This document summarizes an unsupervised project to correct training labels for image segmentation using clustering algorithms. The objectives were to preprocess image data without labels and investigate unsupervised segmentation techniques. The methodology involved applying DBSCAN, K-means, mean shift, N-cut and watershed clustering to satellite images, then using region growing for post-processing. Mean shift performed best overall at segmenting land use types. Future work will tune hyperparameters to improve the region growing accuracy compared to using raw images.