The internal spatial configuration of push-broom multispectral imagers causes different observation directions for various detectors, leading to parallax for the same object across images in different spectral bands. This parallax can be utilized to infer detection information and estimate parameters for moving targets. However, most existing methods rely on traditional paradigms, which suffer from low detection performance and robustness. This paper introduces the first multispectral dataset specifically designed for detecting airborne moving aircraft targets. To address the challenge of detecting small airborne moving targets in multispectral images, we propose modifications to the YOLOv8 network. By introducing skip connections in the neck section, we facilitate information transfer between multi-scale feature layers, thereby enhancing the network’s ability to detect small targets. Experimental results demonstrate that the improved model significantly enhances detection accuracy and robustness in the task of target detection within multispectral images. Finally, the rainbow effect in multispectral images is utilized to estimate the parameters of the detected aircraft targets, and the results show that the proposed parameter estimation method offers high reliability.