancing the edge information of the target. Meanwhile, an IR-Transformer module is devised, which uses the self-attention mechanism to investigate the relationship between the global image, the target, and neighboring pixels. Finally, experiments were conducted on four open datasets, and the results indicate that IR-TransDet achieves state-of-the-art performance in infrared dim and small target detection. To achieve a comparative evaluation of the existing infrared dim and small target detection methods, this study constructed the ISTD-Benchmark tool, which is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://linaom1214.github.io/ISTD-Benchmark</uri> .