Pulse Coupled Neural Network Model Analysis And Its Related Applications

Pulse Coupled Neural Network(PCNN),as a mathematical abstraction of human visual characteristics,has been widely used in various fields of image processing.However,the early classic PCNN model is complex,with many parameters,high computational complexity,and low processing efficiency.In order to solve above problems,many scholars are committed to the simplification of the PCNN model and the setting of parameters.The research results show that the performance of the simplified model has not decreased,and the number of parameters to be set is reduced,which is more conducive to transfer the model in other applications.As for the model simplification,the most famous models are Simplified Pulse Coupled Neural Network(SPCNN),Intersecting Cortical Model(ICM)and Spiking Cortical Model(SCM).These models are widely used in image enhancement,segmentation,denoising,edge detection and target recognition,which greatly promotes the development of related fields.As for parameter setting,there are empirical setting methods,automatic parameter setting methods based on image statistical characteristics,and methods based on data set training parameters.It can be seen that model simplification and parameter setting constitute the two camps of PCNN model related research,and more and more scholars have invested in the research of PCNN model.This paper aims to explore,analyze and study the image processing method based on PCNN model,and proposes several improved PCNN models,which are used in the fields of target detection,image segmentation and quantization.The main contributions and innovations of this thesis are listed as follows:1.The classical PCNN model converts image processing problems into neuron firing problems,where each pixel corresponds to one neuron.However,the mathematical coupling characteristics of the traditional PCNN make the firing time of the neurons an integer,which can not simulate the natural ignition time of the real neurons.Based on this problem,this paper proposes a non-integer step index SPCNN model.By adding a non-integral step size in the iterative process,the mathematical coupling firing characteristics of PCNN are basically disappeared,and the ignition timing of neurons is closer to the true value.At the same time,this non-integer-level step size makes the PCNN model process the image more finely,especially at uncertain boundaries or targets.By reducing the step size,the processing resolution is higher,and the detection and segmentation are more delicate and refined.In addition,by combining with multi-resolution analysis theory,the model has been successfully applied to the mammography calcification detection task,which can effectively detect breast lesions mixed in high-frequency tissues from complex environments.2.Different neurons in mammals have significant differences in structure and connection.This difference(heterogeneous property)enables animal neurons to extract image information efficiently when sensing images,which providing a powerful guarantee for the visual system to process images.Based on the theory of neuron isomerism,this paper proposes a heterogeneous simplified pulse coupled neural network for natural image segmentation.HSPCNN is connected by several SPCNN with different parameters according to different weights to form a parallel processing model.Its parameters are fully adaptive.And each SPCNN unit has a normalized output of different intensities,corresponding to different degrees of response of different regions of the brain to external stimuli.This model divides the image into different regions according to the gray level.This paper introduces the inter-class error and the intra-class error to revise the defects of the image segmentation evaluation index,and successfully applied this index to evaluate the performance of image segmentation method.3.In order to solve the defect of boundary effect caused by heterogeneous pulse coupled neural network(HPCNN)in image quantization,this paper proposes a new heterogeneous network model(Sine-Cosine Heterogeneous Pulse Coupled Neural Network,SC-HPCNN),which introduces the sinusoidal cosine oscillation term into neuron threshold and internal activity term to form an SC-PCNN.The sine-cosine oscillator can add short-term micro-increment in dynamic threshold and internal activity due to oscillation fluctuation.This increment can effectively suppress the system accumulated noise and make the segmentation result quickly converge.This model can reduce the system noise and external noise when combined with morphological operation.After the original image is segmented into the target area and the background area by using this model,two PCNN units with different parameters are used to quantize the target area and the background area,respectively.This heterogeneous structure consisting of SC-PCNN and two different parameters PCNN is SC-HPCNN.Finally,the quantization image of the original image is obtained by superimposing and inverting the time-sharing matrix of the two PCNN outputs.