In the medical field, vein imaging technology is used to assist intravenous injection and the treatment of venous diseases. The reflection-type vein imaging device has a significant advantage in clinical vein locating with the advantages of non-contact and portability. Aiming at the problem that the existing image processing methods are difficult to extract the vein patterns accurately in the reflection-type vein image because of the noises and weak vein features of the reflection-type vein image, a vein pattern extraction method(Reflection-type Imaging Vein Extractor, RIVE) based on convolutional neural network is proposed to improve the accuracy of vein pattern extraction in reflection-type vein images. First, a neural network is trained with the transmission-type vein images and labels; Then the trained network is used to extract the vein patterns in the reflection-type vein images; Finally, comparing the vein patterns extracted from the reflection-type vein images with the results of the transmission-type vein imaging and evaluating the performance of the new method based on the vein extraction rate. The experimental results demonstrate that the vein extraction rate of the RIVE can reach 63.2%, which is a 23.7% improvement compared with the traditional method. Therefore, the proposed method can extract vein patterns more accurately in reflection-type vein images, which is of great significance in clinical vein imaging technology.