A Simple and Effective Surface Defect Detection Method of Power Line Insulators for Difficult Small Objects

Insulator defect detection plays a vital role in maintaining the secure operation of power systems.To address the issues of the difficulty of detecting small objects and missing objects due to the small scale,variable scale,and fuzzy edge morphology of insulator defects,we construct an insulator dataset with 1600 samples containing flashovers and breakages.Then a simple and effective surface defect detection method of power line insulators for difficult small objects is proposed.Firstly,a high-resolution featuremap is introduced and a small object prediction layer is added so that the model can detect tiny objects.Secondly,a simplified adaptive spatial feature fusion(SASFF)module is introduced to perform cross-scale spatial fusion to improve adaptability to variable multi-scale features.Finally,we propose an enhanced deformable attention mechanism(EDAM)module.By integrating a gating activation function,the model is further inspired to learn a small number of critical sampling points near reference points.And the module can improve the perception of object morphology.The experimental results indicate that concerning the dataset of flashover and breakage defects,this method improves the performance of YOLOv5,YOLOv7,and YOLOv8.In practical application,it can simply and effectively improve the precision of power line insulator defect detection and reduce missing detection for difficult small objects.

Principle and realization method of tunnel deformation detection based on image recognition and data transmission technology

基于图象传感器网络检测隧道结构的变丑是空间传感器技术的先进学习和应用程序。为内部、外部的应力在它的长时期操作以后引起的 metro 隧道的垂直解决，全面计划和隧道变丑察觉系统的测量原则被介绍。图象数据获得并且察觉目标处理被图象传感器的合作工作完成，手臂嵌入的系统。RS485 通讯完成在手臂存储器和主机计算机之间的数据传播。在车站站台的数据库系统分析察觉数据并且获得变丑状态掘内部墙，它使它可能到早警告隧道变丑并且及时采取预防措施。

LDNet:structure-focused lane detection based on line deformation

Lane detection is a fundamental necessary task for autonomous driving.The conventional methods mainly treat lane detection as a pixel-wise segmentation problem,which suffers from the challenge of uncontrollable driving road environments and needs post-processing to abstract the lane parameters.In this work,a series of lines are used to represent traffic lanes and a novel line deformation network(LDNet) is proposed to directly predict the coordinates of lane line points.Inspired by the dynamic behavior of classic snake algorithms,LDNet uses a neural network to iteratively deform an initial lane line to match the lane markings.To capture the long and discontinuous structures of lane lines,1 D convolution in LDNet is used for structured feature learning along the lane lines.Based on LDNet,a two-stage pipeline is developed for lane marking detection:(1) initial lane line proposal to predict a list of lane line candidates,and(2) lane line deformation to obtain the coordinates of lane line points.Experiments show that the proposed approach achieves competitive performances on the TuSimple dataset while being efficient for real-time applications on a GTX 1650 GPU.In particular,the accuracy of LDNet with the annotated starting and ending points is up to99.45%,which indicates the improved initial lane line proposal method can further enhance the performance of LDNet.

Single Epoch GPS Deformation Signals Extraction and Gross Error Detection Technique Based on Wavelet Transform

Wavelet theory is efficient as an adequate tool for analyzing single epoch GPS deformation signal. Wavelet analysis technique on gross error detection and recovery is advanced. Criteria of wavelet function choosing and Mallat decomposition levels decision are discussed. An effective deformation signal extracting method is proposed, that is wavelet noise reduction technique considering gross error recovery, which combines wavelet multi-resolution gross error detection results. Time position recognizing of gross errors and their repairing performance are realized. In the experiment, compactly supported orthogonal wavelet with short support block is more efficient than the longer one when discerning gross errors, which can obtain more finely analyses. And the shape of discerned gross error of short support wavelet is simpler than that of the longer one. Meanwhile, the time scale is easier to identify.

