This study introduces Ship-Yolo, a novel algorithm designed to tackle the challenges of detecting small targets against complex backgrounds in remote sensing imagery. Firstly, the proposed method integrates an efficient local attention mechanism into the C3 module of the neck network, forming the EDC module. This enhancement significantly improves the model’s capability to capture critical features, enabling robust performance in scenarios involving intricate backgrounds and multi-scale targets. Secondly, a Lightweight Asymmetric Decoupled Head (LADH-Head) is proposed to separate classification and regression tasks, reducing task conflicts, improving detection performance, and maintaining the model’s lightweight characteristics. Additionally, the LiteConv module is designed to replace the C3 module in the backbone network, leveraging partial convolution to ignore invalid information in occluded regions and avoid misjudgments. Finally, the Content-Aware Reassembly Upsampling Module (CARAFE) is employed to replace the original upsampling module, expanding the receptive field to better capture global information while preserving the lightweight nature of the model. Experiments on the ShipRSImageNet and DOTA datasets demonstrate that Ship-Yolo outperforms other YOLO variants and existing methods in terms of precision, recall, and average precision, exhibiting strong generalization capabilities. Ablation studies further validate the stable performance improvements contributed by the EDC, LADH-Head, LiteConv, and CARAFE modules.

Maritime object detection is essential for resource monitoring, maritime defense, and public safety, yet detecting diverse targets beyond ships remains challenging. This paper presents YOLO-SEA, an efficient detection framework based on the enhanced YOLOv8 architecture. The model incorporates the SESA (SimAM-Enhanced SENetV2 Attention) module, which integrates the channel-adaptive weight adjustment of SENetV2 with the parameter-free spatial-channel modeling of SimAM to enhance feature representation. An improved BiFPN (Bidirectional Feature Pyramid Network) structure enhances multi-scale fusion, particularly for small object detection. In the post-processing stage, Soft-NMS (Soft Non-Maximum Suppression) replaces traditional NMS to reduce false suppression in dense scenes. YOLO-SEA detects eight maritime object types. Experiments show it achieves a 5.8% improvement in mAP@0.5 and 7.2% improvement in mAP@0.5:0.95 over the baseline, demonstrating enhanced accuracy and robustness in complex marine environments.

Non-cooperative spatial target detection plays a vital role in enabling autonomous on-orbit servicing and maintaining space situational awareness (SSA). However, due to the limited computational resources of onboard embedded systems and the complexity of spaceborne imaging environments, where spacecraft images often contain small targets that are easily obscured by background noise and characterized by low local information entropy, many existing object detection frameworks struggle to achieve high accuracy with low computational cost. To address this challenge, we propose YOLO-GRBI, an enhanced detection network designed to balance accuracy and efficiency. A reparameterized ELAN backbone is adopted to improve feature reuse and facilitate gradient propagation. The BiFormer and C2f-iAFF modules are introduced to enhance attention to salient targets, reducing false positives and false negatives. GSConv and VoV-GSCSP modules are integrated into the neck to reduce convolution operations and computational redundancy while preserving information entropy. YOLO-GRBI employs the focal loss for classification and confidence prediction to address class imbalance. Experiments on a self-constructed spacecraft dataset show that YOLO-GRBI outperforms the baseline YOLOv8n, achieving a 4.9% increase in mAP@0.5 and a 6.0% boost in mAP@0.5:0.95, while further reducing model complexity and inference latency.

Determining whether stress overshoot can be predicted from pre-yield structural information remains a central challenge in understanding amorphous yielding. However, conventional analyses usually require the complete stress-strain response to determine whether stress overshoot occurs. To overcome this limitation, a precise predictive model with hybrid convolutional-temporal encoding is developed, and the temporal-spatial class activation maps (TS-CAM) method is introduced to interpret stress-overshoot behavior from pre-yield amorphous configuration sequences. Combined with the TS-CAM analysis, the spatiotemporal evolution of model-salient regions associated with stress-overshoot classification is quantitatively characterized. Furthermore, a model-derived scalar descriptor is constructed to characterize predictive signatures correlated with stress-overshoot yielding within the present dataset. This work demonstrates the capability of interpretable deep learning in exploring the plastic deformation mechanism of amorphous solids, and provides a data-driven insight for the yielding regulation and structural optimization of amorphous materials.