The Geometrical Classification Of Neuronal Soma Based On Integrated Learning Of Spectral Clustering

The brain is the most complex organs with very complicated structure and many advanced features, which it can memory, think and creative consciousness and emotion. At present, we can achieve a complete 3D brain-wide cellular in the mouse brain by technology methods and it is good for carrying out “Human Brain Project”.In this paper, we study the geometrical classification of neuronal soma. Firstly, we change the polar coordinate data into 3D data in order to draw the network chart and make a classification in advance for clustering experiment basis according to the existing knowledge of biology. Secondly, we construct important features and understand the data distribution through the histograms and density curve of data. Next, we reduce the dimension of cell features through extracting main features and eliminating redundant features because it can decrease the complexity of algorithm, weak the coincidence of features and improve the accuracy of clustering. So we use principal component analysis and spectral decomposition to deal with it. In the latter, we use the second method to take k-means, fcm, clustering based on density, hierarchical clustering experiments because it has a bigger accuracy. What it is important; we study clustering methods and find that integrated learning integrates many advantages of algorithm for the study about clustering methods compared with the single algorithm. In the article, we take voting with the weight as consensus function. We randomly select one third of data from original data as new sample. We take four kinds of clustering experiments on the new sample and assign the weight to four kinds of clustering methods according to accuracy. Finally, we use spectral decomposition to deal with the original data, and we take clustering integration experiments on the set made by new features through single algorithm and integrated method. In the end, the result of integrated learning method is satisfying. The result shows that this method is effective to identify different cells and classify different brain areas. We can use the method to find abnormal cells to carry out medical research.In the article, the method based on integrated learning of spectral clustering can be used to extract features and reduce the dimension. Not only is it a kind of effective method to classify different cells, but also it can reduce the algorithm complexity and save the storage space and time, which makes it possible to deal with big data.