Abstract: The plant number is an important field phenotypic trait in monitoring crop growth and estimating output. In order to establish an efficient tobacco plant number automatic counting technology, an optimized tobacco plant detection model YOLOv7-Sim based on YOLOv7 is proposed to solve the miss detection problem of small targets in UAV remote sensing images. First, the SimAM attention mechanism is introduced to enhance the aggregation ability between image features, and a small target detection layer is added to strengthen the detection ability of small targets, then EIOU is used to optimize the positioning loss function, and finally, a slicing strategy is used to solve the problem of small target sampling loss in large image detection. The experimental results on the Vis-Drone2019 dataset and the UAVTob dataset constructed in this study showed that the mean average accuracy rate mAP@0.5 of the detection results was increased by 0.3% and 6.3%, and the mean average accuracy rate mAP@0.5:0.95 was increased by 0.6% and 18.3%, which reflected the superiority of YOLOv7-Sim algorithm for tobacco detection in UAV remote sensing images.