Deep learning-based defect detection of transmission line insulation equipment helps enhance power grid stability and inspection personnel safety. However, existing research still falls short of meeting practical inspection requirements [ 1 ]. To address this issue, this paper proposes a novel Is-YOLO model to improve defect detection accuracy in complex aerial inspection scenarios. The designed C3k2_IDC module adopts a multi-branch parallel structure with diverse convolutional kernels, expanding the receptive field while preserving computational efficiency. The C3k2_StarsBlock module leverages star operations to capture high-dimensional features and further promote detection accuracy. In addition, a newly designed P2/4 tiny-target detection head achieves substantial improvement in small-object detection performance. Experimental results on a UAV-captured dataset of transmission line insulator defects demonstrate that the proposed Is-YOLO model outperforms YOLOv11-n with only an 11.5% increase in model parameters. Its mAP50 rises from 84.5% to 88.7%, mAP75 from 62.3% to 67.1%, and mAP50–95 from 61.4% to 65%. With moderate computational overhead and significant performance gains, Is-YOLO can serve as an efficient solution for insulator defect detection.

Coprime array enables derivation of an extended array with the number of virtual elements beyond the number of physical sensors, resolving the underdetermined direction of-arrival (DOA) estimation problem while mitigating mutual coupling effects. However, the virtual array is discontinuous and contains some holes. To fully exploit the information of the virtual array, conventional methods typically employ complex matrix completion (MC) to iteratively recover the data of the covariance matrix, resulting in high complexity, particularly with large-scale coprime arrays. In this letter, a real-valued DOA estimation via the construction of Toeplitz covariance matrix and spline interpolation is proposed. The algorithm complexity is reduced by real-valued Toeplitz matrix, and then the holes are filled by spline interpolation. The recovered matrix can be used for DOA estimation by conventional subspace approaches. The superior computational performance of our approach is validated through comparative simulations with established methods.

Time-delay estimation (TDE) holds great significance in pavement surveys, especially in the modern transportation system. Established on the ideal Dirac pulse and white Gaussian noise, the performance of the existing TDE methods may degrade in practical ground penetrating radar (GPR) detection, where radar pulse and noise distribution are diverse. In this letter, we develop a gridless TDE method considering the radar pulse and noise pattern in GPR detection. Co-prime sampling strategy is applied to reduce the number of frequency samples compared with conventional uniform sampling. With the prior knowledge of radar pulse and noise distribution, an enhanced measurement vector is generated from the data covariance matrix, thus improving the signal quality compared with the conventional methods which are based on ideal Dirac pulse and white Gaussian noise. Subsequently, the time-delays are estimated by the proposed atomic norm minimization (ANM) method, where the complexity is further reduced compared with the previous works using multiple measurements. Simulation results show the advantages of the proposed method in terms of running time, weak echo detection, and estimation accuracy.

Recently, direction-of-arrival (DOA) estimation in impulsive noise scenarios has been extensively investigated in the field of co-prime array signal processing. This paper proposes a combined enhanced bounded nonlinear covariance and phased fractional low-order moment (EBNC-PFLOM) method, which incorporates the advantages of both EBNC and PFLOM and mitigates the impulsive noise by constructing the equivalent data covariance matrix of the received signals. Furthermore, when dealing with a limited number of input signals, the proposed method is capable of directly estimating the signals’ DOA without spatial smoothing. Simulation results show that the proposed method outperforms the recently reported methods.

In this paper, we propose a nested sampling based off-grid block sparse representation method (Nested-OGBSR) for time-delay estimation (TDE) of coherent ground penetrating radar (GPR) backscattered echoes. Nested sampling strategy is taken to reduce the sampling rate and computational burden. The off-grid data model is adopted to eliminate the effect of basis mismatch caused by the predefined grids in sparse representation (SR) and thus improve the estimation accuracy. The non-circularity of GPR signals is also utilized to enhance the temporal resolution of sparse block representation (BSR). Numerical and experimental results are provided to show the superiority of the proposed method in terms of estimation accuracy, temporal resolution and computational complexity.

Coprime array possesses a larger array aperture and element spacing compared with the conventional uniform linear array (ULA) for the equivalent number of sensors, attracting considerable scholarly attention. However, the direction of arrival (DOA) estimation of coherent signals has been a major challenge for the practical application of the coprime array. In this paper, based upon the perspective of signal subspace reconstruction, we propose two effective approaches to resolve the DOA of completely correlated signals with a coprime array. For the first method, we exploit two selection matrices to separate the signal subspace into two parts and rearrange the elements within them to construct two Hankel matrices. By applying the MUSIC method to these Hankel matrices and finding common solutions, we can determine the DOA of coherent signals. In the second method, we first convert the signal subspace of the coprime array into the signal subspace of ULA using a mapping matrix and the total least squares method. We then construct a Hankel matrix and restore its rank by solving a rank minimization problem. Finally, by applying the MUSIC method to the rank-restored Hankel matrix, we can obtain the angles of the coherent signals. Finally, simulation results are presented to demonstrate the efficiency and superiority of our proposed methods.

