With advancements in remote sensing technologies, high-resolution imagery has become increasingly accessible, supporting applications in urban planning, environmental monitoring, and precision agriculture. However, semantic segmentation of such imagery remains challenging due to complex spatial structures, fine-grained details, and land cover variations. Existing methods often struggle with ineffective feature representation, suboptimal fusion of global and local information, and high computational costs, limiting segmentation accuracy and efficiency. To address these challenges, we propose the dual-level network (DLNet), an enhanced framework incorporating self-attention and cross-attention mechanisms for improved multi-scale feature extraction and fusion. The self-attention module captures long-range dependencies to enhance contextual understanding, while the cross-attention module facilitates bidirectional interaction between global and local features, improving spatial coherence and segmentation quality. Additionally, DLNet optimizes computational efficiency by balancing feature refinement and memory consumption, making it suitable for large-scale remote sensing applications. Extensive experiments on benchmark datasets, including DeepGlobe and Inria Aerial, demonstrate that DLNet achieves state-of-the-art segmentation accuracy while maintaining computational efficiency. On the DeepGlobe dataset, DLNet achieves a $76.9\%$ mean intersection over union (mIoU), outperforming existing models such as GLNet ($71.6\%$) and EHSNet ($76.3\%$), while requiring lower memory (1443 MB) and maintaining a competitive inference speed of 518.3 ms per image. On the Inria Aerial dataset, DLNet attains an mIoU of $73.6\%$, surpassing GLNet ($71.2\%$) while reducing computational cost and achieving an inference speed of 119.4 ms per image. These results highlight DLNet’s effectiveness in achieving precise and efficient segmentation in high-resolution remote sensing imagery.

Multiple Object Tracking (MOT) aims to detect and track multiple targets across consecutive video frames while preserving consistent object identities. While appearance-based approaches have achieved notable success, they often struggle in challenging conditions such as occlusions, motion blur, and the presence of visually similar objects, resulting in identity switches and fragmented trajectories. To address these limitations, we propose Motion-Perception Multi-Object Tracking (MPMOT), a motion-aware tracking framework that emphasizes robust motion modeling and adaptive association. MPMOT incorporates three core components: (1) a Gain Kalman Filter (GKF) that adaptively adjusts detection noise based on confidence scores, stabilizing motion prediction during uncertain observations; (2) an Adaptive Cost Matrix (ACM) that dynamically fuses motion and appearance cues during track–detection association, improving robustness under ambiguity; and (3) a Global Connection Model (GCM) that reconnects fragmented tracklets by modeling spatio-temporal consistency. Extensive experiments on the MOT16, MOT17, and MOT20 benchmarks demonstrate that MPMOT consistently outperforms state-of-the-art trackers, achieving IDF1 scores of 72.8% and 72.6% on MOT16 and MOT17, respectively, surpassing the widely used FairMOT baseline by 1.1% and 1.3%. Additionally, rigorous statistical validation through post hoc analysis confirms that MPMOT’s improvements in tracking accuracy and identity preservation are statistically significant across all datasets. MPMOT delivers these gains while maintaining real-time performance, making it a scalable and reliable solution for multi-object tracking in dynamic and crowded environments.

In recent years, many approved drugs have been discovered using phenotypic screening, which elaborates the exact mechanisms of action or molecular targets of drugs. Drug susceptibility prediction is an important type of phenotypic screening. Large-scale pharmacogenomics studies have provided us with large amounts of drug sensitivity data. By analyzing these data using computational methods, we can effectively build models to predict drug susceptibility. However, due to the differences in data distribution among databases, researchers cannot directly utilize data from multiple sources. In this study, we propose a deep transfer learning model. We integrate the genomic characterization of cancer cell lines with chemical information on compounds, combined with the Encyclopedia of Cancer Cell Lines (CCLE) and the Genomics of Cancer Drug Sensitivity (GDSC) datasets, through a domain-adapted approach and predict the half-maximal inhibitory concentrations (IC50 values). Afterward, the validity of the prediction results of our model is verified. This study effectively addresses the challenge of cross-database distribution discrepancies in drug sensitivity prediction by integrating multi-source heterogeneous data and constructing a deep transfer learning model. This model serves as a reliable computational tool for precision drug development. Its widespread application can facilitate the optimization of therapeutic strategies in personalized medicine while also providing technical support for high-throughput drug screening and the discovery of new drug targets.