The Darknet-19 CNN architecture of YOLOv2. The Darknet-19 architecture consists of 19 convolutional layers and 5 maxpool layers. The Reorg layer combines both high- and mid-level features for better detection accuracy. In order to increase accuracy in YOLOv2, the fully connected layers of the CNN are eliminated, and K-means classification is used for detection and classification (Redmon & Farhadi 2017).