Abstract Aiming at the existing pavement pothole detection algorithms that are susceptible to the interference of the background environment in multiple weather scenarios, and the poor image quality leading to the low detection accuracy of small targets and the lack of real-time performance, this paper proposes a multi-dimensional structural optimization to achieve a balance between accuracy and efficiency of the YOLOv8n model. First, the SPD-Conv convolutional layer is introduced into the backbone network to enhance the feature extraction capability of the model for small- and medium-scale potholes; second, a lightweight GFPN feature fusion architecture is designed to replace part of the C2f module, and a “light backbone-heavy neck” network structure is constructed to improve the efficiency of multi-scale feature interaction; finally, a multi-scale hollow hollow detection system is integrated into the end of the backbone network, and a multi-scale hollow detection system is proposed to achieve the balance between accuracy and efficiency. Finally, a multi-scale void attention (MSDA) mechanism is integrated at the end of the backbone network to enhance spatial context modeling under multiple weather conditions. The experimental results show that the optimized models mAP50 and mAP50-95 reach 86.1% and 63.4%, respectively, which are 1.9 and 3.1 percentage points higher than the original YOLOv8n model, and the detection frame rate reaches 114.9 fps, which meets the real-time requirements of detection. This study can provide real-time and accurate pothole detection technical support for intelligent driving cars in multiple weather scenarios and light intensity.

Abstract This paper addresses the challenge of surface impurity detection in Ribbed Smoked Sheets by proposing a detection algorithm based on an improved YOLOv8 combined with SAHI slicing inference. First, a CSP_MSEIE module is designed in the backbone network to enhance the feature extraction capability for complex-shaped impurities. Second, an AFGC hybrid attention mechanism is integrated into the neck network to improve the sensitivity to impurity contours and suppress background interference. Finally, a lightweight detection head, LSCD, is designed to reduce the number of model parameters. In addition, by incorporating the SAHI slicing strategy, high-resolution input images are divided into sub-images for localized inference, and the results are subsequently fused through coordinate mapping and Non-Maximum Suppression (NMS), significantly enhancing the detection performance for small objects. Experimental results indicate that, compared with the original YOLOv8, the proposed algorithm achieves an 8.7% improvement in mAP50, while reducing the model size to only 2.34M parameters-a 22% decrease relative to the original model. This provides a reliable solution for real-time impurity removal in industrial-grade Ribbed Smoked Sheet processing equipment.

We propose UAD, an end-to-end framework with Unsupervised pretext task for vision-based Autonomous Driving, achieving the best open-loop evaluation performance in nuScenes, meanwhile showing robust closed-loop driving quality in CARLA. Our motivation stems from the observation that current end-to-end autonomous driving (E2EAD) models still mimic the modular architecture in typical driving stacks, with carefully designed supervised perception and prediction subtasks to provide environment information for oriented planning. Although achieving groundbreaking progress, such design has certain drawbacks: 1) preceding subtasks require massive high-quality 3D annotations as supervision, posing a significant impediment to scaling the training data; and 2) each submodule entails substantial computation overhead in both training and inference. To this end, we propose UAD, an E2EAD framework with an unsupervised1 proxy to address all these issues. Firstly, we design a novel Angular Perception Pretext to eliminate the annotation requirement. The pretext perceives the driving scene by predicting the angular-wise spatial objectness and temporal dynamics, without manual annotation. Secondly, a self-supervised training strategy, which learns the consistency of the predicted trajectories under different augment views, is proposed to enhance the planning robustness in steering scenarios. Our UAD achieves 38.7% relative improvements over UniAD on the average collision rate of nuScenes open-loop evaluation and obtains the route completion score of 98.5% in closed-loop evaluation of CARLA's Town05 Long benchmark, which outperforms the recent work VADv2. Moreover, the proposed method consumes only 44.3% training resources of UniAD and runs $3.4\times$3.4× faster in inference when employing the same backbone network. Our innovative design not only for the first time demonstrates unarguable performance advantages over supervised counterparts, but also enjoys unprecedented efficiency in data, training, and inference.

Mussel-inspired nanocellulose scaffold for antibacterial and anti-inflammatory coating for clear aligners.

