A Grading Method of Ginseng (Panax ginseng C. A. Meyer) Appearance Quality Based on an Improved ResNet50 Model

In the academic world, ginseng (Panax C. A. Meyer) has received much attention as most representative element of Chinese medicine. To address lack traditional algorithms in identification appearance quality and further improve manual on ginseng, we propose a grading method based deep learning, taking advantage benefits learning image identification. Firstly, substituted LeakyReLU for conventional activation function ReLU to enhance predictive power model. Secondly, added an ECA module residual block, which allowed be focused input object capture more precise detailed features. Thirdly, used focal loss solve problem imbalanced dataset. Then, self-constructed dataset was processed with data enhancement divided into four different classes ginseng. The trained model transfer finally obtain best applicable quality. experiments showed that, compared classical convolutional neural network models VGG16, GoogLeNet, ResNet50 Densenet121, proposed reported performance, its accuracy test set high 97.39%, value 0.035. This can efficiently classify significant field