CNN‐based infrared dim small target detection algorithm using target‐oriented shallow‐deep features and effective small anchor

• DOI: 10.1049/ipr2.12001
• Publication Date: 2020-12-09T04:12:18Z
• Authors: Jinming Du; Huanzhang Lu; Moufa Hu; Luping Zhang; Xinglin Shen
• Keywords: Computer vision and image processing techniques; QA76.75-76.765; Instrumentation and techniques for geophysical, hydrospheric and lower atmosphere research; Optical, image and video signal processing; Photography; 0211 other engineering and technologies; Computer software; 02 engineering and technology; TR1-1050

Abstract For the extremely small size and low signal‐to‐clutter ratio, target detection in infrared images is still a considerable challenge. Specifically, it very difficult to detect point targets because there no texture shape information can be used. A target‐oriented shallow‐deep feature‐based algorithm proposed, opening up promising direction for convolutional neural network‐based dim algorithms. To ensure that instances used correctly networks, effective anchor designed according shallow layer of ResNet50. determine whether result belongs target, authors depend on centre included ground truth area, rather than Intersection Over Union overlap rate, which avoids misjudging result. In this way, trained detected through More importantly, demonstrate spatially finer features are crucial semantically stronger deep helpful improving probability. Experimental results simulation data sets real show proposed when local ratio approximately 1.3, displaying infinite advantage great potentiality.