I am working in a project with identifying cucumber fruits in a green house environment. i have followed some research articles and found K-means clustering for segmentation process. defining the initial solution is measured in randomly input data and the results will change time to time. Is there any better method to identify initial other than using random data.