Search Results (22)

Real-time perception of coal–biomass blending during conveyor-belt transport remains challenging because of local aggregation, particle overlap, and illumination variation. In this study, a laboratory-scale conveyor-belt image dataset covering different coal mass fractions, illumination conditions, and particle sizes was constructed. Whole-image classification, cropped-ROI classification, direct regression, horizontal bounding box (HBB)-based detection, oriented bounding box (OBB)-based detection, and RT-DETR-L detection baselines were compared using YOLO-series and auxiliary models. Coal mass fraction was estimated using a frequency-weighted statistical strategy that converts frame-level predictions into continuous estimates. YOLOv8-cls achieved an average RMSE of 13.98 percentage points (pp), indicating the influence of background interference in whole-image classification. Among HBB models, YOLOv8m achieved the lowest mean RMSE of 6.10 pp but required higher computational cost. Compared with YOLOv8n, YOLOv8n-OBB reduced the average RMSE from 9.02 to 6.90 pp by providing a more compact material-region representation and reducing background redundancy. These results show that OBB representation improves the stability of lightweight models. The proposed method provides a feasible vision-based soft-sensing approach for online trend monitoring of coal–biomass blending under lightweight deployment. Full article

Defect detection in manufacturing has evolved from manual inspection to deep learning-based Automated Visual Inspection (AVI) systems; however, acquiring sufficient defect samples in real industrial environments remains challenging, causing severe data sparsity and class imbalance. We propose CASDA (Context-Aware Steel Defect Augmentation), a five-stage framework that classifies defect morphology and background surface properties, constructs a compatibility matrix encoding their contextual relationship, and synthesizes defect images via a ControlNet pipeline conditioned on a three-channel hint image. Experiments on the Severstal steel dataset demonstrate that CASDA achieves an 83.0% quality validation pass rate. Under multi-seed evaluation (seeds 42 and 456), CASDA improved EB-YOLOv8’s overall mAP@0.5 by 2.60 pp over the raw baseline and achieved a Class 2 AP gain of 22.09 pp over Copy-Paste, suggesting that context-aware synthesis produces more discriminative minority-class training samples than simple patch reuse under the tested settings. Performance gains are architecture-dependent; YOLO-MFD did not show overall improvement, indicating that augmentation sensitivity varies with backbone feature representation. Full article

To address prevalent challenges in stone inscription character detection—including glyph blurring, incompleteness, densely arranged characters, and substantial inter-object scale variation—this paper proposes PMN-YOLO, a lightweight, enhanced detection model built upon YOLOv11n. To enhance the detection performance of stele characters in complex scenarios, PP-LCNet is introduced to replace the original feature extraction structure. This not only reduces the model complexity but also enhances the feature expression ability for complex textures and blurred characters, providing effective support for subsequent recognition tasks. Consequently, the proposed model achieves a better balance between detection precision and computational efficiency. Second, we designed the MSAM-smallTarget module specifically for detecting small targets. By integrating multi-scale convolutional operations with spatial attention mechanisms, this module significantly enhances the model’s ability to perceive fine-grained features as well as blurred or fragmented characters commonly found in inscriptions. Furthermore, leveraging small convolutional kernels, dilated convolutions, and lightweight convolutional designs enables adaptive expansion of the receptive field while effectively constraining parameter count. Third, the NWD loss function based on the Wasserstein distance is introduced to replace the traditional IoU metric with a distribution-based similarity measure, thereby significantly improving the model’s localization robustness in scenarios involving densely distributed targets and ambiguous boundaries. The experimental results show that on the self-built stone inscription dataset, the precision of PMN-YOLO reaches 98.3%, the recall rate is 75.5%, and the mAP@50 and mAP@50-95 are 82.7% and 65.0% respectively. The model has 8.7% fewer parameters than the baseline. This method achieves lightweight performance and high detection accuracy, delivering a practical approach to automatically detect and digitally safeguard stone inscriptions. Full article

