The realistic images generated by GANs(Generative Adversarial Networks) enrich people’s lives, but they also pose serious threats to personal privacy and society, and it has become essential to study algorithms that can accurately detect GAN-generated images. Existing studies use artifacts to detect GAN-generated images, but the artifacts present in different GAN-generated images vary widely, and thus the cross-model generalization performance of such algorithms is weak. In this thesis, we propose the MaxPix, a new algorithm based on the combination of statistical features and deep learning techniques, for generating image detection. Firstly, MaxPix obtains the filter map of the image by designing the MaxSel filtering algorithm and then designs MA Block embedded in ResNet (Residual Network) to obtain MResNet. MaxPix finally utilizes MResNet to extract features from the filter map to detect GAN-generated images. Experimental results on publicly available datasets such as Wang and Faces-HQ show that the detection accuracy of MaxPix reaches 85.9% and 99.6% on average, which improves 7.6% and 10.2% relative to state-of-the-art algorithms such as the NAFID and the GocNet. Thus MaxPix has strong cross-model generalization performance.