Substrate defect inspection apparatus, substrate defect inspection method, and storage medium

A substrate defect inspection apparatus comprising:

• a feature region extraction part configured to extract a feature region from an image, which is obtained by capturing a substrate, using a deep learning system;
• a two-class classification part configured to output binarized data for defect classification estimation indicating a defect region in which a defect is generated and a surrounding region of the defect region based on an extraction result obtained by the feature region extraction part;
• and
• an estimation part configured to estimate a type of defect based on an attribute of the defect region extracted from the binarized data which is output from the two-class classification part,
• wherein the feature region extraction part is configured to receive image data corresponding to the image, of which a pixel value is assigned to each of a plurality of pixels, as input data, and the feature region extraction part includes a plurality of convolution layers, each of which is provided with a plurality of filters, the plurality of pixels being arranged in a matrix form in each of the plurality of filters, and
• wherein the two-class classification part is configured to use data obtained by associating pixels obtained by a convolution process in each of the plurality of convolution layers with calculation values which are results of the convolution process, to classify whether each of the plurality of pixels is a pixel corresponding to the defect or a pixel corresponding to a normal portion based on an arrangement distribution of the calculation values of the plurality of convolution layers for each of the plurality of pixels, and to obtain the binarized data for the defect classification estimation.