Prior-guided GAN-based interactive airplane engine damage image augmentation method

Deep learning-based methods have achieved remarkable success in object detection, but this success requires the availability of a large number of training images. Collecting sufficient training images is difficult in detecting damages of airplane engines. Directly augmenting images by rotation, flipping, and random cropping cannot further improve the generalization ability of existing deep models. We propose an interactive augmentation method for airplane engine damage images using a prior-guide...     

SRS-Net: Training object detectors from scratch for remote sensing images without pretraining

Most of the current object detection algorithms use pretrained models that are trained on ImageNet and then fine-tuned in the network, which can achieve good performance in terms of general object detectors. However, in the field of remote sensing image object detection, as pretrained models are significantly different from remote sensing data, it is meaningful to explore a train-from-scratch technique for remote sensing images. This paper proposes an object detection framework trained from scratch, SRS-Net, and describes the design of a densely connected backbone network to provide integrated hidden layer supervision for the convolution module. Then, two necessary improvement principles are proposed: studying the role of normalization in the network structure, and improving data augmentation methods for remote sensing images. To evaluate the proposed framework, we performed many ablation experiments on the DIOR, DOTA, and AS datasets. The results show that whether using the improved backbone network, the normalization method or training data enhancement strategy, the performance of the object detection network trained from scratch increased. These principles compensate for the lack of pretrained models. Furthermore, we found that SRS-Net could achieve similar to or slightly better performance than baseline methods, and surpassed most advanced general detectors.     

Bayesian edge detector for SAR imagery using discontinuity-adaptive Markov random feld modeling

Synthetic aperture radar（SAR）image is severely affected by multiplicative speckle noise,which greatly complicates the edge detection.In this paper,by incorporating the discontinuityadaptive Markov random feld（DAMRF）and maximum a posteriori（MAP）estimation criterion into edge detection,a Bayesian edge detector for SAR imagery is accordingly developed.In the proposed detector,the DAMRF is used as the a priori distribution of the local mean reflectivity,and a maximum a posteriori estimation of it is thus obtained by maximizing the posteriori energy using gradient-descent method.Four normalized ratios constructed in different directions are computed,based on which two edge strength maps（ESMs）are formed.The fnal edge detection result is achieved by fusing the results of two thresholded ESMs.The experimental results with synthetic and real SAR images show that the proposed detector could effciently detect edges in SAR images,and achieve better performance than two popular detectors in terms of Pratt＇s fgure of merit and visual evaluation in most cases.     

Neural network directed Bayes decision rule for moving targetclassification

In this paper, a new neural network directed Bayes decision rule is developed for target classification exploiting the dynamic behavior of the target. The system consists of a feature extractor, a neural network directed conditional probability generator and a novel sequential Bayes classifier. The velocity and curvature sequences extracted from each track are used as the primary features. Similar to hidden Markov model scheme, several hidden states are used to train the neural network, the output of which is the conditional probability of occurring the hidden states given the observations. These conditional probabilities are then used as the inputs to the sequential Bayes classifier to make the classification. The classification results are updated recursively whenever a new scan of data is received. Simulation results on multiscan images containing heavy clutter are presented to demonstrate the effectiveness of the proposed methods     

Target localization method of non-cooperative spacecraft on on-orbit service

With the explosion of the number of meteoroid/orbital debris in terrestrial space in recent years, the detection environment of spacecraft becomes more complex. This phenomenon causes most current detection methods based on machine learning intractable to break through the two difficulties of solving scale transformation problem of the targets in image and accelerating detection rate of high-resolution images. To overcome the two challenges, we propose a novel non-cooperative target detection method using the framework of deep convolutional neural network.Firstly, a specific spacecraft simulation dataset using over one thousand images to train and test our detection model is built. The deep separable convolution structure is applied and combined with the residual network module to improve the network’s backbone. To count the different shapes of the spacecrafts in the dataset, a particular prior-box generation method based on K-means cluster algorithm is designed for each detection head with different scales. Finally, a comprehensive loss function is presented considering category confidence, box parameters, as well as box confidence. The experimental results verify that the proposed method has strong robustness against varying degrees of luminance change, and can suppress the interference caused by Gaussian noise and background complexity. The mean accuracy precision of our proposed method reaches 93.28%, and the global loss value is 13.252. The comparative experiment results show that under the same epoch and batchsize, the speed of our method is compressed by about 20% in comparison of YOLOv3, the detection accuracy is increased by about 12%, and the size of the model is reduced by nearly 50%.     

