Infrared Maritime Target Detection Under Wave Interference Based on Long–Short Channel Features and Nonoriginal State Decomposition

Tensor decomposition-based methods exhibit advantages in detecting infrared maritime small targets (IMSTs) but struggle with strong wave interference (SWI) in maritime environments. SWI often presents similar or even higher grayscale values than IMSTs, making it challenging to suppress SWI using spatial information alone. To address this, a new method called long-short channel features and nonoriginal state decomposition (LSCF-NOSD) is proposed. This method leverages the temporal consistency of IMSTs and the temporal variability of SWI, designing a long-channel 3-D structure tensor (3DST) weight to enhance targets. Besides, using the detection results from the previous frame, considering the viewing distance difference (VDD) between IMSTs and SWI, the short-channel pixel encoding weight (PEW) is designed to suppress SWI. Finally, a nonoriginal state decomposition (NOSD) optimization method preseparates target and background components, guiding the model toward a more accurate solution. Testing across multiple datasets demonstrates that LSCF-NOSD effectively mitigates SWI while maintaining target saliency.