Target detection in synthetic aperture radar (SAR) images is constrained by three primary challenges. First, speckle noise overlaps heavily with the high-frequency features of target edges in the frequency domain, so standard convolutions cannot suppress noise without sacrificing edge texture. Second, the scattering signature of a SAR target varies markedly with viewing angle, and a fixed-parameter convolution kernel cannot accommodate this spatial non-stationarity. Third, deep and shallow levels of the feature pyramid differ in semantics and resolution, and a naive element-wise sum either introduces noise interference or loses small-target signals. We propose the Frequency–Spatial Detection Network (FSDNet), whose core FSDBlock cascades three operators to address these failure modes in turn. Wavelet Convolution (WTConv) projects features into Haar sub-bands and applies independent low- and high-frequency kernels prior to inverse-DWT reconstruction, suppressing noise while preserving edges. Receptive-Field Attention Convolution (RFAConv) generates location-conditional kernels and so adapts to non-stationary scattering. Spatial Context Self-Attention (SCSA) aggregates discrete scattering points into coherent target representations via long-range grouped attention. At the fusion stage, CGAFusion replaces FPN element-wise addition with a channel–spatial–pixel triple-attention soft switch that mitigates deep–shallow semantic mismatch. On HRSID, FSDNet attains mAP₅₀ $= 92.3$% and mAP_50:95 $= 68.6$%. On SSDD, it attains mAP₅₀ $= 98.7$% and mAP_50:95 $= 74.2$%. Both sets of results consistently surpass the baseline methods. Against the strongest YOLO baseline (YOLOv11n), FSDNet improves HRSID mAP₅₀ by $+1.7$ percentage points (pp) and mAP_50:95 by $+2.3$ pp, and SSDD mAP₅₀ by $+0.5$ pp and mAP_50:95 by $+2.7$ pp; against the capacity-fair YOLOv11s reference (∼51% more parameters), FSDNet still leads on mAP₅₀, mAP_50:95, recall, and F1. Ablation studies and power-spectral-density analyses corroborate the contribution of each module and confirm WTConv’s role in preserving high-frequency target features. Full article

Frequency-hopping (FH) signals appear as small rectangular pulses in time-frequency spectrograms. At low signal-to-noise ratios (SNRs), noise along the frequency axis, caused by short-time Fourier transform (STFT) spectral leakage, blurs pulse boundaries, while the varying scales of hop rectangles exceed the capacity of a single receptive field. This paper presents YOLO11-FH, a modified YOLO11 detector that introduces two signal-processing-motivated modules. A FreqSmoothBlock (FSB) uses a $(3,1)$ depthwise convolution to smooth exclusively along the frequency axis, while adding only $5C$ parameters. A TFMultiResBlock (TFMRB) fuses three parallel dilated convolution branches (dilation rates of 1, 2, and 3) to cover different hop scales, replacing a heavier C3k2 module. The detection head is further simplified by halving the Bottleneck repeat count and disabling the deep submodule at the P5 scale. On a simulated FH dataset (SNRs ranging from $-15$ dB to $-10$ dB, five jamming types), YOLO11-FH achieves 96.04% mean average precision (mAP)@0.5 and 76.18% mAP@0.5:0.95, outperforming the YOLO11n baseline by 0.95 and 2.91 percentage points (pp) with 2.9% fewer parameters. Full article

Container surface damage detection is critical for ensuring the structural integrity and operational safety of intermodal freight transport. However, visual pseudo-textures arising from rust stains, specular reflections, and paint weathering cause frequent false positives, while the scarcity of puncture-type defects (Hole class) leads to missed detections. Existing YOLO-family detectors address neither the frequency-domain characteristics of such pseudo-textures nor the physical priors inherent in genuine structural damage. In this paper, we propose PPFS-YOLO, a physics-prior frequency-spatial fusion framework built upon YOLOv12s. Two lightweight modules are introduced: (1) Frequency-Spatial Fusion (FSF), which applies a learnable spectral mask in the Fourier domain and performs gated fusion with spatial features to suppress pseudo-texture responses; and (2) Edge-Guided Auxiliary Supervision Module (FIM), which encodes Sobel-derived edge priors as a differentiable $L_1$ constraint (${\mathcal{L}}_{\mathrm{phy}}$) to regularize feature learning toward physically plausible damage boundaries. Three pairs of FSF–FIM are inserted into the YOLOv12s neck and head at P3, P4, and P4-head scales. Experiments on a container damage dataset containing 7013 images and three classes (Dent, Hole, Rusty) demonstrate that PPFS-YOLO achieves 64.86% mAP@50, a +12.35 percentage-point improvement over the YOLO12s baseline (SGD, unified optimizer), with only +0.79 M additional parameters (+8.6%) and a modest latency overhead of 2.9 ms (17.2 ms vs. 14.3 ms at $640\times 640$ on an NVIDIA RTX 3090 GPU (NVIDIA Corporation, Santa Clara, CA, USA)). Ablation analysis reveals that ${\mathcal{L}}_{\mathrm{phy}}$ is the critical catalyst: without it, the combined FSF+FIM modules yield only +0.83 pp, whereas the full model achieves +12.10 pp—underscoring the synergy between frequency-domain fusion and physics-prior regularization. Full article

