“In response to the challenges in the field of initial tobacco grade detection, the research team has proposed an innovative initial tobacco grade detection network FTGDNet. This study aims to address the issue of difficulty in distinguishing initial tobacco leaves with high similarity but different grades, and promote the implementation of refined management of agricultural products. This network significantly enhances the feature extraction capability of the model by combining CSPNet and GhostNet as feature extraction networks. The explicit visual center bottleneck module embedded at the end of the backbone network achieves the fusion of global and local feature information, enhancing the richness of feature representation. Meanwhile, the multi receptive field feature adaptive fusion module utilizes attention feature fusion mechanism to effectively fuse feature maps of different receptive fields, highlighting effective channel information and improving the model's local receptive field capability. To solve the problem of decreased positioning accuracy, the research team designed a new positioning loss function MCIoU_Loss, which combines the area loss of predicted and real boxes to optimize the regression positioning process. In addition, the introduced rectangular similarity attenuation coefficient dynamically adjusts the similarity discrimination term between the real box and the predicted box during the training process, accelerating the model fitting. The experimental results show that FTGDNet performs well in the detection of initial tobacco leaf grades, with a validation accuracy of up to 90.0%, a testing accuracy of 87.4%, and a inference time of only 12.6 ms. Compared with various advanced object detection algorithms, FTGDNet has significant advantages in detection accuracy and speed, providing strong technical support for high-precision initial tobacco leaf grade detection. This research not only brings new breakthroughs in the field of grade detection for newly roasted tobacco leaves, but also provides important technical support for the fine management of agricultural products and the development of intelligent grading equipment.”