The present application relates to the field of
machine vision defect detection, and discloses a COF substrate ink defect
visual detection and analysis
system, which comprises a topology prior module for calculating a
linear polarization degree matrix and generating a topology
mask matrix in combination with a CAD drawing; a variable-frequency projection module for giving high-frequency parameters and low-frequency parameters to high and low reflection areas respectively based on the topology
mask matrix, and generating a fringe sequence; a collection and
demodulation module for acquiring a polarization image by projecting the fringe sequence,
processing light intensity values and extracting an
alternating current matrix; a characteristic
tensor module for calculating a polarization matrix and a frequency-variable
depolarization rate, generating a weight matrix based on the topology
mask matrix, and fusing and constructing a descriptor
tensor; and a defect classification module for searching the descriptor
tensor to extract surface-type and internal-type defects, combining the weight matrix to extract penetrating defects and outputting labels. The present application blocks bottom
metal reflection
crosstalk through
spatial frequency adaptive modulation, fuses multi-dimensional physical parameters and prior weights to remove interference artifacts, and realizes accurate classification of cross-layer defects of composite substrates.