Detecting small infrared targets in vehicle-mounted systems remains challenging due to weak thermal radiation, cross-scale feature loss, and dynamic background interference. To address these issues, this paper proposes MSLCP-DETR, an enhanced RT-DETR-based framework that integrates multi-scale linear attention and sparse fusion mechanisms. The model introduces three novel components: a Multi-Scale Linear Attention Encoder (MSLA-AIFI), which combines multi-branch depth-wise convolution with linear attention to efficiently capture cross-scale features while reducing computational complexity; a Cross-Scale Small Object Feature Optimization module (CSOFO), which enhances the localization of small targets in dense scenes through spatial rearrangement and dynamic modeling; and a Pyramid Sparse Transformer (PST), which replaces traditional dense fusion with a dual-branch sparse attention mechanism to improve both accuracy and real-time performance. Extensive experiments on the M3FD and FLIR datasets demonstrate that MSLCP-DETR achieves an excellent balance between accuracy and efficiency, with its precision, mAP@50, and mAP@50:95 reaching 90.3%, 79.5%, and 86.0%, respectively. Ablation studies and visual analysis further validate the effectiveness of the proposed modules and the overall design strategy.

Robust visual instance segmentation for autonomous maritime surface vehicles remains challenging under adverse weather conditions such as fog and rain, which can cause contrast attenuation, boundary blurring, and visual interference. To address these challenges, we propose WADNet, a weather-adaptive instance segmentation framework for marine ship instance segmentation. At the macro level, a Weather-Adversarial Feature Decoupling (WAFD) mechanism employs gradient-reversal-based adversarial learning during training to encourage the backbone representation to become less dependent on weather-related visual cues, while introducing no additional inference-time computation. At the feature-processing level, a Weather-Adaptive Dynamic Routing Mechanism (WADRM) derives input-dependent routing weights from global feature statistics and softly combines multiple convolutional experts to provide condition-dependent feature transformation under varying weather conditions. Furthermore, Inner-CIoU is incorporated into the bounding-box regression objective to reduce the influence of uncertain peripheral regions during localization. Experiments on the evaluated mixed-weather maritime dataset demonstrate improved instance segmentation performance under the tested conditions. Specifically, under single-pass inference without test-time augmentation, WADNet achieves a Mask $AP$ of 63.40%, outperforming the YOLO11n-seg baseline by 3.18 percentage points while maintaining high-throughput inference on the evaluated GPU platform.