Searching the Effects of Image Scaling for Underground Object Detection Using KMeans and KNN

With the increase in conflicts between countries, underground object detection has become a serious problem today. One of the commonly used technology is Ground penetrating Radar (GPR). There are different variants of GPR devices, but usually they have an array of sensors, which emits electromagnetic waves, and then, collect the reflecting data through its sensors. The signals travel with different speeds in different mediums which yield some beams together and forms holes or peaks in the signal. According to the properties of searching object, depth of the object, or the soil properties, GPR produces different signal signatures. These signals are used to detect searching underground objects. In order to have a detailed view, more sensors are used, and the frequency is changed to be able to detect deeper objects. Increasing the signal quality causes many algorithms to fail or slow down seriously. On the other hand, underground object detection needs fast and accurate detection. In this paper, we have analyzed the effects of image scaling on object detection using KMeans and k-Nearest Neighbor algorithms (kNN). According to our experiments, even after serious image scaling, the results have not change much while increasing the runtime performance and memory efficiency significantly.