关键词: 烟叶主脉, 轻量化, 机械化收获, 精准识别, 边缘化部署

Abstract: In order to accurately identify the main vein of tobacco leaf, realize mechanical grasping and reduce the rate of grasping damage, an improved lightweight tobacco leaf main veins recognition model based on YOLOv7—tiny was proposed. Firstly, the original trunk feature extraction network is replaced by a more lightweight MobileNetV3 based on YOLOv7—tiny network, the default h-swish activation function in the module is replaced by ReLU activation function. Then, the common convolution of the neck is replaced by a lightweight GSConv and a Slim—Neck design is adopted to compress the channel of the model and eliminate the redundant feature redundancy in order to lighten the network structure. At last, the SIoU loss function was introduced to reduce the loss value of the model and enhance the fusion ability of the model to the main vein of tobacco. The results showed that the map value of the improved model on the tobacco leaf dataset was 91.3%, at a cost of only 1.6% loss, the parameter quantity was reduced by 51.1% compared with the original model, and the computational load was 4.3 G, only 32.6% of the original model (13.2 G). Compared with YOLOv5—s (16.5 G), YOLOv6—n (11.4 G), Yolox—s (26.8 G), YOLOv8—n (8.7 G), and YOLOv9—t (7.7 G), all of them were improved. The improved model can be deployed in the marginal equipment with scarce computing resources, which provides some technical support for the mechanized harvesting of tobacco leaves.

Key words: tobacco leaf main veins, lightweight, mechanized harvesting, accurate identification, marginal deployment