A Research And Implementation Of The Sea-Surface Dim Target Detection On Strong Clutter Jamming Infrared Image

The automatic detection technology for far sea-surface targets by using infrared imaging detector has been widely applied in civilian shipslocalization, port monitoring and military ships detection.With the constant improvement of the demand, the accurate positioning and real-time detection of far sea-surface targets have been one hot spot of current research.For the sea-surface dim target detection on strong clutter jamming infrared images, it is the first time to propose a filtering algorithm based on center around suppression(called CAS). On this basis, we have designed and implemented a kind of infrared dim target real-timedetection algorithmbased on CASfilter under strong clutter interference. It is the first time to introduce convolution neural networkCNNmodelto the sea-surface dim target detection fieldbased on infrared image and gained good results.The infrared imaging of sea-surfacetargetwill be affected by cloud cover, illumination, smoke and other complex signal interference, characterized by strong clutter. And imagepixelsof weak infrared target is little, signal strength is weak, present article porphyritic or shape, target shape information is not sufficient, characterized by less pixels weak targets.For the weak targetof the strong sea clutter infrared image, we propose the filter algorithmbased on center around suppression which using probability density distribution of the partial images and by contrasts of neighborhood information from the image neighborhood establish inhibition model. The way is good at background suppression and target enhancement.Experiments show that this method is not only good effect and is very beneficial to hardware implementation.After the image is filtered by CAS, we select a better segmentation and tag algorithm anddesign a real-time infrared small target detection algorithm under complex backgroundbased on CAS filter with extracting the interesting area. In addition, we give a design scheme of infrared target real-time detection systembased on the FPGA/DSP architecturewhich makes the target detection task parallel decomposition and more processors work together. The schemeimproves the real-time performance of the algorithm. The Infrared target real-time detection system verified by the experiments not only can successfully detect weak strong clutter jamming infrared image to the surface of the target, and meet the requirements of high reliability, high real time capability of practical application.In order to further improve the strong clutter jamming infrared image under the weak target detection accuracy, the deep learning technology is introduced into the surface of the weak infrared image targets detection field.With taking target area as positive samples and background area as negative samples, we build the convolutionalneural network(called CNN) model and give the model to configuration and trainingunder the deep learning framework CAFFE.Through contrast experiment, in terms of performance, the target detection algorithm based on CNN target detection rate is high, strong adaptability, better than that of the target detection algorithm based on CAS filter.