The invention belongs to the technical field of
image processing and
computer vision, and particularly relates to an
infrared sequence
small target detection method based on
multilevel memory fusion, which comprises the following steps: coding
infrared video data after frame extraction
processing of
infrared video sequence data to obtain different scale features and generate pseudo coding information of a first frame; a memory value vector and a memory key vector are obtained through coding of the
working memory vector coding module, a memory matching result is obtained through the
working memory vector reading module, and meanwhile, a short-
term memory strategy and a long-
term memory strategy are fused; and the memory reading result and the cross-layer connection feature information are sent into a cross-layer attention gate module to obtain fused information,
feature mapping of the
small target is obtained through layer-by-layer up-sampling of a decoder, and the
small target is successfully segmented from the
infrared image. According to the invention, through mutual cooperation of multi-level memory strategies, association information of front and back frames is fully extracted, the detection precision of a small target under a complex background is effectively improved, and the
false alarm rate is significantly reduced.