Betel nut classification algorithm based on improved Xception

Abstract

Objective: In order to reduce the manual demand of betel nut classification improve the accuracy of betel nut classification and reduce the size of classification model. Methods: Expanded the input layer of Xception as the feature extraction backbone network. Added a dual-channel sequeeze and excitation module after the feature extraction network. Used the ELU activation function instead of ReLU. Used the data enhancement to expand the dataset of betel nuts, divided the dataset into training sets, validation sets and test sets in 9∶3∶1, and trained the improved Xception models. Results: When the improved Xception was used to classify 1 100 betel nut images in the test set, the classification accuracy reached 99.182%, and the model size was 15.7 MB. Conclusion: The improved model can meet the accuracy requirements and model size requirements for betel nut classification.

Keywords

• xception
• dual channel attention
• elu
• accuracy
• light weight