Character Reconstruction Based On Local Field Potential Signals Of Pigeon Optic Tectum Neurons

Visual loss caused by pathological changes in the visual system such as the retina or optic nerve has become the most serious disability affecting human life.The study of visual prostheses has made it possible to repair blindness caused by lesions and injuries of the retina,optic nerve and visual cortex.In the past study of animal models of retina or optic nerve prosthesis,the evaluation of stimulating effect was mainly performed through animal behavior feedback methods.There is no direct quantitative evaluation method.Therefore,it is of great significance to develop a method for stimulating reconstruction from the visual system site downstream to evaluate the stimulating effect of the visual prosthesis.This thesis presented a method for reconstructing visual information by collecting neuron cluster response signals in optic tectum(OT).The micro-electrode array was used to record the multi-channel local field potential signal(LFP)under character stimulation,and various features of the local field potential signal were extracted,the reconstruction model was constructed,and the stimulating character image was reconstructed.The main work and research results of this thesis are as follows:(1)Obtained neuron cluster response data for reconstruction.A microelectrode array was implanted in the OT region of the pigeons and LFP signals were collected to filter out the power frequency and myoelectric interference of the LFP signal.The position of the receptive field in the OT region of the pigeon was determined using a checkerboard experiment to determine the position of the visual stimulus.A scheme of moving the character images in sequence was designed to overcome the problem that the receptive fields corresponding to the signals collected by the electrodes cannot cover the entire image.(2)The characteristics of local field potential signals characterizing character stimulation information were determined?The information carried by the phase and amplitude features of the local fieldpotential was compared by information theory.It was found that the information carried by the phase feature is greater than the information carried by the amplitude feature.The information carried by the amplitude at the frequency points 43 Hz,54Hz,62 Hz,and 74 Hz is the largest,and the information carried at the frequency points74 Hz,105Hz,144 Hz,and 175 Hz is the largest.In addition,the main carrying band of the amplitude is concentrated between 35 Hz and 100 Hz.When the frequency band is greater than 100 Hz,the information carried by the amplitude is relatively small,and the phase also carries information after 100 Hz.The effect of different feature combinations on the reconstruction performance was compared through reconstruction accuracy,from which it was discovered that the features of the phase-nonlinear mapping and the combined features of phase and amplitude were used to reconstruct the characters Image stimuli with getting higher accuracy.(3)Determined the feature parameters of the reconstruction of the channel,feature start time,feature interception time length,etc.The factors affecting the reconstruction performance include the frequency band where the feature is located,the channel where the feature is located,the start time of interception of the feature data,and the time length of interception of the feature data.Analyze the influence of the frequency band factor of the feature.It was found that when the frequency band contains 70-110 Hz,the correct rate of reconstruction is greater than 0.96.The correct rate of reconstruction with the amplitude feature after100 Hz is less than the correct rate of phase feature reconstruction.Analyze the effect of channel factors and found that when the number of channels is greater than 2,the correct rate of reconstruction begins to saturate.Analyze the influence of the start time factor and found that within 10 ms after the start of the stimulus,the neuron response already contains the external stimulus information.Analyze the effect of data length time and found that when the data length is greater than 0.17 s,the reconstruction accuracy tends to be saturated.(4)The reconstruction model was designed to achieve the reconstruction of character images.In this thesis,we designed two reconstruction models based on logistic regression and KNN,and compared the difference of reconstruction quality of the two models.Itwas found that the accuracy of both algorithms is above 0.93 no matter which features were used.For character image reconstruction,the correctness of the reconstruction with KNN is greater than with logistic regression.Using features that map the phase non-linearly and features that combine phase and amplitude in a particular way,the accuracy of the logistic regression algorithm is above 0.96.The results of this study show that for reconstructing a character image,a simple algorithm model can be reconstructed well.From the comparison of the algorithms,the importance of finding a combination of excellent features was further illustrated.At the end of this thesis,the versatility of the model was verified,and it was found that the character reconstruction model constructed in this thesis is universal,and can be reconstructed by any character image to rebuild the other characters.