基于改进Deformable-DETR的水下图像目标检测方法

针对由于水下复杂环境造成的目标检测效果较差、检测精度较低的问题,基于Deformable-DETR算法提出一种改进的水下目标检测算法Deformable-DETR-DA。使用空间注意力模块结合标准Transformer块设计了一个用于增加模型深度的深度特征金字塔(deep feature pyramid networks,DFPN)模块,将其嵌入到模型中提高模型对深层纹理信息的提取能力。使用注意力引导的方式对原模型中编码器部分进行改进,加强了对特征信息的聚合能力,提高了模型在复杂环境下的检测能力。针对URPC数据集,模型各交并比尺度的平均准确度(average precision,AP)为39.5%,相比原模型提升1%,与一些DETR(detection transformer)类的模型相比,不同目标尺度的平均准确度均有1%~4%左右的提高,表明改进的模型能够很好解决复杂环境的水下目标检测的问题。本文提出的模型可作为其他水下目标检测模型设计的参考。

RFI Detection for Multichannel HRWS SAR System Based on Spatial Cross Correlation

Multichannel high-resolution and wide-swath(HRWS)imaging is an advanced digital beamforming technique for future synthetic aperture radar(SAR)systems.However,radio frequency interference(RFI)is a critical concern for HRWS SAR missions,which distorts measure-ments and produces image artifacts.In this paper,the spatial cross-correlation coefficients of multichannel HRWS SAR signals are investigated for RFI detection.It is found when the two channels are correlated,RFI-polluted areas present lower coherence values than non-polluted areas in the same scenarios,which makes previous methods fail.Further,this paper studies the case of two fully decorrelated channels to maximize the coherence difference among RFI and target echoes,and RFI detection is realized by exploiting the anomaly value of coherence.Experimental results of real air-borne multichannel SAR data demonstrate that the RFI can be detected successfully.

Research on Facial Fatigue Detection of Drivers with Multi-feature Fusion

In order to solve the shortcomings of current fatigue detection methods such as low accuracy or poor real-time performance,a fatigue detection method based on multi-feature fusion is proposed.Firstly,the HOG face detection algorithm and KCF target tracking algorithm are integrated and deformable convolutional neural network is introduced to identify the state of extracted eyes and mouth,fast track the detected faces and extract continuous and stable target faces for more efficient extraction.Then the head pose algorithm is introduced to detect the driver’s head in real time and obtain the driver’s head state information.Finally,a multi-feature fusion fatigue detection method is proposed based on the state of the eyes,mouth and head.According to the experimental results,the proposed method can detect the driver’s fatigue state in real time with high accuracy and good robustness compared with the current fatigue detection algorithms.

基于改进Deformable DETR的无人机视频流车辆目标检测算法

针对无人机视频流检测中小目标数量多、因图像传输质量较低而导致的上下文语义信息不充分、传统算法融合特征推理速度慢、数据集类别样本不均衡导致的训练效果差等问题,提出一种基于改进Deformable DETR的无人机视频流车辆目标检测算法。在模型结构方面,该算法设计了跨尺度特征融合模块以增大感受野,提升小目标检测能力,并采用针对object_query的挤压-激励模块提升关键目标的响应值,减少重要目标的漏检与错检率;在数据处理方面,使用了在线困难样本挖掘技术,改善数据集中类别样本分布不均的问题。在UAVDT数据集上进行了实验,实验结果表明,改进后的算法相较于基线算法在平均检测精度上提升了1.5%,在小目标检测精度上提升了0.8%,并在保持参数量较少增长的情况下,维持了原有的检测速度。

Traffic Accident Detection Based on Deformable Frustum Proposal and Adaptive Space Segmentation

Road accident detection plays an important role in abnormal scene reconstruction for Intelligent Transportation Systems and abnormal events warning for autonomous driving.This paper presents a novel 3D object detector and adaptive space partitioning algorithm to infer traffic accidents quantitatively.Using 2D region proposals in an RGB image,this method generates deformable frustums based on point cloud for each 2D region proposal and then frustum-wisely extracts features based on the farthest point sampling network(FPS-Net)and feature extraction network(FE-Net).Subsequently,the encoder-decoder network(ED-Net)implements 3D-oriented bounding box(OBB)regression.Meanwhile,the adaptive least square regression(ALSR)method is proposed to split 3D OBB.Finally,the reduced OBB intersection test is carried out to detect traffic accidents via separating surface theorem(SST).In the experiments of KITTI benchmark,our proposed 3D object detector outperforms other state-of-theartmethods.Meanwhile,collision detection algorithm achieves the satisfactory performance of 91.8%accuracy on our SHTA dataset.