Time-delay estimation (TDE) for thin top layers of asphalt pavement is a challenging task due to the limited resolution of ground penetrating radar (GPR) as well as small permittivity difference between top layers. Echoes backscattered from the interfaces of top layers with similar permittivity have usually much smaller amplitudes compared with other echoes, which can be called weak signals. The weak backscattered echoes are usually too sensitive to the noise and other strong echoes that current signal processing approaches (subspace-based methods and compressed sensing based methods) might have false estimation results even failures without proper processing of them. Therefore, in this paper, an enhanced orthogonal matching pursuit (OMP) method is proposed to deal with weak signals resulting from similar permittivity of adjacent asphalt layers. Based on the orthogonality between signal and noise subspaces, we firstly apply the truncated singular value decomposition (SVD) on the received signals, in order to reduce the noise impact. Secondly, we build an orthogonal matrix to the mode matrix of the pre-estimated strong backscattered echoes, and map it to the overcomplete dictionary matrix, such that the influence of the residual of the strong backscattered echoes can be reduced. Finally, the time-delays of backscattered echoes and layer thicknesses are estimated. Compared with conventional approaches, the proposed method is more suitable for TDE in thin asphalt pavement detection. The accuracy of the proposed method is validated by both numerical and experimental data.

Named entity recognition (NER) is usually developed and tested on text from well-written sources. However, in intelligent voice assistants, where NER is an important component, input to NER may be noisy because of user or speech recognition error. In applications, entity labels may change frequently, and non-textual properties like topicality or popularity may be needed to choose among alternatives. We describe a NER system intended to address these problems. We test and train this system on a proprietary user-derived dataset. We compare with a baseline text-only NER system; the baseline enhanced with external gazetteers; and the baseline enhanced with the search and indirect labelling techniques we describe below. The final configuration gives around 6% reduction in NER error rate. We also show that this technique improves related tasks, such as semantic parsing, with an improvement of up to 5% in error rate.

Pavement survey is one of the most important applications for ground penetrating radar (GPR) in civil engineering. In the case of centimeter scale of GPR waves, the influence of interface roughness cannot be neglected and should be taken into account in the radar data model. The objective of this article is to estimate the time-delay in the presence of interface roughness by GPR. Using the property of noncircular signals, we propose a modified orthogonal matching pursuit method to estimate the pavement parameters for both overlapped and nonoverlapped echoes. Compared with subspace-based methods in coherent scenarios, the proposed method can estimate the time delays of backscattered echoes without applying the cumbersome interpolation and spatial smoothing procedures, which are more practical in real applications. The performance of the proposed method is tested on both simulated and experimental data. The estimation results show the good performance of the proposed method.

The classification framework has been popularly adopted to perform sound event detection. However, the existing neural network based classification based approaches treat each feature dimension equally and the varying influence of feature dimensions has not been taken into consideration. To deal with this, we propose a feature space attention based convolution recurrent neural network approach utilizing the varying importance of each feature dimension to perform acoustic event detection. The convolution layers are used to extract the high level information from the audio signals. Then the feature space attention scheme is applied to the extracted features to automatically determine the importance of each feature dimension. Experimental results on the latest TUT Sound Event 2017 dataset demonstrate the improved performance of the proposed approach compared to the existing acoustic event detection systems.

Ground Penetrating Radar (GPR) is a widely used tool in the management and monitoring of pavement structures. In this paper, we focus on the detection of thin inter-layer debondings between the hot mix asphalt layers of pavement structures. A Stepped-Frequency Radar (SFR) associated with a Support Vector Regression based Linear Prediction (LP-SVR) method is used to detect thin debondings. The performance of SFR with the LP-SVR method is analyzed according to various used frequency bandwidths on the experimental data.

In civil engineering, roadway structure evaluation is an important application which can be carried out by ground penetrating radar. This paper focuses on the estimation of the time delay and interface roughness of civil engineering structure, like pavements. The influence of interface roughness is taken into account in the signal model. Therefore, we propose a new method which allows to efficiently estimate the time delay and interface roughness. Like in [1], the modified MUSIC is used for time delay estimation. In interface roughness estimation, we propose a modified orthogonal propagator method (OPM) to estimate the interface roughness with estimated time delay. While in [1], maximum likelihood method is applied, which needs multiple dimensional search. The proposed method is tested on the experimental data. The experimental results show the performance of the proposed method.