In recent years, object detection models based on Transformers have demonstrated outstanding performance in general scenarios due to their powerful global feature modeling capabilities. However, when directly applied to aerial image detection tasks, their performance often falls short of expectations. The root cause lies in the nature of aerial imagery, which typically contains numerous small objects. These objects occupy an extremely low proportion of pixels, resulting in weak feature representation. They are also susceptible to factors such as complex background noise and mutual interference from densely distributed targets, making it difficult for Transformer models to effectively capture and distinguish small object features. To address these challenges, this paper proposes an enhanced Transformer architecture for aerial small object detection: Dynamic Interactive Fusion DETR (DIF-DETR). Its core innovations comprise two aspects: First, introducing the DIENet backbone feature extraction network embedded with DIEBlocks. These DIEBlocks serve as feature enhancement units within the backbone network, leveraging dynamic Inception multi-branch deep convolutions and adaptive weight allocation mechanisms to efficiently capture multi-scale, long-range contextual information. Second, it introduces Context-Aware Bidirectional Fusion (CABF), which enables adaptive complementary fusion of high-level semantic features and low-level detail features within the FPN-PAN architecture of the neck network, effectively mitigating the issue of small target features being obscured by background interference. Experimental results demonstrate that on the highly challenging VisDrone and HIT-UAV aerial datasets, the proposed DIF-DETR network outperforms existing mainstream models with 30.5% mAP and 82.3% mAPtest, respectively. Simultaneously, it significantly reduces computational cost to 43.6 GFLOPs with only 13.4M parameters, achieving an optimal balance between detection accuracy and computational efficiency. This demonstrates that through the synergistic effects of three core innovations, DIF-DETR significantly enhances detection accuracy and robustness for small objects in aerial images, providing an effective solution for object detection tasks in aerial scenarios.

Currently, research on cherry detection and recognition is relatively limited, and existing methods for agricultural product inspection often suffer from slow speed and low classification accuracy. To address these issues, this paper introduces an improved YOLOv11n-based model for detecting cherry ripeness, designed to enhance both the accuracy and efficiency of identifying cherries at different maturity stages. First, improvements were made to the backbone network of the YOLOv11n model by replacing the original backbone with ConvNeXtv2. This replacement achieved a broader global receptive field and enhanced multi-scale learning, which helped reduce computational costs and significantly improve efficiency while maintaining high performance. Second, a DCNv4 convolution module—an advanced convolutional layer with adaptive receptive fields—was added to the neck of the model. The neck is an intermediate stage that combines features from different layers, and the DCNv4 adapts the receptive field to help accurately locate occluded cherries of any shape and scale. This improves detection performance for small cherries without increasing computational complexity. Finally, the convolutional attention module CBAM was introduced. CBAM adaptively focuses on important image features while suppressing irrelevant background by using both channel and spatial attention mechanisms. Together, these additions significantly improve cherry detection accuracy and robustness. Our experimental results show that the improved M-YOLOv11n algorithm achieved a 4.84% increase in mAP@50 compared to the original YOLOv11n model. Precision and recall also improved by 1.25% and 0.4%, respectively. Overall, the enhanced model outperformed not only its base version but also the YOLOv5n and YOLOv8n models. Compared to multi-stage models, the proposed model demonstrates superior accuracy, speed, and reduced computational requirements. This improvement enables more efficient and precise identification of cherry ripeness, thereby enhancing the efficiency of cherry harvesting and facilitating optimal harvest timing. These advancements support the optimization of storage and transportation conditions for cherries and provide robust technical support for intelligent orchard management and the advancement of automated fruit sorting systems.

(1) Background: With the continuous development of intelligent education, classroom behavior recognition has become increasingly important in teaching evaluation and learning analytics. In response to challenges such as occlusion, scale differences, and fine-grained behavior recognition in complex classroom environments, this paper proposes an improved YOLOv11-ASV detection framework; (2) Methods: This framework introduces the Adaptive Spatial Pyramid Network (ASPN) based on YOLOv11, enhancing contextual modeling capabilities through block-level channel partitioning and multi-scale feature fusion mechanisms. Additionally, VanillaNet is adopted as the backbone network to improve the global semantic feature representation; (3) Conclusions: Experimental results show that on our self-built classroom behavior dataset (ClassroomDatasets), YOLOv11-ASV achieves 81.5% mAP50 and 62.1% mAP50–95, improving by 1.6% and 2.9%, respectively, compared to the baseline model. Notably, performance shows significant improvement in recognizing behavior classes such as “reading” and “writing” which are often confused. The experimental results validate the effectiveness of the YOLOv11-ASV model in improving behavior recognition accuracy and robustness in complex classroom scenarios, providing reliable technical support for the practical application of smart classroom systems.

