Multilayer Semantic Perceptual Network for Aerial Tracking

The challenges of scale variation and background interference in UAV tracking severely impact tracking performance. Currently, most existing algorithms make improvements for equiproportional scale change, but they cannot effectively deal with the challenges of aspect ratio variation. This paper proposes a multilayer semantic perceptual network consisting of a scale adaptive module (SAM) and a depth fusion module (DFM) for robust tracking. SAM captures multi-scale features at the channel level through the proposed aggregation expansion and aggregates the extracted information by a top-down path. DFM fuses the global information extracted from the search image into the multi-scale response map hierarchically, alleviating the problem of missing detail features of the target. The proposed tracker MSPTrack, which incorporates SAM and DFM, enhances the ability of target location and scale perception. Comprehensive experiments on different UAV tracking benchmarks demonstrate that our method outperforms other state-of-the-art tracking methods in terms of precision and success rate.