Blindly Selecting Method of Training Samples Baded Data''s Intrinsic Character for Machine Learning

摘要

The supervised machine learning is the main analyzing method for the object recognition, but, when we analyze the multidimensional data using the supervised learning method, how can we get the training data from the data itself without other previous knowledge? Based on the intrinsic assembling feature of the multidimensional data, we present a method to select the training samples for machine learning. Firstly, we calculate each dimension''s probability density estimating (PDE) to find the easily separable dimensions of the multidimensional data, then gain the smallest representative sample sets of all objects through intersecting the data of the same object of each easily separable dimensions, and get the object''s number and the training data sources for the machine learning at the same time; secondly, train the neural network ensembles using the data selected from the representative sample sets to label the other data. Lastly, we analyzed the hyper-spectral images to detect red tide using this method, which proved this method could recognize the red tide effectively