Method for detecting object image using hyperspectral imaging by band selection

A method for detecting object image using hyperspectral imaging by band selection, comprising steps of:

• acquiring the hyperspectral image information formed by converting a reference image to a hyperspectral reference image, said reference image including one or more object reference image and a background reference image;
• an image extraction unit extracting an input image to a host, said input image including one or more object-under-test image and a background image;
• said host converting said input image according to said hyperspectral image information for giving a hyperspectral image;
• said host analyzing said hyperspectral image for giving a plurality of first hyperspectral characteristic values;
• said host performing band selection on said plurality of first hyperspectral characteristic values according to a cell and performing principal component analysis to simplify said hyperspectral image and generate a plurality of second characteristic values correspondingly;
• said host performing one or more layer of convolution calculation on said plurality of second characteristic values according to a plurality of kernels to filter out said background image and give a convolution result, and giving one or more selected image according to said convolution result and said one or more object-under-test image, said plurality of kernels including a plurality of selected characteristic values and a plurality of peripheral characteristic values, said one or more object-under-test image including a plurality of peripheral images and said one or more selected image, said plurality of peripheral images surrounding said one or more selected image, said one or more selected image corresponding to said plurality of selected characteristics, and said plurality of peripheral images corresponding to said plurality of peripheral characteristic values;
• said host generating one or more prediction box according to the edge of said one or more selected image;
• said host extracting a bounding box of said input image for comparing a first central point of said prediction box with a second central point of said bounding box and giving a center displacement between said prediction box and said bounding box, and said bounding box corresponding to an edge of said input image;
• said host performing a regression calculation according to said center displacement and giving a regression result;
• said host aligning said object-under-test image according to said regression result and said prediction box so that said selected image can move towards said second central point as said first central point moves towards said second central point;
• said host matching and comparing said aligned object-under-test image with one or more sample image for generating a comparison result;
• and
• said host classifying said input image as a target-object image according to said comparison result.