CUDA-based Moving Target Detection In Remote Sensing Image

Moving target detection in remote sensing image is an important research directionin remote sensing field, and its real-time detection has always been a hot topic. Becauseremote sensing image data has large data storage capacity, high resolution and hugescale information and the moving target detection algorithm such as mathematicalmorphology algorithm holds large computation and high complexity, the current CPUserial calculation method leds to long operation time and low operation efficiency thuscan not reach the real-time requirements. It is urgent to seek a new processing method.In this paper’s main algorithm of moving target detection algorithm (expansion,corrosion, open and close, gray quantization, linear interpolation, etc.) the operation ofeach pixel is usually only related with local pixels and the basic process of differentpixel data are similar, which makes it very suitable for parallel processing. GPU generalcalculation is a kind of concept highly stressed deserialize calculation, which could beused to deal with repeated large data quantity computing. If optimized proper, thecomputing speed could be one hundred times and even higher than that of CPU. By thediscussion of the parallel architecture Graphics processing unit (GPU) and unifiedcomputing equipment structure (CUDA) with agile programmability, the moving targetdetection in remote sensing image on CUDA platform is realized achieving data parallelprocessing, with high real-time and high calculation accuracy.