Old-vine vineyards often contain dozens of grapevine varieties intermingled and irregularly distributed, making plant-level varietal identification slow and expensive when based on ampelography or molecular approaches. This paper proposes a field-oriented computer-vision pipeline for Vitis vinifera variety identification using images with a natural background from the historic “Vinha Maria Teresa” parcel (Quinta do Crasto, Portugal). A single-class YOLO11 detector is trained to localize the vine leaf and generate standardized crops, and a YOLO11 classifier is then fine-tuned on leaf regions of interest (ROIs) for eight selected varieties in the Douro UNESCO region. We annotated 2015 vineyard images for classification and supplemented detection training with 2648 additional leaf images; detectors (YOLO11n/s/m) were benchmarked under four augmentation regimes and evaluated on a fixed 48-image subset, including runtime on CPU and GPU. The best detector reached mAP@50–95 of 0.918 on the benchmark, while YOLO11n achieved ∼27 FPS on CPU for fast cropping. On a 303-image test set, the best classifier (YOLO11s with mixed augmentations) achieved 94.06% Top-1 accuracy, 93.92% macro-F1, and 100% Top-5 accuracy with remaining errors concentrated among morphologically similar varieties. To assess deployment-oriented performance, classifiers trained under three input settings (manual crops, detector-generated crops, and full images) were evaluated on a held-out 48-image benchmark subset; removing the detection step reduced Top-1 accuracy from 75.00% to 68.75%, while the gap between manual and automatic crops was only 2.44 pp on successfully detected images with detection failures (14.6%) representing the primary operational bottleneck. Repeated retraining of the best manual-crop YOLO11s configuration across multiple random seeds showed stable performance with low variability in Top-1 accuracy and macro-F1. Under identical training conditions, ResNet50 and EfficientNet-B0 provided competitive baselines, but YOLO11s remained the strongest overall model on the held-out field benchmark. These results indicate that lightweight leaf detection plus crop-based classification can support scalable varietal identification in old vineyards under realistic acquisition conditions. Full article

Continuous unidirectional (UD) thermoplastic composite tapes are increasingly used in aerospace, automotive, and energy applications because of their high specific strength, low weight, recyclability, and compatibility with automated manufacturing. Since final component performance strongly depends on tape quality, reliable defect characterization and detection are essential. In this study, manufacturing-induced defects in polypropylene-based UD tapes reinforced with carbon and glass fibers were investigated using real images acquired directly from laboratory-scale production without synthetic data. Defects related to interfacial integrity, matrix distribution, fiber architecture, and surface irregularities were systematically analyzed, and a practical four-class defect taxonomy was established. To enable automated inspection under limited-data conditions, lightweight YOLOv8, YOLOv11, and the new YOLO26 models were comparatively evaluated using a UD tape-specific augmentation strategy combining physically constrained Albumentations and on-the-fly augmentation. Among the tested models, YOLO26-s achieved the best overall performance, reaching a mean mAP@0.5 of 0.87 ± 0.03, outperforming YOLOv11 (0.83) and YOLOv8 (0.78), with 0.90 precision and 0.85 recall. Interfacial (0.92 mAP) and matrix-related (0.90 mAP) defects were detected most reliably, whereas fiber-related (0.89 mAP) and surface defects (0.79 mAP) remained more challenging, particularly in glass-fiber-reinforced tapes due to transparency-masking effects. The results demonstrate the potential of compact deep learning models for computationally efficient and manufacturing-relevant in-line quality monitoring of UD tape production. Full article

