Proposed methods for error detection in GPS data using shape clustering

GPS technology is becoming a big part of life, especially in agriculture. Methods exist to convert this data into geometric shapes representing the underlying field, but these processes are not perfect and typically produce holes in the interior of the geometry. Some of these holes are erroneous while others are legitimate, perhaps representing a tree or boulder. Therefore, a method is needed for classifying a hole as erroneous or legitimate.

The field of shape analysis has developed rapidly over the past several decades and allows for the comparison of multiple shapes. This includes methods for calculating distances between shapes and averages of sets of shapes. This study explores the use of shape analysis and clustering to propose a method of classification for these holes and a corresponding hypothesis test. Using this method, two sets of simulations are run and three different types of rejection regions are calculated. The resulting error rates are explored to show that the method performs quite well and warrants further study.