Abstract We present the COordinate-attention ResiduAL Fusion and Efficient Multi-Scale Head (CORAL-EMH) Framework to overcome the problems of target missed detections, misidentification, and the difficulties associated with low visibility as well as intricate underwater settings in target detection. This framework integrates the Coordinate Attention Residual Fusion (CARF) Block in the backbone network to improve the extraction of multi-scale features. Furthermore, GCBlock has been incorporated into the last layer of the neck network to capture global context and fuse hierarchical features. The EMHead, a new detecting head, is designed to enhance the model's capability to detect minute details of underwater objects while maintaining a lightweight structure. Lastly, a new loss function known as inner-WIoU is introduced. Inner-WIoU minimizes the adverse effects of extreme samples on gradient contributions and modifies auxiliary bounding box sizes to more precisely match the actual ground-truth (GT) target boundaries, which enhances overall model performance. The experimental results demonstrate that the model is effective, attaining mean average precision scores of 86.5%, 85.1%, and 85.2% on the DUO, UTDAC2020, and RUOD datasets, respectively. Furthermore, it meets the essential criteria for real-time detection by operating at a detection rate of 85.5 FPS. Its moderate computational demand of 12.8 GFLOPs and succinct model parameter of 5.5 M render it a suitable option for incorporation into underwater detection systems.

Peanut, as an important economic crop, is widely cultivated and rich in nutrients. Classifying peanuts based on the number of seeds helps assess yield and economic value, providing a basis for selection and breeding. However, traditional peanut grading relies on manual labor, which is inefficient and time-consuming. To improve detection efficiency and accuracy, this study proposes an improved BTM-YOLOv8 model and tests it on an independently designed pod detection device. In the backbone network, the BiFormer module is introduced, employing a dual-route attention mechanism with dynamic, content-aware, and query-adaptive sparse attention to extract features from densely packed peanuts. In addition, the Triple Attention mechanism is incorporated to strengthen the model’s multidimensional interaction and feature responsiveness. Finally, the original CIoU loss function is replaced with MPDIoU loss, simplifying distance metric computation and enabling more scale-focused optimization in bounding box regression. The results show that BTM-YOLOv8 has stronger detection performance for ‘Quan Hua 557’ peanut pods, with precision, recall, mAP50, and F1 score reaching 98.40%, 96.20%, 99.00%, and 97.29%, respectively. Compared to the original YOLOv8, these values improved by 3.9%, 2.4%, 1.2%, and 3.14%, respectively. Ablation experiments further validate the effectiveness of the introduced modules, showing reduced attention to irrelevant information, enhanced target feature capture, and lower false detection rates. Through comparisons with various mainstream deep learning models, it was further demonstrated that BTM-YOLOv8 performs well in detecting ‘Quan Hua 557’ peanut pods. When comparing the device’s detection results with manual counts, the R2 value was 0.999, and the RMSE value was 12.69, indicating high accuracy. This study improves the efficiency of ‘Quan Hua 557’ peanut pod detection, reduces labor costs, and provides quantifiable data support for breeding, offering a new technical reference for the detection of other crops.

In the context of neural system structure modeling and complex visual tasks, the effective integration of multi-scale features and contextual information is critical for enhancing model performance. This paper proposes a biologically inspired hybrid neural network architecture - CompEyeNet - which combines the global modeling capacity of transformers with the efficiency of lightweight convolutional structures. The backbone network, multi-attention transformer backbone network (MATBN), integrates multiple attention mechanisms to collaboratively model local details and long-range dependencies. The neck network, compound eye neck network (CENN), introduces high-resolution feature layers and efficient attention fusion modules to significantly enhance multi-scale information representation and reconstruction capability. CompEyeNet is evaluated on three authoritative medical image segmentation datasets: MICCAI-CVC-ClinicDB, ISIC2018, and MICCAI-tooth-segmentation, demonstrating its superior performance. Experimental results show that compared to models such as Deeplab, Unet, and the YOLO series, CompEyeNet achieves better performance with fewer parameters. Specifically, compared to the baseline model YOLOv11, CompEyeNet reduces the number of parameters by an average of 38.31%. On key performance metrics, the average Dice coefficient improves by 0.87%, the Jaccard index by 1.53%, Precision by 0.58%, and Recall by 1.11%. These findings verify the advantages of the proposed architecture in terms of parameter efficiency and accuracy, highlighting the broad application potential of bio-inspired attention-fusion hybrid neural networks in neural system modeling and image analysis.

