The algorithm helped to narrow the area of search for unexploded bombs by using satellite images

American scientists have developed a method that improves the efficiency of the search for unexploded bombs remaining in the territory of Cambodia since the Vietnam war. First, a two stage algorithm notes on satellite images of the crater from the exploded bombs, distinguishing them from similar items of relief, and then researchers on the basis of the analysis of these data to determine region around the craters, which contain 98 percent of both types of bombs. Researchers for example, a region of 100 square kilometers compared the results of the algorithm, information about the bombing of the U.S. air force and the results of demining in the area, and came to the conclusion that about half of the bombs dropped on this area remains unexploded and pose a danger. Article published in PLoS ONE.

Carpet bombing cause not only a direct attack on the enemy positions during the war, but often leave behind a vast territory with a large number of unexploded ordnance. This is especially true of cluster bombs — they are after a reset are divided into tens and hundreds of sub-munitions, many of which often remain unbroken. In fact, they become analogues min, and then many decades undermined peaceful people during field and other activities.

Cambodia was heavily affected after the Vietnam war in which the territory of modern Cambodia were dropped about polumilliona tons of ammunition, and just about every second of the Cambodian countryside are faced with the damage caused by unexploded bombs. Usually demining is done manually with metal detectors, which a lot of time time is spent including the test areas which do not turns out to be unexploded bombs. Erin Lin (Erin Lin) and her colleagues from Ohio state University have developed a method to significantly narrow the area for searches based on satellite imagery.

The algorithm works in two stages. In the first stage, he is using the support vector machine and histogram oriented gradients lays out on fragments of satellite images round elements with spectral values that are similar to bomb craters from the training sample. In the second phase, the selected crater-candidates are separated from surrounding objects, and their size and color options are analyzed by two methods, and the results of these methods combine into a single vector and random forest assigns a true or false value.

Leave a Reply

Your email address will not be published.