Capacitance detection report document retrieval method based on text image alignment

The invention discloses a capacitance detection report document retrieval method based on text image alignment. The method comprises the following steps: extracting a semantic feature # imgabs0 # of a document image and a semantic feature # imgabs1 # of a keyword to be searched; the image semantic feature # imgabs2 # and the text semantic feature # imgabs3 # are fused to obtain a probability feature graph # imgabs4 # after self-attention alignment correction; based on the probability feature graph # imgabs5 #, obtaining a binary image Bm of the document image by setting a threshold Hr, and obtaining a textbox position; performing feature selection by setting a threshold value Ht based on the probability feature graph # imgabs6 #, extracting a feature region of each text line, and constructing a selection feature set of all existing texts on the probability feature graph # imgabs7 #; and performing GLR prediction based on the selection feature set to obtain position information containing a keyword to be searc