
Object detection in foggy and dynamic environments remains challenging, especially for small objects whose features are easily degraded by low visibility. Existing methods often suffer from insufficient robustness and limited real-time performance. This study aims to develop a high-efficiency and highly robust detection framework tailored for foggy small-object detection. We propose HS-MambaDet, a hybrid state-space detection network built upon the Mamba-Transformer architecture. A hidden-state compression-enhanced backbone(HSM-SSD) is designed to reduce computational overhead while preserving global dependency modeling.Additionally, a Multi-scale Frequency-Spatial Fusion (MFSF) module-consisting of a high-frequency perception branch and a spatial dependency modeling branch-is introduced to enhance fine-grained details and long-range spatial relations for small objects. The overall framework integrates CNN, Transformer, and Mamba components to jointly capture local textures, global context, and efficient sequential dependencies. Extensive experiments on the RTTS and CityScapes datasets demonstrate that HS-MambaDet outperforms mainstream SOTA models in both accuracy and efficiency. On RTTS, the complete model achieves 87.3 % precision, 73.1 % recall, 81.2 % mAP@0.5, and 51.0% mAP@(0.5 similar to 0.95), exceeding the best comparison baselines by up to 3.8 %, 3.9 %, 3.6 %, and 3.3 %, respectively. The inference time remains low at 0.26 s, ensuring real-time capability. In small-object scenarios, HS-MambaDet improves mAP@0.5 by up to 4.4 %, and cross-domain tests under varying fog intensities further verify its superior generalization and robustness. By combining hidden state compression with multi-scale frequency-spatial fusion, HS-MambaDet effectively enhances fine-detail perception and spatial dependency modeling under foggy conditions. The proposed framework achieves strong performance in accuracy, robustness, and inference speed, offering a practical and efficient solution for real-time object detection in adverse weather and dynamic environments.
Existing image super-resolution algorithms face issues such as high computational cost, limited perception of local high-frequency textures, and insufficient extraction of key information. To address these challenges, this paper proposes a lightweight image super-resolution reconstruction network based on high-frequency fusion attention. First, a Dual-Path Enhancement module is designed within the network architecture to explicitly decouple features into parallel high-frequency and low-frequency branches. The low-frequency branch captures smooth structures, while the high-frequency branch employs learnable high-pass filters to accurately extract edges and textures. Second, a spatial refinement module is introduced, utilizing directionally separated convolutions to enhance local features and generate queries rich in spatial information. Finally, an efficient asymmetric frequency-aware attention mechanism is designed to achieve effective interaction between local and global information. This mechanism uses the local information output from the spatial refinement module as a query to dynamically retrieve complementary information across the entire image, effectively overcoming the limitation of local windows on long-range dependencies. Experiments on the Urban100 dataset demonstrate that at & times;4 magnification scales, the proposed algorithm achieves an average PSNR improvement of 0. 53 dB and an average SSIM improvement of 0.014 7 compared to other models, while running at only 14% of SwinIR's computational time. Regarding visual quality in image reconstruction, the model generates sharper edges and more realistic textures, fully validating its effectiveness and advanced capabilities in image reconstruction while maintaining lightweight efficiency.
Spectral Domain Optical Coherence Tomography(SD-OCT) systems have been extensively studied for biological tissue measurement, featuring non-invasiveness, micron-level resolution, and real-time detection capabilities. To expand their application scope to the measurement of industrial products with multilayer structures, it is necessary to optimize the performance of SD-OCT systems. Firstly, this study optimizes the parameters of the sample arm field lens using ZEMAX. Aberrations are reduced by adjusting the lens curvature, focal length, and optical path layout, and a matched scanning lens is designed to expand the detection range. Secondly, the spectrometer parameters are optimized: appropriate lenses and gratings are selected, and the optical path layout and device tilt angle are adjusted to improve spectral resolution and imaging depth. Finally, MATLAB is used to establish a simulation model for the system's axial resolution, which supports tomographic imaging simulation experiments of multilayer sample structures. The results show that the comprehensive performance of the optimized SD-OCT system is significantly improved. The lateral resolution is enhanced to 20 & micro;m, while a large scanning range of 34 mm & times; 34 mm is achieved. The imaging depth is increased to 2. 44 mm, enabling effective penetration of multilayer industrial structures. The axial resolution is verified by simulation to be 6. 40 & micro;m, with the simulation results highly consistent with the theoretical calculation values. Additionally, the tomographic imaging simulation of multilayer sample structures is successfully realized. The proposed optimization design method for the SD-OCT system provides valuable reference and promotion for its applications in other fields.
Surface defects on liquid crystal displays(LCDs) impair appearance and reliability, presenting challenges such as wide-scale variations, complex backgrounds, and difficulty in detecting small targets. This paper proposes FFD-YOLO, an LCD defect detection algorithm based on the lightweight YOLOv8n framework. The algorithm uses the FasterNet backbone to enhance feature extraction. It designs the Feature Pyramid Shared Convolution(FPSC) module, which uses multi-expansion-rate convolutions and shared convolutional mechanisms to enhance multi-scale feature modeling. Additionally, it proposes the Multi-Scale Adaptive Convolution Module(MACM), which employs dynamic convolution weights and multi-scale convolution kernels to enhance the representation of small objects and stability in complex backgrounds. Experimental results demonstrate that on the self-built industrial-grade LCD-NET dataset, FFD-YOLO achieves 5. 0%, 4. 7%, and 3. 2% improvements in Precision, Recall, and mAP50, respectively, compared to baseline models, with a 6. 4% boost in accuracy for detecting small stain-type objects. These results demonstrate that FFD-YOLO significantly enhances LCD defect detection performance while maintaining lightweight efficiency, offering an effective and reliable solution for industrial vision inspection systems.
To achieve efficient and accurate defect detection for wind turbine blades,this study develops a UAV-based intelligent inspection system and investigates its core algorithms,including multi-stage defect screening,representative frame selection,and deep learning-based defect recognition.First,based on image features such as structural integrity and texture patterns,an improved EfficientNetV2-S network is proposed for preliminary damage screening(accuracy 87.5%).Next,Gaussian Mixture Clustering is employed to select representative damaged frames from video streams,effectively reducing data redundancy.Then,a YOLO-v11 network integrated with Transformer modules and dynamic anchor strategies is developed for precise defect localization and classification.Additionally,a comprehensive dataset containing 15 000 annotated UAV images of various defects(e.g.,cracks,delamination)is established for algorithm validation.Experimental results demonstrate that the system achieves 88.4%recall and 71.5%precision in single-frame detection,while video-level analysis improves to 95.0%recall and 86.4%precision.The method outperforms existing approaches in detecting multiple defect types under complex environmental conditions.The proposed system meets the requirements for non-contact inspection,real-time processing,high accuracy,and strong anti-interference capability,providing an effective solution for intelligent wind turbine maintenance.