Betel Nut Classification Method Based on Transfer Learning

Betel nut is one of the main chewing hobby in the world. The traditional methods of betel nut classification are mostly based on handcrafted features, and the classification stability is poor. To solve this problem and reduce labor costs for manual extraction of image features, this paper applies convolutional neural network to betel nut, and adopts parameter transfer strategy to realize transfer learning. The dataset of betel nut was constructed by shooting different types of betel nut with betel nut grading equipment. Aiming at the problem of insufficient samples, some parameters of AlexNet are transferred and fixed, new fully connected layers are build. Make full use of betel dataset to train and test the network. The experimental results indicate the proposed method is effective and low labor cost.