Flue-Cured Tobacco Grading Method Based on a Convolutional Neural Network

Artificial flue-cured tobacco grading often has problems such as unstable grading results and low qualification rates, which lead to waste of tobacco resources and conflicts between tobacco farmers and collection stations. To rapidly and accurately recognize tobacco grades, a tobacco grading network called Tacc-Net was proposed. First, the convolution layer, pooling layer, batchnorm layer and SENet are used to build the Tacc-Block. Second, three Tacc-Blocks are connected to extract features, followed by the full connection layer to compose Tacc-Net for classification. Finally, the training was carried out in the tobacco leaf image dataset with 10 grades and more than 4000 images to realize the grade of tobacco leaf images. Compared with the classical neural network, this method has higher accuracy and can classify tobacco images of different grades more accurately.