Abstract: To automatically recognize and count tobacco plants in fields, an automatic tobacco plant recognizing and counting method was proposed based on U-Net and Simple Linear Iterative Clustering (SLIC) superpixel segmentation. Firstly, the area of tobacco fields was extracted by training semantic segmentation network U-Net, then excess green difference index (EGDI) algorithm was constructed to remove any weeds and to extract tobacco plant coverage, and further the tobacco plant coverage was segmented using the SLIC algorithm. Finally, the segmented tobacco plants were automatically labeled and counted. The results showed that the Mean Intersection over Union (MIoU) obtained by the U-Net network for tobacco field extraction reached 98.24%. Average overall accuracy of 99.21% and average precision of 93.42% were obtained by using this method to extract tobacco plants, which indicated the feasibility of this method for large-scale tobacco plant extraction and counting.