Exogenous fires in underground coal mines are characterized by low illumination, smoke occlusion, heavy dust loading and pseudo fire sources, which jointly degrade image quality and cause missed and false alarms in visual detection. To achieve accurate and real-time early warning under such conditions, this paper proposes a mine exogenous fire detection algorithm based on an improved YOLOv9m, termed PPL-YOLO-F-C. First, a lightweight PP-LCNet backbone is embedded into YOLOv9m to reduce the number of parameters and GFLOPs while maintaining multi-scale feature representation suitable for deployment on resource-constrained edge devices. Second, a Fully Connected Attention (FCAttention) module is introduced to perform fine-grained frequency–channel calibration, enhancing discriminative flame and smoke features and suppressing low-frequency background clutter and non-flame textures. Third, the original upsampling operators in the neck are replaced by the CARAFE content-aware dynamic upsampler to recover blurred flame contours and tenuous smoke edges and to strengthen small-object perception. In addition, an MPDIoU-based bounding-box regression loss is adopted to improve geometric sensitivity and localization accuracy for small fire spots. Experiments on a self-constructed mine fire image dataset comprising 3000 samples show that the proposed PPL-YOLO-F-C model achieves a precision of 97.36%, a recall of 84.91%, mAP@50 of 96.49% and mAP@50:95 of 76.6%, outperforming Faster R-CNN, YOLOv5m, YOLOv7 and YOLOv8m while using fewer parameters and lower computational cost. The results demonstrate that the proposed algorithm provides a robust and efficient solution for real-time exogenous fire detection and edge deployment in complex underground mine environments. Full article

Detecting low-altitude, slow-speed, small (LSS) UAVs is especially challenging in low-visibility scenes (low light, haze, motion blur), where inherent uncertainties in sensor data and object appearance dominate. We propose GAME-YOLO, a novel detector that integrates a Bayesian-inspired probabilistic reasoning framework with Global Attention and Multi-Scale Enhancement to improve small-object perception and sub-pixel-level localization. Built on YOLOv11, our framework comprises: (i) a visibility restoration front-end that probabilistically infers and enhances latent image clarity; (ii) a global-attention-augmented backbone that performs context-aware feature selection; (iii) an adaptive multi-scale fusion neck that dynamically weights feature contributions; (iv) a sub-pixel-aware small-object detection head (SOH) that leverages high-resolution feature grids to model sub-pixel offsets; and (v) a novel Shape-Aware IoU loss combined with focal loss. Extensive experiments on the LSS2025-DET dataset demonstrate that GAME-YOLO achieves state-of-the-art performance, with an AP@50 of 52.0% and AP@[0.50:0.95] of 32.0%, significantly outperforming strong baselines such as LEAF-YOLO (48.3% AP@50) and YOLOv11 (36.2% AP@50). The model maintains high efficiency, operating at 48 FPS with only 7.6 M parameters and 19.6 GFLOPs. Ablation studies confirm the complementary gains from our probabilistic design choices, including a +10.5 pp improvement in AP@50 over the baseline. Cross-dataset evaluation on VisDrone-DET2021 further validates its generalization capability, achieving 39.2% AP@50. These results indicate that GAME-YOLO offers a practical and reliable solution for vision-based UAV surveillance by effectively marrying the efficiency of deterministic detectors with the robustness principles of Bayesian inference. Full article

Insulators serve as critical insulating components in power transmission lines, and their defects are one of the primary causes of power outages in power grids. Power companies widely utilize unmanned aerial vehicle (UAV) inspections to collect image data of power transmission lines. However, existing methods face two core challenges: at the data level, insulator defect samples are extremely scarce in massive image datasets, leading to severe data imbalance issues. At the algorithm level, deep learning-based defect detection methods rely on data-driven feature extraction, ignoring quantifiable prior knowledge such as insulator installation specifications and mechanical structure. This factor results in low localization efficiency and poor robustness in complex scenarios. To address these issues, this paper proposes an insulator defect detection method based on Vision–Language models and prior knowledge. It extracts prior knowledge about the physical characteristics of insulators, quantifies spatial structure and installation specifications as prior constraints, embeds prior knowledge into the vision–language model’s feature space to generate insulator defect samples, addresses the data imbalance issue, and detects insulator defects using an improved You Only Look Once (YOLO) algorithm. This approach reduces model parameters while maintaining detection accuracy, constructing a lightweight model for insulator defect detection. The experimental results show that, compared with PP-YOLOE-m and RT-DETR-R18 models, the method proposed in this paper can significantly improve the detection accuracy. The mean average precision indicator of the model in this paper has reached 95.7%. Full article