Single-Shot 360-Degree Cranial Deformity Detection System Using Digital Image Correlation

In this paper,a single-shot 360-degree cranial deformity detection system using digital image correlation(DIC)is presented to quickly obtain and detect accurate 3D data of infants’cra-nium.By introducing plane mirrors into a stereo 3D DIC measurement system,a multi-view 3D imaging model is established to convert 3D data from real and virtual perspectives into 360-degree 3D data of the tested infant cranium,achieving single-shot and panoramic 3D measurement.Exper-imental results showed that the performance and measurement accuracy of the proposed system can meet the requirements for cranial deformity detection,which provides a fast,accurate,and low-cost solution medically.

基于Deformable DETR的自然场景任意形状文本检测

自然场景下的文本区域形状复杂多变,直接使用轮廓坐标描述文本区域会使得建模不充分,导致文本检测准确性低。针对自然场景下文本区域不规则的问题,提出了一种基于Deformable DETR的任意形状文本检测模型,不同于传统的直接预测轮廓点的方法,使用B-样条对文字区域进行建模使得文本轮廓平滑精确的同时减少了需要预测的参数。提出的文本检测模型无需手工设计锚点、区域建议等组件,极大地简化了模型设计并提高了通用性。提出的模型在无需额外数据集的情况下在任意形状文本数据集CTW1500和Total-Text上的平均精度(F值)分别达到了85.4%和85.0%,证明了模型的有效性。

基于Deformable DETR的红外图像目标检测方法研究

基于Transformer架构的DETR系列网络在计算机视觉目标检测领域不断刷新目标检测的精度与速度。然而,基于红外图像的非合作目标检测的应用环境复杂,而且红外图像质量较差。针对该问题,提出了一种新的以Deformable DETR算法为基线的具有高检测精度的目标检测算法:首先设计了对红外图像进行图像增强处理的图像增强模块CLAHE-GB,并将其与Deformable DETR进行有机结合;然后在大型通用数据集上进行预训练;最后引入数据增强和迁移学习方法在自制的空中飞行物小型红外图像数据集中对检测头网络参数进行再训练,并对结果进行分析。结果表明:所提出的算法对红外图像数据具有较好的图像增强效果和检测精度。

Using LBG quantization for particle-based collision detection algorithm

Most collision detection algorithms can be efficiently used only with solid and rigid objects, for instance, Hierarchical methods which must have their bounding representation recalculated every time deformation occurs. An alternative algorithm using particle-based method is then proposed which can detect the collision among non-rigid deformable polygonal models. However, the original particle-based collision detection algorithm might not be sufficient enough in some situations due to the improper particle dispersion. Therefore, this research presents an improved algorithm which provides a particle to detect in each separated area so that particles always covered all over the object. The surface partitioning can be efficiently performed by using LBG quantization since it can classify object vertices into several groups base on a number of factors as required. A particle is then assigned to move between vertices in a group by the attractive forces received from other particles on neighbouring objects. Collision is detected when the distance between a pair of corresponding particles becomes very small. Lastly, the proposed algo- rithm has been implemented to show that collision detection can be conducted in real-time.

A new approach for distinguishing different deformation trend blocks with displacement observations

The process for dividing the different deformation trend blocks with displacement observations includes three steps. They are datum detection, block scope told part and anomalous deformations detection in blocks. The three steps are implemented by Quasi-Accurate Detection(QUAD) in the paper. In the previous two steps, the prelimi-nary selection for Quasi-Accurate Observations (QAOs) is key. The preliminary selection is according to the size of deformation displacement for datum detection and according to the direction of deformation for block scope told part. At last through an example, each implementation process is introduced simply and the detection effect of QUAD is compared with that of the robust estimation (Huber) and the statistic test. The result indicates that the three steps can be implemented successfully with QUAD and that the anomalous deformations in blocks can be detected, but the steps of the datum detection and block scope told part are failed by robust estimation. The detec-tions of three steps are failure by the statistic test. The results show that the QUAD has the virtues that the location of gross errors is much accurate and the breakdown point is higher than the other two methods.

