ANTENNA DOWNTILT ANGLE MEASUREMENT METHOD BASED ON DEPTH INSTANCE SEGMENTATION NETWORK

摘要

An antenna downtilt angle measurement method based on a deep instance segmentation network. The method comprises the following steps: shooting a full range of antenna video by an unmanned aerial vehicle; and transmitting the antenna video to a server end in real time, and the server end measuring an antenna downtilt angle in real time by means of a deep learning algorithm. The deep learning algorithm comprises: a feature extraction network module, an instance segmentation module, an antenna candidate frame module, and an antenna downtilt angle measurement module, wherein the instance segmentation module comprises the following steps: mapping a region of interest directly to a feature map; dividing a candidate region into k*k units, calculating four fixed coordinate positions for determining the center point of a unit for each unit, and calculating the values of the four fixed coordinate positions by means of bilinear interpolation; and performing a maximum pooling operation and achieving back propagation. By means of distinguishing an antenna pixel and a background pixel of an instance segmentation module, the antenna downtilt angle is more accurate, and a convenient, secure, effective and accurate antenna measurement method is established.