Fruit Recognition Using Color Statistics

Because of aging and very low birthrate, agricultural labor force has become seriously insufficient. In order to solve this problem, researchers have developed a series of agricultural robots for different purposes, including fruit picking robots. For fruit picking robots, detection and recognition of fruits is an important task. Here a fruit recognition technique based on the statistical characteristics of HSV color was developed. First, the RGB color fruit images were converted into HSV color. Then the hue distribution of HSV color of fruit is approximated with a Laplace distribution. Further, this Laplace distribution can be adopted as the characteristic description of this fruit. The fruit was segmented out of the input image. If the segmented fruit image falls in some Laplace distribution with 90% confidence interval, then input fruit belongs to this special fruit. In practice, the Mahalanobis distance (MD) corresponding to the 90% confidence interval of the Laplace distribution for each fruit class was set as the reference evaluation. If the input fruit data has a smaller Mahalanobis distance than the reference evaluation, the input belongs to this type fruit. The experimental results have shown the good performance for this fruit recognition technique on different kinds of fruits.