Research on Synchronous Recognition Method of Multiple Tobacco Leaf Grades Based on MDCNet

In order to improve the actual efficiency of tobacco collection, it is urgent to transform the single tobacco leaf level recognition into multiple tobacco leaves. Due to the lack of obvious difference in the features of adjacent grades of tobacco leaves in multiple tobacco leaf images, and the large difference in the position and direction of tobacco leaves, the existing synchronous recognition methods of multiple tobacco leaf grades have low detection and positioning accuracy. In view of this, this paper proposes a synchronous recognition method of multiple tobacco leaves level based on MDCNet. Firstly, based on Mobilenet-v3-YOLOv7, a deformable convolutional DCN that can adapt to target shape changes was introduced to improve the robustness of the model. Secondly, the convolutional attention module (CBAM) was embedded in the neck network of the model to make the model pay more attention to the important features of tobacco leaf, suppress the general feature information such as background, and improve the recognition accuracy. Then, SIoU Loss was used as the localization loss function to improve the localization accuracy of the network. Finally, the transfer learning strategy was used for model pre-training to accelerate the convergence speed of the model. The experimental results show that the mAP value on the test set is 10.5% higher than that of the original Mobilenet-v3-YOLOv7. The proposed method is compared with the other two mainstream object detection algorithms (Mobileone-YOLOv7 and Ghostnet-YOLOv7). It also shows significant advantages in accuracy and detection speed. Therefore, the proposed method can meet the practical application requirements of classification recognition for multiple tobacco leaves synchronously.