Ping Lin

Railway track surface defect detection is essential for ensuring operational safety and maintaining the structural integrity of rail systems. However, existing detection frameworks often struggle with complex background interference, irregular defect shapes, and the difficulty of identifying small and subtle defects under real-world railway conditions. To overcome these challenges, this study presents an enhanced real-time detection framework built upon RT-DETR and tailored for railway track inspection. A frequency–state space modeling block (FSSM) is proposed to jointly capture global contextual information and fine-grained texture cues, while a spatially-enhanced feedforward network is introduced to strengthen spatial consistency and improve the localization of small defects. In addition, a multi-scale reverse fusion module is designed to reconstruct high-frequency structural details through reverse-convolution upsampling and enrich multi-scale semantic features. Experiments on the TrackDefect dataset show that the proposed method achieves an AP of 60.7%, markedly surpassing the baseline RT-DETR. Further evaluation on the GC10-DET dataset confirms strong generalization capability, with the model maintaining robust detection performance across diverse defect types. These results demonstrate the effectiveness of the proposed framework in improving both accuracy and robustness for practical railway track surface defect detection. The RailDefect dataset is publicly available at https://github.com/ganyang007/TrackDefect .

Detecting aircraft targets in optical remote sensing(RS) imagery is essential for airspace management, airport operations, and defense applications. However, RS aircraft detection remains challenging due to the prevalence of small targets, dense spatial arrangements, multi-scale target variations, and complex backgrounds. To address these issues, a lightweight detection model, AR-DETR, is developed for aircraft detection in RS imagery, building upon the Real-Time Detection Transformer (RT-DETR) framework. Firstly, a cross-aware feature enhancement module is proposed to strengthen both spatial and channel feature interactions, effectively enhancing the representation of small and weakly textured aircraft under cluttered backgrounds. Next, a dual-scale frequency fusion module is designed, which combines multi-scale depth-wise convolutions in the spatial domain with discriminative frequency selection in the frequency domain, thereby improving structural pattern extraction across different scales while suppressing background noise. Finally, an edge-Gaussian contextual enhancement module is developed to replace the conventional neck block, enabling fine-grained detail preservation and contextual enhancement through edge-guided Gaussian modulation with reduced computational cost. Results on the DOTA1.0 dataset demonstrate that AR-DETR achieves superior accuracy and efficiency compared to state-of-the-art detectors. Further validation on the SODA-A dataset, which emphasizes small-and medium-sized aircraft, confirms the proposed model's robustness and generalization under complex RS conditions.

Ship remote sensing detection technology plays a crucial role in modern maritime navigation and ocean management. However, ship targets exhibit multi-scale characteristics, diverse categories, and dense distributions, while image quality is often affected by weather variations and complex sea surface backgrounds, making detection highly challenging. In this paper, we propose RS-YOLO, an accurate ship detection algorithm based on YOLO11. Firstly, an Adaptive Convolutional Fourier Frequency Fusion Module is designed with a spatial-frequency cascading structure to effectively integrate local and global information, thereby enhancing feature extraction under low-resolution and complex background conditions. In addition, a Synergistic Convolution-Attention Mechanism is developed to strengthen the interaction between local feature extraction and global relationship modeling, enabling the model to accurately recognize ships in multi-resolution and densely distributed scenarios. Finally, a Shape IoU loss function is incorporated to better adapt to the complex shapes and scale variations of ships, improving the robustness and accuracy of bounding box regression. Experimental results demonstrate that the proposed method consistently outperforms the YOLO11n baseline across four public datasets, achieving mAP0.5 improvements of 2.1%, 2.2%, 1.6%, and 1.8% on HRSC2016, NWPU VHR-10, LEVIR-Ship, and DIOR, respectively. Moreover, comparative evaluations with several state-of-the-art detection algorithms further confirm the superior performance and robustness of RS-YOLO in complex maritime environments.

Detecting ship targets in optical remote sensing images is crucial for ensuring the safety of maritime traffic. However, remote sensing ship detection faces challenges such as low resolution, dense distribution, and multi-scale variations. Additionally, complex backgrounds and changing weather conditions further complicate the detection process. To address these challenges, a lightweight remote sensing ship detection architecture derived from YOLO11 is proposed. Firstly, a Fourier convolution dual-domain fusion module is designed to efficiently integrate multi-scale convolution operations in both the spatial and frequency domains, thereby enhancing the model’s capacity to recognize ship targets under complex geographical and weather conditions. Furthermore, a partially parallel synergistic attention mechanism is developed to comprehensively model both global and local features, effectively mitigating the difficulty of extracting ship features caused by complex background interference. Finally, a Ship intersection over union loss function is proposed, which dynamically calculates the loss by focusing on the shape and scale of the ship targets, ensuring that the geometric properties of the targets are better captured by the model. The improved model demonstrates superior detection performance and lower parameter parameters on the HRSC2016 and DOTA1.0 public remote sensing datasets compared to other advanced algorithms.

Aircraft target detection in remote sensing images faces numerous challenges, including target size variations, low resolution, and complex backgrounds. To address these challenges, an enhanced end-to-end aircraft detection framework (DIMD-DETR) is developed based on an improved metric space. Initially, a bilayer targeted prediction method is proposed to strengthen gradient interaction across decoder layers, thereby enhancing detection accuracy and sensitivity in complex scenarios. The pyramid structure and self-attention mechanism from pyramid vision transformer V2 are incorporated to enable effective joint learning of both global and local features, which significantly boosts performance for low-resolution targets. To further enhance the model's generalization capabilities, an aircraft-specific data augmentation strategy is meticulously devised, thereby improving the model's adaptability to variations in scale and appearance. Additionally, a metric-space-based loss function is developed to optimize the collaborative effects of the modular architecture, enhancing detection performance in complex backgrounds and under varying target conditions. Finally, a dynamic learning rate scheduling strategy is proposed to balance rapid convergence with global exploration, thereby elevating the model's robustness in challenging environments. Compared to current popular networks, our model demonstrated superior detection performance with fewer parameters.

The existing strawberry ripeness detection algorithm has the problems of a low precision and a high missing rate in real complex scenes. Therefore, we propose a novel model based on a hybrid attention mechanism. Firstly, a partial convolution-based compact inverted block is developed, which significantly enhances the feature extraction capability of the model. Secondly, an efficient partial hybrid attention mechanism is established, which realizes the remote dependence and accurate localization of strawberry fruit. Meanwhile, a multi-scale progressive feature pyramid network is constructed, and the fine-grained features of strawberry targets of different sizes are accurately extracted. Finally, a Focaler-shape-IoU loss function is proposed to effectively solve the problem of the difficulty imbalance between strawberry samples and the influence of the shape and size of the bounding box on the regression. The experimental results show that the model’s precision and mAP0.5 reach 92.1% and 92.7%, respectively, which are 2.0% and 1.7% higher than the baseline model. Additionally, our model is better in detection performance than most models with fewer parameters and lower FLOPs. In summary, the model can accurately identify the maturity of strawberry fruit under complex farmland environments and provide certain technical guidance for automated strawberry-picking robots.