The invention relates to a method for detecting a platelet object and for determining a number of platelets in the platelet object in a sample, a platelet object having at least one platelet. In particular, the sample comprises whole blood. At least one image is recorded from the sample, in particular the image showing a platelet object, the at least one image being generated in particular by means of an imaging flow cytometer. The at least one image of the sample is examined with the aid of a trained convolution network, which is executed in particular on a computer device. The trained convolution network is set up to recognize the platelet object on the at least one image, the trained convolution network being configured to determine and output a number of platelets in the recognized platelet object.