Remote sensing object detection based on the combination of infrared and visible images can effectively adapt to the around-the-clock and changeable illumination conditions. However, most of the existing infrared and visible object detection networks need two backbone networks to extract the features of two modalities, respectively. Compared with the single modality detection network, this greatly increases the amount of calculation, which limits its real-time processing on the vehicle and unmanned aerial vehicle (UAV) platforms. Therefore, this paper proposes a local adaptive illumination-driven input-level fusion module (LAIIFusion). The previous methods for illumination perception only focus on the global illumination, ignoring the local differences. In this regard, we design a new illumination perception submodule, and newly define the value of illumination. With more accurate area selection and label design, the module can more effectively perceive the scene illumination condition. In addition, aiming at the problem of incomplete alignment between infrared and visible images, a submodule is designed for the rapid estimation of slight shifts. The experimental results show that the single modality detection algorithm based on LAIIFusion can ensure a large improvement in accuracy with a small loss of speed. On the DroneVehicle dataset, our module combined with YOLOv5L could achieve the best performance.

Multi-Object Tracking (MOT) is an important topic in computer vision. Recent MOT methods based on the anchor-free paradigm trade complicated hierarchical structures for tracking performance. However, existing anchor-free MOT methods ignore the noise in detection, data association, and trajectory reconnection stages, which results in serious problems, such as missing detection of small objects, insufficient motion information, and trajectory drifting. To solve these problems, this paper proposes Noise-Control Tracker (NCT), which focuses on the noise-control design of detection, association, and reconnection. First, a prior depth denoise method is introduced to suppress the fusion feature redundant noise, which can recover the gradient information of the heatmap fusion features. Then, the Smoothing Gain Kalman filter is designed, which combines the Gaussian function with the adaptive observation coefficient matrix to stabilize the mutation noise of Kalman gain. Finally, to address the drift noise issue, the gradient boosting reconnection context mechanism is designed, which realizes adaptive trajectory reconnection to effectively fill the gaps in trajectories. With the assistance of the plug-and-play noise-control method, the experimental results on MOTChallenge 16 &17 datasets indicate that the NCT can achieve better performance than other state-of-the-art trackers.

Model partitioning techniques, which decompose and collaboratively execute subtasks of deep neural networks (DNNs), have emerged as a critical strategy for enhancing distributed inference efficiency. However, in mobile edge computing (MEC), dynamic load fluctuations at edge nodes and the complexity of cross-node task dependencies make the delay minimization problem extremely challenging. Existing studies predominantly adopt a decoupled optimization framework that separately addresses partitioning-offloading and pipeline scheduling, neglecting their inherent cyclic state-dependent coupling. This oversight leads to suboptimal solutions, such as pipeline stagnation caused by mismatched computation and communication timestamps. To address these challenges, we propose a multi-endpoint directed acyclic graph (DAG)-driven cooperative optimization approach, enabling partitioning-offloading and pipeline scheduling in MEC. Specifically, the approach involves two core steps: 1)Dynamic pre-scheduling: We propose an improved DNN scheduling algorithm for constrained subtasks, which simulates node-level queuing delays and pipeline stalls under real-world constraints, translating runtime states into latency objectives. 2)Partitioning and offloading solution retrieval: Based on latency objectives, we introduce a novel multi-endpoint DAG structure and design a multi-node collaborative optimization retrieval algorithm, enabling adaptive partitioning-offloading remapping of subtasks. Experiments demonstrate the superiority of the proposed method over other advanced methods, reducing the time overhead by an average of 24% and 75% in two different scenarios, respectively. The resource code can be found at: https://github.com/aiheiheiheii/Partition_Scheduling.git.

Domain generalization (DG) aims to learn knowledge from multiple related domains to achieve a robust generalization performance in unseen target domains, which is an effective approach to mitigate domain shift in remote sensing image classification. Although the sharpness-aware minimization (SAM) method enhances DG capability and improves remote sensing image classification performance by promoting the convergence of the loss minimum to a flatter loss surface, the perturbation loss (maximum loss within the neighborhood of a local minimum) of SAM fails to accurately measure the true sharpness of the loss landscape. Furthermore, its variants often overlook gradient conflicts, thereby limiting further improvement in DG performance. In this paper, we introduce implicit sharpness-aware minimization (ISAM), a novel method that addresses the deficiencies of SAM and mitigates gradient conflicts. Specifically, we demonstrate that the discrepancy in training loss during gradient ascent or descent serves as an equivalent measure of the dominant eigenvalue of the Hessian matrix. This discrepancy provides a reliable measure for sharpness. ISAM effectively reduces sharpness and mitigates potential conflicts between gradients by implicitly minimizing the discrepancy between training losses while ensuring a sufficiently low minimum through minimizing perturbation loss. Extensive experiments and analyses demonstrate that ISAM significantly enhances the model’s generalization ability on remote sensing and DG datasets, outperforming existing state-of-the-art methods.

As an important part of autonomous driving intelligence perception, pedestrian detection has high requirements for parameter size, real-time, and model performance. Firstly, a novel multiplexed connection residual block is proposed to construct the lightweight network for improving the ability to extract pedestrian features. Secondly, the lightweight scalable attention module is investigated to expand the local perceptual field of the model based on dilated convolution that can maintain the most important feature channels. Finally, we verify the proposed model on the Caltech pedestrian dataset and BDD 100 K datasets. The results show that the proposed method is superior to existing lightweight pedestrian detection methods in terms of model size and detection performance.