METHOD FOR DETERMINING DEFECTS IN THE IMAGE OF COMPOSITE PRODUCTS Russian patent published in 2023 - IPC G01N23/46 

FIELD: determining defects in the image of composite products.

SUBSTANCE: invention relates to a method for determining defects in the image of composite products. The technical result is achieved by uploading an image of the finished product and selecting areas for defect detection; by means of a trained deep neural network, each pixel is determined to belong to the "defect" or "no defect" class and a segmentation mask is obtained; post-processing is carried out, in which a segmentation mask and an image of the finished product are received at the input of the module and filtering is carried out assigned pixels to the "defect" class by the signs images of the finished product, namely the area and linear dimensions of connected areas of defect pixels, deviations of the average value of defect pixels from the average value of pixels in the area of defect detection, and also aggregate filtered pixels by the depth of the object and isolate volumetric defects from them; calculate the characteristics of the selected volumetric defects by measuring the linear dimensions of connected pixel areas defects, and classification of defects based on the measured linear dimensions of connected areas of pixels of defects.

EFFECT: technical result is to increase the accuracy of determining the defects of composite products.

8 cl, 3 dwg