Abstract
Air traffic congestion-induced flight accidents pose a significant challenge in the aviation sector. Currently, aviation navigation systems primarily rely on GPS and Inertial Navigation Systems (INS) to track aircraft, yet they lack the capability to recognize and provide early warnings about the surrounding environment. To address this issue, this paper proposes a multi-aircraft parallel approach aimed at enabling coordinated flight along the same route. This method utilizes a neural network-based semantic segmentation model to monitor aircraft and perform situational awareness of the surrounding environment, thereby assisting in multi-aircraft route scheduling. When wake turbulence is generated, the model can identify the wake, further enhancing flight safety. Recently, state-space models (SSMs) based on Mamba have demonstrated outstanding performance in computational efficiency and inference speed. Based on this, we designed a U-shaped State Space Block UNet (USS-Net), which consists of StateConvBlock and ResConvBlock. The StateConvBlock integrates Mamba as a fundamental module for understanding temporal dynamics and contextual information. By constructing a symmetrical encoder-decoder structure, the model progressively extracts image features and performs multi-scale fusion to achieve high-precision pixel-level segmentation. Experimental results show that USS-Net achieved outstanding performance on the aircraft simulation dataset. On an NVIDIA A100-SXM4-40GB GPU, USS-Net attained a mean Intersection over Union (mIoU) of 95.70% and a pixel accuracy (PA) of 97.80% on the simulation training dataset. These results demonstrate that USS-Net performs effectively in assisting multi-aircraft parallel route scheduling tasks.