Research on Classification of Wild Fungi Based on Improved Resnet50 Network

摘要

Because of the wide variety of wild fungi, it is so hard to distinguish whether it is edible or not that wild fungi poisoning incidents occur frequently. In order to help people effectively identify wild fungi species and reduce the occurrence of bromatoxism, a method for wild fungi classification based on the improved ResNet50 is proposed. The ResNet50 transfer learning model is used to reduce training time. Using the spatial pyramid pooling (SPP) layer, the network can use the original image with more detailed features for training, and retain the discriminative image features as much as possible. Besides, EvoNorm-S0 is used to solve the impact of the decreasing model accuracy with smaller batch size because of using the original image size to train the model. The improved model has an accuracy rate of 96.54%, which is 2.21% higher than the original ResNet50 network. The experimental results show that the proposed method has a better effect on wild fungi classification, which is better than the original model and the other four classic models. It has achieved a better recognition results and makes the practical application for wild fungi classification possible.