The rapid development of deep learning-based image inpainting poses serious challenges to image authenticity. As inpainting methods continue to evolve, the inpainted images exhibit extremely high visual fidelity, presenting recognition difficulties to the forgery detection model due to differences in operational mode and forgery traces among methods. In particular, the detection performance tends to drop significantly in the testing phase when the test samples differ from the training data. To address this issue, we propose a test-time adaptive detection framework for image inpainting forgeries. First, we propose an image gradient-based metric that quantifies model uncertainty and orchestrates the entire adaptation process. Integrating this metric with sample-specific batch normalization (BN) statistics enhances the ability of pretrained models in the inference stage. Second, we introduce a cross-attention module as a side-tuning module, enabling the model to adapt dynamically to reliable test samples without altering the backbone network. To validate the effectiveness of the proposed method, we construct a dataset comprising synthetic images of multiple inpainting methods and design experiments under two scenarios of distributional bias. The results demonstrate that our proposed framework outperforms the existing baseline method, enhancing the adaptability and detection performance of the forgery detection model in dynamic environments.

This article examines the growth and structural transformation of South Korea's network media economy from 2010 to 2022 as part of the Global Media and Internet Concentration Project. Using CR4 and Herfindahl–Hirschman Index measures, it analyzes revenue trends and market concentration across telecommunications, media services, and core internet sectors. Despite overall industry growth, concentration levels declined due to the rapid expansion of digital cultural industries. However, oligopolistic control persists in the telecoms sector, while global platforms and major domestic players are increasing their influence in online media services. As a result, Korea's media economy has become increasingly dualized, characterized by concentrated infrastructure sectors alongside diversified media services. The article highlights that rapid industrial expansion does not necessarily ensure the decentralization of market power, and it calls for critical perspectives on market concentration and industrial restructuring in the digital age.

Cup rupture failures in strip steel welds can lead to strip breakage, resulting in unplanned downtime of high-speed continuous rolling mills and scrap steel losses. Manual visual inspection suffers from a high false positive rate and cannot meet the production cycle time requirements. This paper proposes a lightweight online cup rupture visual inspection method based on an improved YOLOv10 algorithm. The backbone feature extraction network is replaced with ShuffleNetV2 to reduce the model’s parameter count and computational complexity. An ECA attention mechanism is incorporated into the backbone network to enhance the model’s focus on cup rupture micro-cracks. A Slim-Neck design is adopted, utilizing a dual optimization with GSConv and VoV-GSCSP, significantly improving the balance between real-time performance and accuracy. Based on the results, the optimized model achieves a precision of 98.8% and a recall of 99.2%, with a mean average precision (mAP) of 99.5%—an improvement of 0.2 percentage points over the baseline. The model has a computational load of 4.4 GFLOPs and a compact size of only 3.24 MB, approximately half that of the original model. On embedded devices, it achieves a real-time inference speed of 122 FPS, which is about 2.5, 11, and 1.8 times faster than SSD, Faster R-CNN, and YOLOv10n, respectively. Therefore, the lightweight model based on the improved YOLOv10 not only enhances detection accuracy but also significantly reduces computational cost and model size, enabling efficient real-time cup rupture detection in industrial production environments on embedded platforms.

This paper proposes a lightweight video action recognition framework that integrates 3D Convolutional Neural Networks (CNNs), the Histogram Transformer Block (HTB), and the Split-Attention Residual Block (SAB), while also introducing Spatiotemporal Tensor Factorization (ST-Factor) technology in an innovative manner. The method first incorporates the HTB module into each computational unit of the AR3D backbone network to leverage local statistical features for improve the granularity of spatiotemporal modeling. Next, the SAB module is introduced into the residual path to utilize dynamic channel re-weighting for optimizing feature selection across dimensions. Finally, the ST-Factor decouples the 4D convolution kernels into independent spatial (H W) and temporal (T C) operations, which significantly reducing computational redundancy. Experiments on the UCF101/HMDB51 datasets demonstrate that the proposed method not only maintains real-time inference speed but also outperforms existing state-of-the-art (SOTA) methods in recognition accuracy, providing a new paradigm for video understanding research.

