The Weak Signal Detection Based On Stochastic Resonance Mechanism And Its Application To Image Enhancement

Stochastic resonance mechanism, which is the internal factor to coordinate the signal and noise in nonlinear system, plays a significant role in the detection and enhancement of weak signals. It is considered that the stochastic resonance mechanism of aperiodic response exists in nervous system and acts on the biological visual processing. Therefore, this paper studied on the characteristics of biological vision, such as image restoration and target enhancement, by the way of aperiodic stochastic resonance mechanism based on FitzHugh-Nagumo (FHN) neuron and bistable system. In succession, the crucial role of stochastic resonance in the visual system was interpreted. Furthermore, stochastic resonance implemented the enhancement of low-dose lung CT, which stood for the practical application of weak signal detection.First of all, the response coding in FHN neuron, analyzed by circle map mechanism, was applied to the frequency measurement of highly sensitive stimulus. The first step of research verified the nonlinear characteristic of FHN neuron, which is necessary for stochastic resonance phenomenon. And then, it was discovered that the appropriate intensity noise could enhance the FHN neuron and bistable response to the stimulus, including aperiodic square wave and continuos wave. Following this, a new image enhancement method based on stochastic resonance of FHN neuron was proposed for the low dose lung CT referred to medical field. The new method dramatically enhanced the object region of CT with suppressing noise. The main contributions of this thesis are summarized as follows:(1). A new weak signal detection method based on chaotic circle map of FHN neuron was presented. The work researched the phase relation of FHN neuron response, which satisfied with the typical characteristics of chaotic circle map. Afterwards, the frequency measurement of highly sensitive stimulus was simulated by combination chaotic circle map of FHN neuron with symbolic dynamics approach. The simulation showed that the measurement resolution of pulse period could be up to 0.01ms with the 32 bits length of symbol sequence. It was demonstrated that significant nonlinear characteristics existed in FHN neuron, which meant a theoretical basis for the application of FHN neuron to stochastic resonance.(2). The image restoration method based on bistable stochastic resonance was applied to the low SNR and gray image. Previously, this thesis studied on the restoring ability of bistable stochastic resonance on the noisy signal with continuos amplitude. The research laid the theoretical foundation for gray image restoration. Simultaneously, the Hilbert scanning method converted the image signal into one-dimensional signal to ensure the spatial correlation of image pixels, which was beneficial to the coordinate relations among bistable system, pixel signal and noise. The new method overcome the limitation of traditional stochastic resonance and gave fully play for its feature of weak signal enhancement by noise. Compared with the traditional image restoration method based on filter, the suggested one had advantages on better robustness and less judgment error in results.(3). The method based on FHN neuron stochastic resonance was used in image enhancement. To begin with, the FHN neuron stochastic resonance was used for the research on the response to the noisy and aperiodic signal. And the study provided a theoretical basis for the two-dimensional image enhancement. The raster scanning method was selected to achieve the image dimensional reduction. Since then, the response of stochastic resonance was obtained after the synergism between FHN neuron and noise. Furthermore, a new discriminator was proposed to get the final judgment result of image enhancement. The experimental results showed that the suggested method could reduce the noise and stretched the contrast between object region and background in CT while retaining the image details. And this method performed better on the noise filtering and contrast stretching than the conventional ones.