Detecting Mines in Minefields With Linear Characteristics

摘要

We consider the problem of detecting minefields using aerial images. A first stage of image processing has reduced the image to a set of points, each one representing a possible mine. Our task is to decide which ones are actual mines. We assume that the minefield consists of approximately parallel rows of mines laid Out according to a probability distribution that encourages evenly spaced, linear patterns. The noise points are assumed to be distributed as a Poisson process. We Construct a Markov chain Monte Carlo algorithm to estimate the model and obtain posterior probabilities for each point being a mine. The algorithm performs well on several real minefield datasets.