Personnel and equipment target detection algorithms in open pit mines have significantly improved mining safety, production efficiency, and management optimization. However, achieving precise target localization in complex backgrounds, addressing mutual occlusion among multiple targets, and detecting large-scale and spatially extensive targets remain challenges for current target detection algorithms in open pit mining areas. To address these issues, this study proposes a novel target detection algorithm named RSLH-YOLO, specifically designed for personnel and equipment detection in complex open pit mining scenarios. Based on the YOLOv11 (You Only Look Once version 11) framework, the algorithm enhances the backbone network by introducing receptive field attention convolution and dilated convolution to expand the model’s receptive field and reduce information loss, thereby improving target localization capability in complex environments. Additionally, a bidirectional fusion mechanism between high-resolution and low-resolution features is adopted, along with a dedicated small-target detection layer, to strengthen multi-scale target recognition. Finally, a lightweight detection head is implemented to reduce model parameters and computational costs while improving occlusion handling, making the model more suitable for personnel and vehicle detection in mining environments. Experimental results demonstrate that RSLH-YOLO achieves a mAP (mean average precision) of 89.1%, surpassing the baseline model by 3.2 percentage points while maintaining detection efficiency. These findings indicate that the proposed model is applicable to open pit mining scenarios with limited computational resources, providing effective technical support for personnel and equipment detection in mining operations.

Galvanized steel sheets are vital anti-corrosion materials, yet their surface quality is prone to defects that impact performance. Manual inspection is inefficient, while conventional machine vision struggles with complex, small-scale defects in industrial settings. Although deep learning offers promising solutions, standard object detection models like YOLOv5 (which is short for ‘You Only Look Once’) exhibit limitations in handling the subtle textures, scale variations, and reflective surfaces characteristic of galvanized sheet defects. To address these challenges, this paper proposes S-C-B-YOLO, an enhanced detection model based on YOLOv5. First, a Squeeze-and-Excitation (SE) attention mechanism is integrated into the deep layers of the backbone network to adaptively recalibrate channel-wise features, improving focus on defect-relevant information. Second, a Transformer block is combined with a C3 module to form a C3TR module, enhancing the model’s ability to capture global contextual relationships for irregular defects. Finally, the original path aggregation network (PANet) is replaced with a bidirectional feature pyramid network (Bi-FPN) to facilitate more efficient multi-scale feature fusion, significantly boosting sensitivity to small defects. Extensive experiments on a dedicated galvanized sheet defect dataset show that S-C-B-YOLO achieves a mean average precision (mAP@0.5) of 92.6% and an inference speed of 62 FPS, outperforming several baseline models including YOLOv3, YOLOv7, and Faster R-CNN. The proposed model demonstrates a favorable balance between accuracy and speed, offering a robust and practical solution for automated, real-time defect inspection in galvanized steel production.

With the increasing frequency of traffic safety issues and the rapid development of autonomous driving technology, traffic sign detection is highly susceptible to adverse weather conditions such as changes in light intensity, fog, rain, snow, and partial occlusion, which places higher demands on the accurate recognition of traffic signs. This paper proposes an improved DAYOLO model based on YOLOv8n, aiming to balance detection accuracy and model complexity. First, the Bottleneck in the C2f module of the YOLOv8n backbone network is replaced with Bottleneck DAttention. Introducing DAttention allows for more effective feature extraction, thereby improving model performance. Second, an ultra-lightweight and efficient upsampler, Dysample, is introduced into the neck network to further improve performance and reduce computational overhead. Finally, a Task-Aligned Dynamic Detection Head (TADDH) is introduced. TADDH enhances task interaction through a dynamic mechanism and utilizes shared convolutional modules to reduce parameters and improve efficiency. Simultaneously, an additional Layer2 detection head is added to the model to strengthen the extraction and fusion of features at different scales, thereby improving the detection accuracy of small traffic signs. Furthermore, replacing SlideLoss with NWDLoss can better handle prediction results with more complex distributions and more accurately measure the distance between predicted and ground truth boxes in the feature space during object detection. Experimental results show that DAYOLO achieves 97.2% mAP on the SDCCVP dataset, which is 5.3 higher than the baseline model YOLOv8n; the frame rate reaches 120, which is 37.8% higher than YOLOv8; and the number of parameters is reduced by 6.2%, outperforming models such as YOLOv3, YOLOv5, YOLOv6, and YOLOv7. In addition, DAYOLO achieves 80.8 mAP on the TT100K dataset, which is 9.2% higher than the baseline model YOLOv8n. The proposed method achieves a balance between model size and detection accuracy, meets the needs of traffic sign detection, and provides new ideas and methods for future research in the field of traffic sign detection.

