An experimental research for automatic classification of unbalanced single-channel protein sub-cellular location fluorescence image set

To Model the protein sub-cellular localization pattern which responds to drugs' treatment is an important problem for medical applications named as high content screening (HCS) with bio-imaging and machine learning. Traditionally, at least, three channels' images have to be retrieved for auto-focusing and segmentation on DNA channel, background correction on auto-fluorescence channel and protein sub-cellular localization analysis on GFP channel which is a time consuming and error accumulating procedure. An automatic classification of Single-channel Protein Sub-cellular Location Fluorescence Images without segmentation is desired for speeding up the pattern analysis. But in the real world, the data imbalance often occurred and affected the classification accuracy. By now there are a variety of approaches proposed for improved the classification accuracy including the re-sampling and reweighting. This experiment engaged in comparing existed solutions for further research.