On Application of Convolutional Neural Network for Classification of Plant Images

Now, the convolutional neural network (CNN) represents a powerful visual model that demonstrates high performance in solving object recognition problems. However, due to the insufficient amount of training samples of image data, an efficient application of CNN models still remains a challenging problem. In this paper, CNN architecture has been researched to increase its performance capability for plant image classification. The open database of plant images, consisting 12 various species, is used to the training of CNN models. In order to better deal with the images information of plants, the CNN model includes maximum pooling layers and dropout layers. The experiments on benchmark dataset have demonstrated that the proposed CNN architecture considerably outperforms other state-of-the-art methods. The high accuracy of classification makes the model a very useful to support an integrated plant identification system to operate in real conditions.