Knitted sock defect detection is a critical step in ensuring sock quality. Due to the complex texture of knitted socks and the diverse forms of defects, traditional detection methods are inefficient and lack precision. While current deep-learning-based defect detection models show great potential, they still face challenges in detecting small defects in complex backgrounds and deploying them on resource-constrained devices. To address these challenges, this study proposes a lightweight knitted sock defect detection model called Sock-YOLO. First, RepViTBlock is introduced into the backbone network to construct the reparameterization-based feature extraction module C3k2-RVB, which enhances feature extraction capabilities and eliminates the computational overhead caused by skip connections. Second, a CA-HSFPN module is designed in the Neck section, which utilizes HSFPN to dynamically filter features and suppress background noise, while integrating coordinate attention to improve the localization accuracy of small-sized defects. Next, the lightweight and efficient detection head (LEDH) replaces the original detection head, utilizing a depth-separable convolution structure to reduce computational complexity. In addition, the MPDIoU loss function is introduced to improve the regression accuracy of bounding boxes. Finally, the LAMP channel pruning strategy is adopted to alleviate the deployment pressure on edge devices. Experimental results show that compared to YOLOv11n, the pruned Sock-YOLO reduces the number of parameters and computational complexity by 72% and 47%, respectively, to 700,000 and 3.3 GFLOPS, while improving mAP 50 by 3.4% to 89.8%. The model weights were reduced by 65%, with a size of only 1.8 MB, and the inference speed reached 134.8 FPS. The research results indicate that the proposed method effectively improves the accuracy of knit sock defect detection while balancing detection accuracy and deployment costs, providing a reliable solution for knit sock defect detection tasks in industrial scenarios.

The substantial domain discrepancy inherent in multi-source and multi-view imagery presents formidable challenges to achieving precise drone-based multi-view geo-localization. Existing methodologies primarily focus on designing sophisticated backbone architectures to extract view-invariant representations within abstract feature spaces, yet they often overlook the rich and discriminative frequency-domain cues embedded in multi-view data. Inspired by the principles of π-Noise theory, this paper proposes a frequency-domain Positive-Incentive Information Screening (PIIS) mechanism that adaptively identifies and preserves task-relevant frequency components based on entropy-guided information metrics. This principled approach selectively enhances discriminative spectral signatures while suppressing redundant or noisy components, thereby improving multi-view feature alignment under substantial appearance and geometric variations. The proposed PIIS strategy demonstrates strong architectural generality, as it can be seamlessly integrated into various backbone networks including convolutional-based and Transformer-based architectures while maintaining consistent performance improvements across different models. Extensive evaluations on the University-1652 and SUES-200 datasets have validated the great potential of the proposed method. Specifically, the PIIS-N model achieves a Recall@1 of 94.56% and a mean Average Precision (mAP) of 95.44% on the University-1652 dataset, exhibiting competitive accuracy among contemporary approaches. These findings underscore the considerable promise of frequency-domain analysis in advancing multi-view geo-localization.

Purpose Sensors are vital for autonomous driving, but they are susceptible to weather noise. To address the performance degradation of point cloud object detection in foggy conditions, this paper aims to propose an efficient dynamic pillar-based point cloud detection network tailored for foggy traffic scenarios. Design/methodology/approach The point cloud data are first pillarized, and then the model adaptively adjusts the feature extraction process through DynamicPillarVFE to significantly enhance the processing capability for sparse and irregular point cloud data. Second, a bird's eye view (BEV) backbone network based on residual connections is used to effectively mitigate the gradient vanishing problem in deep networks. In addition, this paper introduces Triplet Attention (TA) in the feature enhancement part, which can focus on important features more comprehensively and suppress noise interference. Findings In this paper, we mixed a small portion of the Multifog KITTI dataset on top of the KITTI dataset. Experimental results demonstrate that under moderate difficulty 3D evaluation metrics, our method achieves accuracy improvements of 5.03%, 6.66% and 7.78% for cars, pedestrians and cyclists, respectively, compared to the PointPillars baseline, significantly enhancing point cloud object detection performance in foggy conditions. Furthermore, the DPSF-Net architecture achieves an inference speed of 32.36 ms per frame, fully meeting the real-time processing requirements for autonomous driving applications. Originality/value This method is more economical to simulate the influence of sensors under foggy conditions, and through the training of mixed data sets, the network model can better cope with foggy interference.