A method for detecting defects during laser additive manufacturing, comprising the following steps:
- B21) a first image is captured, said first image being the image, captured in the infrared spectrum, of an upper surface of a layer of powder exposed to laser scan;
- B23) said first image is processed using a first convolutional neural network of self-encoding type, in such a way as to produce a defect mask indicating a location of defects at the upper surface of said layer of powder;
- wherein during the capturing step B21, a second image of the upper surface exposed to laser radiation is also captured;
- the second image is an image of said surface, captured in the infra-red spectrum, wherein for each pixel, the intensity of the pixel represents the maximum luminous energy received by the pixel during a capturing period;
- and
- during the processing step B23, the processing carried out takes the second image into account as input, in addition to the first image.