The paper proposed a method for segmentation of finger veins based on U-net and ResNet. The deep residual network is used to replace the feature extraction part of U-net. Considering the risk that some vein information may be lost in the vein extraction of finger vein by using pooling layers, Dilated Convolution is used instead of traditional convolution, which can increase the receptive field without pooling and extract the vein information from the finger vein image better. To further improve the expression ability of the model, Mish activation function is used instead of the ReLU activation function, which can make the extracted vein patterns more continuous. The experimental results show that the proposed novel VeinsegNet network in this paper has a excellent vein segmentation performance on the multi-modal vein dataset, outperforming other deep learning networks in terms of the mean intersection over union (i.e. MIOU). The segmentation of the finger vein image is shown to be greatly beneficial to the person identification and certain disease recognition. KEYWORLD: Deep learning; Image segmentation for finger vein; Applications to person identification and certain disease recognition