关键词: 图像识别, 深度卷积神经网络, 空间映射网络, 密集神经网络, 树种识别

Abstract: Image recognition is one of the important branches of computer vision, which has important theoretical and practical significance. In recent years, image classification methods represented by deep convolutional neural networks have been successfully applied in various fields. In view of the sensitivity of neural networks to input data and the long training time, and combining the characteristics of Spatial Transform Network（STN） and Dense Neural Network（DenseNet）, a new network structure, namely ST-DenseNet is proposed. The network solves the problem of data sensitivity and improves the effect of image recognition by way of normalized processing and immutability of the input image. On Leafsnap, the open data set of tree leaves, the experimental results of identification accuracy rate of 90.43%, recall of 87.75% and F-Measure of 89.07% are achieved. The model ST-DenseNet is obviously superior to other network models.

Key words: image recognition, deep convolutional neural network, space mapping network, dense neural network, tree species identification