人工智能技术在航空发动机孔探检测中的应用进展

孔探是当前航空发动机检修过程中应用最多的无损检测方法,也是孔探图像的唯一获取途径。近年来,深度学习等人工智能方法开始被应用到航空发动机损伤分类、检测中,为实现航空发动机检修智能化提出了一些现行有效的方法,具有重要的工业应用价值。本文概述了航空发动机孔探检测的发展和优缺点,综述了专家系统和机器学习人工智能方法在发动机孔探图像方面的应用进展,总结了基于孔探图像实现航空发动机孔探检测智能化面临的一些挑战。     

基于机器视觉技术的飞机绕机外观检查方法研究

大型民用飞机试飞和航线运营期间,对其外观表面进行绕机外观检查是适航性安全检查的必要工作.目前飞机的绕机检查主要采用人工绕机方式,该方式,且成本高、效率低,易出现漏检、误检等人为因素,因此智能外观表面检查方法的研究是一项迫切的任务.相比于其他工业检测任务,飞机外观检查智能识别目前无公开数据集,且飞机真实外观损伤类型多样....     

结合核相关滤波器和深度学习的运动相机中无人机目标检测

针对无人机与相机快速相对运动造成的运动模糊问题，以及小型无人机外观信息缺失和背景复杂造成漏警和虚警问题，提出了一种新的无人机检测-跟踪方法。针对成像尺寸小于32像素×32像素的无人机目标，提出改进的多层特征金字塔的分类和目标框回归器作为目标检测器，克服漏警。利用检测结果初始化基于核相关滤波的目标跟踪器，并持续修正跟踪结果，跟踪结果为剔除检测器虚警提供依据。在跟踪过程中，引入对观测场景纹理自适应的相机运动补偿策略实现目标重定位。多场景下的实验结果表明：提出的方法在对高速运动小目标的检测和跟踪指标上显著优于传统方法，且运动补偿机制的引入进一步增强了方法在极端复杂场景下的鲁棒性。     

Radar detection and preclassification based on multiple hypothesis

This work presents a single-scan-processing approach to the problem of detecting and preclassifying a radar target that may belong to different target classes. The proposed method is based on a hybrid of the maximum a posteriori (MAP) and Neyman-Pearson (NP) criteria and guarantees the desired constant false alarm rate (CFAR) behavior. The targets are modeled as subspace random signals having zero mean and given covariance matrix. Different target classes are discriminated based on their different signal subspaces, which are specified by their corresponding projection matrices. Performance is investigated by means of numerical analysis and Monte Carlo simulation in terms of probability of false alarm, detection and classification; the extra signal-to-noise power ratio (SNR) necessary to classify once target detection has occurred is also derived.     

卷积神经网络和峭度在轴承故障诊断中的应用

针对传统智能诊断方法依靠专家知识和人工提取数据特征工作量大的问题,结合深度学习方法在特征提取和处理大数据方面的优势,研究了一种基于卷积神经网络和振动信号峭度指标的滚动轴承故障诊断方法。该方法将深度学习应用于轴承故障诊断,提取滚动轴承正常状态、内圈故障、外圈故障和滚动体故障4种状态的振动信号,将振动信号分段处理得到峭度指标,使用数据到图像的转换方法将峭度指标转换为灰度图,送入卷积神经网络模型完成故障分类。在进行滚动轴承故障诊断的实验时,所提的模型诊断准确率达到99.5%,高于传统支持向量机(SVM)算法的95.8%。      

Novel approach of crater detection by crater candidate region selection and matrix-pattern-oriented least squares support vector machine

Impacted craters are commonly found on the surface of planets, satellites, asteroids and other solar system bodies. In order to speed up the rate of constructing the database of craters, it is important to develop crater detection algorithms. This paper presents a novel approach to automatically detect craters on planetary surfaces. The approach contains two parts: crater candidate region selection and crater detection. In the first part, crater candidate region selection is achieved by Kanade-Lucas-Tomasi (KLT) detector. Matrix-pattern-oriented least squares support vector machine (MatLSSVM), as the matrixization version of least square support vector machine (SVM), inherits the advantages of least squares support vector machine (LSSVM), reduces storage space greatly and reserves spatial redundancies within each image matrix compared with general LSSVM. The second part of the approach employs MatLSSVM to design classifier for crater detection. Experimental results on the dataset which comprises 160 preprocessed image patches from Google Mars demonstrate that the accuracy rate of crater detection can be up to 88%. In addition, the outstanding feature of the approach introduced in this paper is that it takes resized crater candidate region as input pattern directly to finish crater detection. The results of the last experiment demonstrate that MatLSSVM-based classifier can detect crater regions effectively on the basis of KLT-based crater candidate region selection.     

