Multi-Period Tobacco Plant Identification Based on Drone Image and Improved Yolov7 Model

Taking Pingdingshan and Xuchang areas as an example, an improved YOLOv7 model was used to accurately and dynamically identify the number of tobacco plants based on the tobacco images obtained by UAV during the flourishing period and early maturity period. The results show that after adding Focus to MPConv module, introducing SimAM parameterless attention mechanism and improving it with WIoU loss function, YOLOv7 model has faster convergence speed and better generalization performance during training. The average accuracy rate, recall rate and accuracy rate of tobacco plant number identification reached 98.62%, 97.20% and 92.85%, respectively, which were increased by 0.32%,1.52% and 0.85% compared with the original model, and the F1-score reached 0.95, 1% higher than that of the original model, which was better than YOLOv5 and YOLOv4. Traditional target detection models such as SSD and Faster-RCNN and YOLOv4+CBAM models; The identification accuracy of the improved YOLOv7 model was 99.3%, 99.0% and 98.7% for the number of tobacco plants from the cluster planting stage to the early mature stage, and the identification effect was the best in the flourishing period.