Leaf image recognition and classification based on GBDT-probabilistic neural network

Abstract

Abstract In this paper, the binary images of 100 kinds of leaves are used for leaf recognition. Firstly, we screen 35 important features and use the grey clustering analysis to establish the quantitative feature system of leaves. Then we use the gradient descent tree algorithm (GBDT) to select core features and use probabilistic neural network (PNN) to recognize and classify leaves, constructing a hybrid GBDT-PNN model. In the end, we obtain the classification results of leaves to evaluate model performance and the influence of core features on the model. The results show that the accuracy rate of GBDT-PNN model using 12 core features is 92.75%. And the accuracy rate with all 35 features is 93.5%. It illustrates that the model has great performance and core features have high influence on the model. By comparing with other commonly used deep learning algorithms and models, it is verified that the GBDT-PNN image recognition and classification model is effective and has high accuracy.