The Target Recognition Algorithm based on the Coordinate Attention Mechanism and the Mutual-channel Loss

Abstract

To order to meet the target detection and recognition needs of the helicopter / UA V cooperative situation awareness system, the target recognition algorithm based on the coordinate attention mechanism and mutual-channel Loss of UA V is proposed. Our method uses deep separable convolution and a double threshold attention mechanism to build a lightweight deep convolution neural network with strong feature extraction ability and uses a loss function based on feature channel information to drive the network model to obtain convolution features with strong discrimination. The experimental results show that the proposed algorithm can effectively obtain higher UAV target recognition accuracy, and reduce neural network parameters and computing costs. It can provide new ideas and solutions for the realization of element awareness tasks in a helicopter/UAV cooperative situation awareness system.