Research On Dim Infrared Small Target Detection Based On BEMD And Spatial-temporal Method

Infrared detection system has a good concealment, strong anti-interference ability, can detect camouflage target, which can be an important supplement of the radar warning system. However, small infrared target detection under complex and low SNR background is always a very difficult task, and because of this, the subject becomes the core of remote precision system and early warning system. Aiming at this problem, many scholars paid a lot of energy, and made a lot of fruitful infrared small target detection algorithm, but at this stage these algorithms are also some limitations. This paper puts forward a new method of small infrared target detection under complex and low SNR background based on BEMD and spatial–temporal. EMD algorithm is a powerful tool for the one-dimensional non-stationary signal analysis,unlike Fourier transform and wavelet transform, EMD is entirely data-driven decomposition, do not need transform kernel function, has a strong ability of adaptability. In this paper, the 1D EMD algorithm is extended to 2D applications,the BEMD is used to process infrared small target image to get BIMF1, which contains small target signal characteristics, then the BIMF1 is fused with Phase-shifting, at last, the segmentation process is made on fusion image to obtain the result of infrared small target detection. Experimental results show that the detection method in this paper has a good effect against the complex background and low SNR to detect infrared small target image.The traditional BEMD has the non-uniform convergence and mode aliasing problems, which greatly limit the applications of BEM. In order to solve these two problems, we reference the “Fishbone diagram analysis method” to trace the reason, and point that, Extreme Point Extraction inadequate is the main reason of the non-uniform convergence and mode aliasing problems. To address this issue, we put forward an improved BEMD algorithm to settle the non-uniform convergence and mode aliasing problem. The experience data shows that the improved BEMD has solved the non-uniform convergence and overcome the mode aliasing problem.After that, we research on the small infrared target feature extraction by the improved BEM, though theory analysis and experiment, we point that, under the premise of overcoming the mode aliasing, the BIMF1 can considered as the target feature information,which,not only makes the study of BEMD applied to infrared small weak target detection more targeted but also improves the efficiency of the system greatly.At last, we analysis the characteristics of BIMF1 and phase-shifting difference results, and point that, the fluctuations in infrared image background can be cut by multiplicative fusion of BIMF1 and phase-shifting difference results. After this, the SNR of small infrared target image can be greatly improved. And using the fused image, we can easily get the accurate segmentation of small target. Experimental data show that, our method of small infrared target detection under complex and low SNR background based on BEMD and spatial–temporal has high accuracy and low false alarm rate.