In the field of civil engineering, ground penetrating radar (GPR) is a highly efficient nondestructive testing tool for sustainable management of pavement infrastructures. GPR allows to evaluate the structure of the roadway over large distances (with contactless configurations) and to detect significant subsurface defects. This letter presents a new method to detect thin debondings within pavement structures with the step-frequency GPR. The proposed method enables us to carry out the detection with only a small number of frequency samples and A-scans. It is based on the linear prediction and support vector regression theories. Two experimental results show its effectiveness.

The development of methods and tools for the road infrastructure sustainable management is a research challenge, especially for nondestructive testing methods. This letter focuses on the estimation of the thickness of civil engineering structures, like pavements, and more precisely, the time delay and interface roughness. We propose a modified Min-Norm algorithm which allows efficiently estimating the time delay and interface roughness without the eigenvalue decomposition. Therefore, it has a smaller computational load compared with subspace-based methods. The experimental results show the efficiency of the proposed algorithm.

Ground-penetrating radars (GPR) are widely used in media parameters' estimation and targets' localization. This paper focuses on time-delay estimation (TDE) using the GPR signal, which contains important information about the probed media structure. However, TDE tends to be a challenging task in GPR applications, in the scenarios of overlapping, coherent signals and limited snapshots. Forward-backward linear prediction (FBLP) is a high time-resolution method, which is able to directly deal with coherent signals. Support vector regression (SVR) is robust with small samples. Therefore, we propose to combine the theory of FBLP and SVR together to enhance the robustness of TDE in the case of coherent, overlapping signals as well as limited snapshots. The proposed method is tested with both numerical and experimental data. Both the results demonstrate the effectiveness of the proposed method.

In civil engineering, ground penetrating radar is a common technique for evaluating the structure and quality of road pavement. This paper focuses on the estimation of the time delay and interface roughness of civil engineering structure, like pavements. The influence of interface roughness is taken into account in the signal model. A modified estimation of signal parameters via rotational invariance technique (ESPRIT) algorithm combined with an interpolated spatial smoothing technique is proposed. It allows us to jointly and efficiently estimate the time delay and interface roughness by ultrawideband radar (the upper frequency up to 8-10 GHz) with low computational complexity. The proposed algorithm is tested on both numerical and experimental data. Simulation and experimental results show the good performance of the proposed algorithm.

We tackle the problem of joint discovery and segmentation of the object of interest from noisy image sets collected via web crawling (e.g., Figure 1). Existing methods [1] [2] [3] employ region-wise comparison in order to separate noise images (images not containing target objects) from the rest, which may be a bottleneck for scaling up to larger datasets. Our idea to avoid such computationally intensive operations is to use image-wise reconstruction errors. Specifically, based on the assumption that images containing target objects are easier to be reconstructed by a pool of target objects than noise images, we first reconstruct each image using a small number of similar target objects. The resulting error is then combined with some other criteria (e.g., saliency) so as to delineate only target object regions. Experimental evaluations on a noisy image dataset [1] demonstrate that our approach achieves state-of-the-art results on every subset of the dataset 5-7 times faster than existing methods.

The ill-posed inversion problem is to be solved urgently. Priori knowledge is introduced to increase the inversion information to improve the inversion quality. MODIS time-series leaf area index (LAI) data is used to extract the priori knowledge. Savitzky-Golay (SG) filter and bi-Gaussian curve-fit are performed to reconstruct the original time-series LAI data, then, the upper envelope of the smoothed time-series curve is achieved to get a fixed range of LAI in a certain growing season. The variation of LAI was restricted with the range based on Look-up Table (LUT) method with the PROSAIL model. The validation results show that the priori knowledge extracted from LAI time-series data is efficiency on improving the LAI inversion precision.

This paper extends the standard support vector regression (SVR) into multi-dimensional case to estimate different biophysical parameters simultaneously. The improvement is made by the Vapnik loss function of L2 form. The proposed multi-output SVR (MO-SVR) is implemented in the joint inversion of Leaf Area Index (LAI) and vegetation cover fraction (fCover) over SPOT2/HRV1 data in the Fundulea site (VALERI), Romania. Comparison between standard SVR and MO-SVR, and the validation using biophysical maps both indicate the better fitness and accuracy of MO-SVR than the traditional form.