A method of green litchi recognition in natural environment based on improved LDA classifier

Abstract

Green litchi is always difficultly recognized by picking robot under the natural environment because of its similar color feature with background. A method of green litchi recognition based on improved LDA classifier is proposed by this paper. The color features of RGB components of litchi images were firstly analyzed. Then a linear discriminant analysis (LDA) method used for extracting convolutional features for classifying pixels of image was proposed to train the convolution kernel based on 1600 sample pixels. Simultaneously, an idea of ‘maximal margin’ from SVM to calculate the threshold of LDA classifier was introduced, and the corresponding threshold calculation method was put forward. The AdaBoost method was used in integration of a strong multiple LDA classifier. After classifying pixels, the Hough transform circle detection was used to locate the fruit of litchi by the sphere shape feature. Experiments with the proposed method show that green litchi recognition precision rate is 80.4% and the recall rate is 76.4%. This study provides technical support for the visual identification of green litchi and even other green fruits in natural environment.