Abstract：In response to the problems existing in the detection of maritime crashed aircraft targets, such as insufficient background noise suppression, incomplete target contours, weakened target features, and difficulty in identifying small target pixels, a dual-stream fusion gray value discrimination algorithm based on time difference accumulation optimization is proposed. By constructing a three-dimensional optimization of "dual-target feature enhancement-noise suppression-spatiotemporal registration", a five-level progressive image processing is realized. An airborne platform image acquisition and evaluation system under typical scenarios is built; "day-night" stepwise experiments are conducted, and it is found that the maximum dual-target signal ratio in daytime experiments can reach 1.739, and the signal-to-background ratio in nighttime environments can reach 25.

Abstract：Based on the actual needs of landing support in UAV after-sales service, this paper carries out research on technologies that use solar-blind ultraviolet detectors to support UAV after-sales maintenance and support. Through the analysis of atmospheric transmission characteristics in the ultraviolet band, the design of ultraviolet detection system, the design of ultraviolet cooperative beacon points, and the experimental verification of P4P photogrammetry algorithm, it provides technical support for landing performance evaluation, fault diagnosis, and maintenance effect verification in UAV after-sales support, and verifies the feasibility and rationality of this technology in ensuring the reliability of UAV landing in after-sales service.

Abstract：In response to the complex and changeable marine environment, such as fog, strong light reflection, low illumination at night, and the limited characteristic information of small targets at sea, a small target detection method based on YOLOv11 for marine multispectral features is proposed. By designing a dual-branch YOLOv11 model to handle the fusion of cross-modal data features and introducing a global attention mechanism module for training, the improved multispectral image small target detection model can fully utilize the multispectral features, achieving accurate detection and positioning of small target objects in multispectral images, especially performing well from the perspective of unmanned aerial vehicle (UAV) aerial photography. This cross-modal fusion method can significantly enhance the robustness and accuracy of small target detection at sea. Experiments show that the improved model can achieve an mAP@50 of 96.5% on the VTSaR dataset, an increase of 0.4% compared to YOLOv11n, providing a new solution for the detection of small targets in marine aerial unmanned search and rescue.