透過您的圖書館登入
IP:3.133.160.75
  • 學位論文

利用時間延遲神經網路改進SD大鼠大腦運動區皮質層誘發電位的動作命令

An Improved Action Command Encoding of Cerebral Cortex M1 Evoked Potential of SD Rat Using Time Delay Neural Networks

指導教授 : 駱榮欽

摘要


當我們在思考、感覺和行動時,神經元就會在腦中傳遞訊息,近幾十年來,雖然我們對大腦功能以及細胞已經有明顯的瞭解,然而對於神經元集群響應的研究卻還未完整。為了瞭解在進行不同動作時,大腦神經元間傳遞的訊息是否存在著固定的微指令,本研究先取得大鼠不同動作時M1皮質層的訊號,其中會利用NLEO、ICA及DDIA找出特徵,接著使用TDNN建立一套編碼系統,將特徵編碼成Code,再選取各個區段的Code編碼成Symbolic,最後再由各類動作訊號的Symbolic組成Command,並觀察其分類結果。透過對各種動作的神經訊號進行編碼,進而了解大腦是如何進行溝通的以及了解不同動作間的訊號是否存在著固定的微指令命令。本研究利用TDNN建立編碼系統,當不計算訓練的樣本時,利用模擬訊號模擬三種動作的正確率分別為72.7%、72.7和81.8%,而在實際動作訊號方面,走路、站立及頭動的正確率為分別為33.3%、74.1%和40.7%。

並列摘要


When we think, feel and act, the neuron will transmit message in the brain. In recent decades, although we have understanding of functions of the brain and the cells, research of neuronal population response has not yet to complete. In order to understand whether it have a regular microinstructions that during different actions transmitting message between neurons in primary motor cortex. In the research, first obtain signals of M1 cortex of rat during different action. We will use nonlinear energy operator (NLEO), independent component analysis (ICA) and dynamic dimension increasing algorithm (DDIA) to extract features of signals. Then use time delay neural network (TDNN) to establish encoding system. The features of neuronal signals will encode Code, then we select the Code of all sections encode Symbolic. Finally, Symbolic from various kinds of the action signals comprise Command and observe the classification system. Through encode neuronal signals of various actions, we hope to understand whether several regular microinstructions exist among neurons of M1 while brain transmitting different action messages and understand the relationship between action behaviors and the functional areas of the brain. In our research, we use TDNN to build encoding system. If we not calculate the number of training samples in simulation respect, the accuracy rate for three actions are 72.7%, 72.7% 81.8%, respectively. In real capture signal respect, the accuracy rate for three actions are 33.3%, 74.1% and 40.7%, respectively.

參考文獻


[4] D.R Huw, P. Stefano, Z.Fashan, P. Giannni, W. Jack, Scannel, V. Dimitrios, J.M Rojer, J.T Martin and P.Y Malcolm, "Correlations, feature-binding and population coding in primary visual cortex", Neuroreport, 2002, pp1045-1050.
[5] C. Kayser, RF. Salazar and P. Koing, "Responses to natural scenes in cat V1", J Neurophysiol, 90: 1910-1920, 2003.
[6] L. Longnian, C. Guifen, X. Kun, A.Z. Kimberly, Z.Shuqing and T. Joe, "Large-scale neural ensemble recording in the brains of freely behaviong mice", Journal of neuroscience methods, NSM-4150.
[7] T. Kevin, "An interactive tool for visualization of spike train synchronization", Journal of neuroscience methods, Volume 191, Issue 1, 15 August 2010, Pages 145–148.
[8] A.L. Mikhail and A.L. Nicolelis, "Brain-machine interfaces : past, present and future", TRENDS in Neuosciences Vol.29 NO.9, pp:536-546, 2006.

延伸閱讀