Classification of Flammulina Velutipes Heads via Convolution Neural Network

As one of the most common fungal food, Flammulina Velutipes (FV) is an important source of food in China. At the same time, the popularization of industrialization has greatly improved the yield of FV. However, in its industrialized production, there are still many artificial factors in the selection and classification of the FV. It will bring about mistakes after the workers' long hours working, which will increase the classification error rate, low production efficiency, thus resulting in low production of FV and damage to the factory interests. In order to solve these problems, we use machine instead of workers to complete the classification of FV via its heads by adopting the currently popular Deep Learning (DL) of computers. And the corresponding methods are as follows: (1) Collect the data of FV heads and then make a dataset according to the classification standard proposed by the "FV Factory" in this paper. (2) Preprocess the image, augment and normalize the dataset. (3) Retrain dataset of the FV heads respectively in the following three convolution neural network models as in Alexnet, Vgg-16, Resnet-50 as well as an improved Resnet-50 one by using the Transfer Learning method. (4) By analyzing and comparing the three network training models, this paper comes to a conclusion that the results obtained by Data Augmentation in the improved Resnet-50 model with a test accuracy of 79.9%, are superior to that of the other neural networks.