Research On Noise Reduction Method Based On PCNN And Its Application

Pulse Coupled Neural Network(PCNN) is in recent years, as the research anddevelopment of biological neurology, the emerging of a new artificial neural networkmodel. It has the unique features for a biological background, the dual channel mechanism,output pulse, variable threshold value, the product coupling of internal behavior in thecomposition of neurons. PCNN is an excellent image processing tool, and it has beenwidely used in various fields of image processing, such as image denoising, imagesegmentation and image fusion, particularly in terms of image noise reduction. Althoughthe PCNN model has excellent performance, but there are some defects. Therefore, the in-depth research on the theory of PCNN is opened, some of the shortcomings of PCNNmodel is improved. The main innovations of this article are as follows:1. The fixed value for synaptic links strength can not reflect the level of mutualinfluence between different neurons. In order to solve the problem, an adaptive synapticlink strength PCNN model on image filtering method is proposed. The synaptic linksstrength is modified as the adaptive value according to similarity degree of differentneurons and neighborhood neurons, and the time series matrix to record the number ofneurons firing is added. It can identify the noise points accurately according to the numberof neurons ignition, so the misjudgment of noise points is avoided. The proposed algorithmcan achieve to remove image noise effectively, and possesses the better capability toprotect edges and details of images.2. For the threshold function of PCNN model decayed according as a singleexponential, the algorithm running speed slowly, an adaptive threshold function PCNNmodel is proposed. On the basis of the improved synaptic link strength for adaptive value,the neuron ignition frequency matrix are added. The threshold function attenuation curve ismodified to the linear attenuation for thresholds without ignition neurons, and thethresholds of ignition neurons are still according to the exponential damping. Suchimprovement realizes the adaptive regulation attenuation speed. Noise reduction for data using this method, which maintains the original data characteristics, shortens the runningtime of the algorithm and achieves quickly and efficiently filter out the noise data.3. The improved PCNN model is applied to the coal to methanol data and strip hot-dipgalvanizing data. The data of noise reduction and non noise reduction is forecasted andanalysed used BP neural network. The results show that the accuracy of data predictionafter noise reduction is obviously improved.In this paper, the synaptic link strength and the threshold function of PCNN model aremodified, thus the accuracy of the algorithm is improved and the running time of thealgorithm is shortened. The noise reduction experiments using the simulation data and theactual production data have both obtained better noise reduction effects. So it shows thatthe proposed methods have certain effectiveness and practicability.