Abstract:

Aerial refueling is an important technology to improve the air endurance of Unmanned Aerial Vehicles (UAVs), and detection and identification of the refueling cone sleeve are the primary tasks to achieve aerial refueling. In order to meet the needs of fast and accurate target detection for aerial refueling, based on the keen vision mechanism of the eagle eye, a saliency detection algorithm imitating the eagle eye vision is designed. According to the characteristics of aerial refueling targets and scenes, the eagle-eye color antagonism and light-dark adaptation mechanism model is used to extract multi-channel image features from the primary image. The channel feature saliency map is subjected to spatial gating processing. On this basis, in furtherance of the salient target, the saliency map is enhanced by using the eagle-eye vision midbrain network gain model to obtain the final saliency map. According to the characteristics of aerial refueling targets and scenes, the eagle-eye color antagonism and light-dark adaptation mechanism model extract multi-channel image features from the primary image. The channel feature saliency map is subjected to spatial gating processing. To further enhance the salient target, the saliency map is enhanced by using the eagle-eye vision mid-brain network gain model to obtain the final saliency map. The algorithm is tested through the public data set and the aerial refueling simulation data set, and is then compared with other algorithms. The experimental results show that the saliency detection algorithm proposed in this paper can suppress the background influence by setting the proportion of the salient target area, highlight the salient target, and extract the aerial refueling drogue or the salient target whose size is similar.

Key words: aerial refueling, eagle-eye vision, color opponency, midbrain network, saliency detection, UAV