Recognition of Macrofungi by Convolutional Neural Networks with Attention Mechanism

Abstract

Macrofungi refer to fungi with large fruiting bodies. In terms of systematic classification, the species of macrofungi come from the Discomycetes of Basidiomycota and Ascomycota. As the decomposer of nature, macrofungi play a key role in the carbon cycle of the earth and have strong research significance. However, there are many kinds of macrofungi, with a population of more than 10 000. It requires profound professional knowledge to identify them, which is a waste of manpower and material resources. In this study, we creatively proposed a new method based on convolutional neural network (CNN) to recognize macrofungi images. By combining the attention mechanism with the lightweight backbone model densely connected convolutional network (DenseNet), A-DenseNet model is proposed to complete the efficient classification task of macrofungi. The recognition accuracy of our model on the public macrofungi dataset reached 84.3%, and the recognition accuracy on the local macrofungi dataset reached 82.2%, which illustrates the excellent performance of our network in the macrofungi recognition task. This method is an effective supplement and reference for macrofungi classification task.