In order to improve the ability of real time computation

an attention-based IR ship detection is proposed to reduce the data to be processed and enhance the processing efficiency. The attention is divided into "pre-attention" and "attention". A nonlinear sampling model is adopted to reduce the resolution in pre-attention while keeping adaptive to size variance of the target; the hot region which always refers to the engine or chimney of a ship is adopted as the guidance of attention to the areas of interest(AOI)

then the waterline

which is taken as the less salient feature of infrared ship target

is detected in the AOI. If the waterline feature exists in an AOI

it means a target is detected; otherwise

the AOI is taken as false alarm. To test the performance of the approach proposed

an algorithm is designed and realized both on PC and on a multi-DSPs system. Experiments demonstrate that the approach proposed can enhance the detection efficiency

and it is adaptive to the size of target.