To improve the anti attack ability and adaptability of the watermarking scheme, a blind watermarking scheme based on GoogLeNet is proposed. Firstly, the proposed network is relatively simple, and the deepest path (that is, the path through preprocessing network, embedding network and extracting network) only contains 17 layers. The resolution of the host image is maintained by increasing the watermark resolution in the watermark preprocessing network, thus enhancing the transparency of the watermark. The average pooling is used in the watermark preprocessing network to combine the binary value of the watermark data with the host image properly, so it can enhance the transparency of the watermark. Finally, The extractor uses cross entropy as the loss function to achieve the training balance between the embeder and the extractor. The experimental results show that the performance of the proposed scheme is excellent, the watermark capacity is 0.0038, and the average PSNR in the dataset is 40.57 dB. The performance under meaningful attack is better than other advanced methods.