关键词: K-means算法, Grabcut算法, 特征提取, PCA, 距离集成kNN, 果蔬识别

Abstract:

In view of the poor fruit and vegetable recognition rate of traditional algorithm, a principal component analysis and distances ensemble for K-nearest neighbor combine recognition method was proposed. This method realized the recognition of fruits and vegetables from feature description, feature dimension reduction and classifier design. For the problems of uneven light and shadow in the picture of fruits and vegetables, K-means clustering was used to divide the picture by combining with the second watershed. Aiming at the problem that the recognition rate of fruit and vegetable recognition model is not high, the color and texture features of the extracted fruit and vegetable images were composed into the feature matrix. The matrix was normalized by the PCA and integrated kNN algorithm and the dimension reduction was obtained to get the low-dimensional classification feature and to finally realize the classification of fruit and vegetable agricultural products. The experimental results showed that the algorithm had the highest recognition rate of 92.6% in the category of fruits and vegetables, and was robust to the changes of illumination and angle of view.

Key words: K-means algorithm, Grabcut algorithm, feature extraction, PCA, distances ensemble kNN, fruits and vegetables recognition