Performance analysis of linearly combined order statistic CFARdetectors

Linearly combined order statistic (LCOS) constant false-alarm rate (CFAR) detectors are examined for efficient and robust threshold estimation applied to exponentially distributed background observations for improved detection. Two optimization philosophies have been employed to determine the weighting coefficients of the order statistics. The first method optimizes the coefficients to obtain efficient estimates of clutter referred to the censored maximum likelihood (CML) and best linear unbiased (BLU) CFAR detectors. The second optimization involves maximizing the probability of detection under Swerling II targets and is referred to as the most powerful linear (MPL) CFAR detector. The BLU-CFAR detector assumes no knowledge of the target distribution in contrast to the MPL-CFAR detector which requires partial knowledge of the target distribution. The design of these CFAR detectors and the probability of detection performance are mathematically analyzed for background observations having homogeneous and heterogeneous distributions wherein the trade-offs between robustness and detection performance are illustrated     

A new and robust CFAR detection algorithm

A constant false alarm rate (CFAR) detection method which is based on a combination of median and morphological filters (MEMO) is proposed. The MEMO algorithm has robust performance with small CFAR loss, very good behavior at clutter edges and high detection performance in the case of closely spaced narrowband signals (targets). The proposed MEMO method is favourably compared with cell averaging (CA) and ordered statistics (OS) CFAR detectors. The Monte Carlo method is employed to analyze the MEMO-CFAR detector     

基于遥感图像分块直线特征检测的机场跑道检测方法

针对遥感图像机场跑道检测问题,提出了一种基于图像分块直线特征检测的机场跑道检测方法。首先,针对遥感图像数据量大带来的计算处理问题,设计了基于直线分割检测子（LSD）的遥感图像分块直线特征检测环节;然后,在总结归纳机场跑道数学特性的基础上,对提取的直线特征进行平行线分组、直线生长、平行线合并,并以Radon变换为基础,找出候选机场跑道区域;最后,使用灰度统计信息并结合梯度方向直方图对候选区域进行处理,筛选出最终的机场道路区域。实验结果表明,在能够提取出有效直线特征的情况下,该方法可以对多类机场跑道进行有效定位。     

视频序列中的人脸检测算法

根据视频序列的特性,首先采用背景差分法提取视频中的运动区域,之后对检测的运动区域采用对数扩展的方式调整光照的影响。采用基于YCbCr颜色空间的肤色模型进行检测得到人脸候选区域,并采用一定的阈值去除小的非人脸的区域,再通过在人脸候选区域检测眼睛位置,利用人脸几何特征精确定位人脸。实验表明,该方法提高了检测速度,降低了误检率,可以应用在视频监控等实时系统中。     

基于深度学习的发动机叶片故障检测技术

为了解决航空发动机叶片故障检测中存在的检测精度欠佳、检测效率不高的问题,提出了一种基于深度学习的目标检测方法。针对小样本数据集检测精度低、模型训练速度慢等问题,对Faster R-CNN目标检测算法进行结构优化,引入Res2Net结构,通过分割串联的策略强化残差模块的卷积学习能力,搭建了细粒级的多尺度残差模型Res2Net-50,以提升模型的特征提取能力。同时,在网络的训练过程中,采用多次余弦退火衰减法对学习率进行调整,以加快模型的训练速度,提升模型的训练质量。针对航空发动机叶片裂纹和缺损2种故障类型进行网络训练与检测试验,试验结果表明：优化后的模型识别准确率提高了0.7%,模型的平均检测精度提高了1.8%,训练时间缩短了5.56%,取得了比较好的检测效果。     

Mutual and conditional mutual informations for optimizingdistributed Bayes detectors

This paper considers optimization of distributed detectors under the Bayes criterion. A distributed detector consists of multiple local detectors and a fusion center that combines the local decision results to obtain a final decision. Introduced first are distributional distance measures, the mutual information (MI) and the conditional mutual information (CMI), that are obtained by applying information theoretic concepts to detection problems. Error bound analyses show that these distance measures approximate the Bayesian probability of error better than the conventional ones regardless of the operational environments. Then, a new optimization technique is proposed for distributed Bayes detectors. The method uses the distributional distances instead of the original Bayes criterion to avoid the complexity barrier of the optimization problem. Numerical examples show that the proposed distance measures yield solutions far better than the existing ones