Application of wavelet analysis to crustal deformation data processing

The time-frequency analysis and anomaly detection of wavelet transformation make the method irresistibly advan- tageous in non-stable signal processing. In the paper, the two characteristics are analyzed and demonstrated with synthetic signal. By applying wavelet transformation to deformation data processing, we find that about 4 months before strong earthquakes, several deformation stations near the epicenter received at the same time the abnormal signal with the same frequency and the period from several days to more than ten days. The GPS observation sta- tions near the epicenter all received the abnormal signal whose period is from 3 months to half a year. These ab- normal signals are possibly earthquake precursors.

Methodology for Road Defect Detection and Administration Based on Mobile Mapping Data

A detailed inspection of roads requires highly detailed spatial data with sufficient precision to deliver an accurate geometry and to describe road defects visually.This paper presents a novel method for the detection of road defects.The input data for road defect detection included point clouds and orthomosaics gathered by mobile mapping technology.The defects were categorized in three major groups with the following geometric primitives:points,lines and polygons.The method suggests the detection of point objects from matched point clouds,panoramic images and ortho photos.Defects were mapped as point,line or polygon geometries,directly derived from orthomosaics and panoramic images.Besides the geometric position of road defects,all objects were assigned to a variety of attributes:defect type,surface material,center-of-gravity,area,length,corresponding image of the defect and degree of damage.A spatial dataset comprising defect values with a matching data type was created to perform the attribute analysis quickly and correctly.The final product is a spatial vector data set,consisting of points,lines and polygons,which contains attributes with further information and geometry.This paper demonstrates that mobile mapping suits a large-scale feature extraction of road infrastructure defects.By its simplicity and flexibility,the presented methodology allows it to be easily adapted to extract further feature types with their attributes.This makes the proposed approach a vital tool for data extraction settings with multiple mobile mapping data analysts,e.g.,offline crowdsourcing.

An Improvement of Pedestrian Detection Method with Multiple Resolutions

In object detection, detecting an object with 100 pixels is substantially different from detecting an object with 10 pixels. Many object detection algorithms assume that the pedestrian scale is fixed during detection, such as the DPM detector. However, detectors often give rise to different detection effects under the circumstance of different scales. If a detector is used to perform pedestrian detection in different scales, the accuracy of pedestrian detection could be improved. A multi-resolution DPM pedestrian detection algorithm is proposed in this paper. During the stage of model training, a resolution factor is added to a set of hidden variables of a latent SVM model. Then, in the stage of detection, a standard DPM model is used for the high resolution objects and a rigid template is adopted in case of the low resolution objects. In our experiments, we find that in case of low resolution objects the detection accuracy of a standard DPM model is lower than that of a rigid template. In Caltech, the omission ratio of a multi-resolution DPM detector is 52% with 1 false positive per image (1FPPI);and the omission ratio rises to 59% (1FPPI) as far as a standard DPM detector is concerned. In the large-scale sample set of Caltech, the omission ratios given by the multi-resolution and the standard DPM detectors are 18% (1FPPI) and 26% (1FPPI), respectively.

大直径盾构隧道成型质量巡检方法研究

针对因工业应用成本限制,中、小盾构隧道成型质量无损检测技术迁移至大直径盾构隧道时精度、速度折损严重的问题,以巡检车为载体,集成二维激光扫描仪、编码器和计算机等设备,研制了大盾构隧道成型质量巡检车,并提出一种基于数字图像的盾构质量非对称巡检方法.分析大直径盾构的施工环境,滤除地面、车体点云,并采用邻域向量法提取中轴线,建立隧道中心坐标系.偏心布置巡检路线,按照点云密度将采样点云分为稠密侧和稀疏侧点云,通过不同方法实现对管片接缝特征的拾取:将稠密侧点云绕中轴线展开为二维灰度图像,并通过缩放、归一化、梯度阈值分割等方法实现接缝图像分割;基于直线方程对接缝进行分类,结合管片结构、布置点位,推导出稀疏侧接缝与稠密侧接缝的线性分布公式,间接拾取稀疏侧接缝.根据接缝特征点计算两侧管片边缘点云簇,计算管片错台量;剔除接缝点云簇,使用最小二乘法拟合隧道点云,计算隧道椭圆度.最后在某大直径盾构隧道进行巡检试验,试验结果表明:成型质量巡检车在十四米盾构隧道中巡检速度为3 km·h-1,与传统方法的错台量检测偏差小于2 mm,椭圆度检测偏差小于0.1%,可以满足大直径盾构隧道成型质量巡检的高速度、高精度、低成本需求.