1. Tobacco top flowering period recognition and detection model based on improved YOLOv4 and Mask R-CNN network.
- Author
-
Weihua Qin, Longfei Wang, Xiangguo Cheng, Chaofei Yang, Shengli Chen, Lianhao Li, and Chenhui Zhu
- Subjects
- *
NICOTIANA , *IMAGE recognition (Computer vision) , *WATER levels , *DISEASE management , *AGRICULTURAL productivity , *DEEP learning - Abstract
Tobacco is an important crop. During the flowering period of tobacco cultivation, the top part of plant contains high levels of water and impurities, which affect the taste of tobacco. Therefore, topping operations are carried out. Currently, deep learning is widely used in modern agricultural image recognition, pest and disease management, and other scenarios. However, due to the cost and technology limitations, its application in agricultural production is relatively low. This study proposed a tobacco flowering period recognition technology based on deep learning. An attention mechanism was introduced into the backbone residual module of deep learning to collect tobacco flower characteristic data at different stages to realize tobacco flower identification. Meanwhile, considering the large computational load of the tobacco flower detection model, a cascade strategy and transmembrane segment splitting were introduced to optimize the repeated features in backpropagation and improve the segmentation effect of the model. In the detection model test, compared with the convolutional and the unimproved models, the proposed model improved the recognition effect of tobacco flowers by 45.65% and 28.65%, respectively. In the multi-model comparison, the accuracy of the proposed model in dark light scenes was 0.853, which was better than other models. In the segmentation model test, the proposed model performed best in terms of confidence, time consumption, and number of frames. Improving the segmentation model during training took 3.5 seconds and detected 5 flowers, which was significantly better than the improved tobacco flower detection model and taking less time. The proposed tobacco flower detection model showed good application effects in tobacco flower detection scenarios, took less time, and had high detection accuracy, meeting the production requirements of modern farmers. The research provided important technical references for the management and cultivation of tobacco varieties. [ABSTRACT FROM AUTHOR]
- Published
- 2024