In the domain of automatic parking systems, parking space detection and localization represent fundamental challenges that must be addressed. As a core research focus within the field of intelligent automatic parking, they constitute the essential prerequisite for the realization of fully autonomous parking. Accurate and effective detection of parking spaces is still the core problem that needs to be solved in automatic parking systems. In this study, building upon existing public parking space datasets, a comprehensive panoramic parking space dataset named PSEX (Parking Slot Extended) with complex environmental diversity was constructed by integrating the concept of GAN (Generative Adversarial Network)-based image style transfer. Meanwhile, an improved algorithm based on PP-Yoloe (Paddle-Paddle Yoloe) is used to detect the state (free or occupied) and angle (T-shaped or L-shaped) of the parking space in real-time. For the many and small labels of the parking space, the ResSpp in it is replaced by the ResSimSppf module, the SimSppf structure is introduced at the neck end, and Silu is replaced by Relu in the basic structure of the CBS (Conv-BN-SiLU), and finally an auxiliary detector head is added at the prediction head. Experimental results show that the proposed SimSppf_mepre-Yoloe model achieves an average improvement of 4.5% in mAP50 and 2.95% in mAP50:95 over the baseline PP-Yoloe across various parking space detection tasks. In terms of efficiency, the model maintains comparable inference latency with the baseline, reaching up to 33.7 FPS on the Jetson AGX Xavier platform under TensorRT optimization. And the improved enhancement algorithm can greatly enrich the diversity of parking space data. These results demonstrate that the proposed model achieves a better balance between detection accuracy and real-time performance, making it suitable for deployment in intelligent vehicle and robotic perception systems. Full article

In cold and highly humid regions, wind turbine blades (WTB) are susceptible to icing, which can have a significant impact on the security and economic operation of turbines. Therefore, precise and prompt icing status detection is pivotal for maintaining wind turbine operational normalcy. In this research, an improved PP-YOLOE network is developed for classifying and detecting the icing state of WTB. First, a dataset of WTB icing is constructed based on a wind tunnel laboratory and expanded to improve the generalization of the model. To enhance feature representation, the network architecture was improved by embedding a coordinate attention (CA) mechanism and integrating atrous spatial pyramid pooling (ASPP) to better capture multi-scale contextual information. Moreover, a key innovation of this work is the novel application of a particle swarm optimization (PSO) algorithm to systematically automate hyperparameter tuning. Through ablation experiments and comparative tests, the improved PP-YOLOE network demonstrates superior overall performance on this dataset, achieving a multiple average precision of 0.94. It surpasses the original model across multiple evaluation metrics, indicating a robust and meaningful enhancement. The improved PP-YOLOE network proposed in this study provides a promising and effective method for WTB icing detection. This work provides a paradigm for applying advanced deep learning techniques to enhance intelligent industrial inspection tasks. Full article

Timely detection and prevention of citrus leaf diseases and pests are crucial for improving citrus yield. To address the issue of low efficiency in citrus disease and pest detection, this paper proposes a lightweight detection model named PEW-YOLO. First, the PP-LCNet backbone is optimized using a novel GSConv convolution, and a lightweight PGNet backbone is introduced to reduce model parameters while enhancing detection performance. Next, the C2f_EMA module, which integrates efficient multi-scale attention (EMA), replaces the original C2f module in the neck, thereby improving feature fusion capabilities. Finally, the Wise-IoU loss function is employed to address the challenge of identifying low-quality samples, further improving both convergence speed and detection accuracy. Experimental results demonstrate that PEW-YOLO achieves a 1.8% increase in mAP50, a 32.2% reduction in parameters, and a detection speed of 1.6 milliseconds per frame on the citrus disease and pest dataset, thereby meeting practical real-time detection requirements. Full article

In object detection networks, one widely used and effective approach to address the challenge of detecting small-sized objects in images is to employ multiscale pyramid features for prediction. Based on the fundamental principles of pyramid feature extraction, shallow features with small receptive fields are responsible for predicting small-sized objects, while deep features with large receptive fields handle large-sized objects. However, during the actual network training process using this structure, the loss function only provides the error between all positive samples and labels, treating them equally without considering the relationship between the actual size of the label and the feature layer where the sample resides, which to some extent affects the object detection performance. To address this, this paper proposes a novel method for constructing a loss function, termed Predicted Probability Loss (PP-Loss). It determines the probability of each feature layer predicting the objects labeled by the labels based on the size of the labels and uses this probability to adjust the weights of different sample anchors in the loss function, thereby guiding the network training. The prediction probability values for each feature layer are obtained from a prediction probability function established on a statistical basis. The algorithm has been experimentally validated on different networks with YOLO as the core. The results show that the convergence speed and accuracy of the network during training have been improved to varying degrees. Full article