The invention provides a leukocyte classification method and system based on a convolutional generative neural network. The leukocyte classification method based on the convolutional generative neuralnetwork comprises the following steps: inputting a leukocyte image into a generator network, selecting a classified region of interest, and carrying out normalization processing to generate an initialized cell contour; inputting the initialized cell contour image into an initialized discrimination network to obtain an initialized pixel-level cell classification result; taking the cell classification result of the initial pixel level as an initial value of a deep adversarial generation classification network, and extracting contour and edge information as edge characteristic values; adding anattention mechanism module in the generator network, and classifying the edge characteristic values to obtain a pixel-level cell reclassification result; and inputting the pixel-level cell reclassification result into a support vector machine classifier cascaded with a generator network to obtain an accurate image classification result.