基于梯度双面互补特性的级联快速目标检测

针对目标检测中精度和速度难以兼顾的问题,借助视觉注意理论中的目标感知与识别机制,分析目标描述中梯度幅值与梯度方向信息之间具有的互补性,提出了基于两层级联梯度特征的快速目标检测模型,可有效描述类无关和类相关检测器.一方面,采用梯度幅值特征,从滑动窗口采样中获得候选目标提议,大幅降低了验证窗口的数量,确保检测速度,另一方面,利用级联方式学习训练多个子检测器,可更好实现不同尺度变化下的目标检测精度.PASCAL数据集上的实验结果,解释了级联梯度特征对目标结构描述的有效性,表明了该文方法在与现有先进方法的检测精度相当的前提下,可极大提升检测速度.     

滑窗式航迹自动起始

雷达数据处理的困难在于：地物、海杂波、气象、干扰、噪声产生了大量的虚假点迹，对于航迹的快速自动起始，带来很大困难。本文通过目标的速度特性，采用滑窗式分类判定逻辑，快速自动建航，并根据输出自动调整建航起始逻辑，达到自适应目的。在滑窗后级联多级滤波器，用以消除虚假航迹。     

基于Gabor展开的水下目标激光探测的信号检测与识别

针对水下目标激光探测的信号检测与识别的问题 ,通过对回波信号的Gabor展开进行分析 ,建立了局部窗口多重相关的峰值检测方法 ,以及利用系数矩阵的迹和特征值描述目标的反射性能 ,实现了对目标的分类识别。实验表明 ,上述方法对目标的有效检测率达 93 0 % ,分类识别有效率达 83 2 %。     

基于机器学习的太赫兹光谱分析与识别

国内对转基因作物的监管非常严格,但是对转基因作物的检测缺乏快速准确的计量方法。太赫兹时域光谱结合机器学习分类算法可以实现对转基因作物快速有效地检测识别。通过太赫兹时域光谱技术提取了2种转基因油菜种子和一种非转基因油菜种子的太赫兹吸收谱,朴素贝叶斯算法、基于朴素贝叶斯的自适应提升算法、主成分分析结合随机森林算法、主成分分析结合支持向量计算法被应用于转基因油菜种子的太赫兹吸收谱的分类识别。通过实验对比,基于朴素贝叶斯的自适应提升算法获得了高达96.6%的检测准确率。该研究为运用太赫兹光谱技术手段开展转基因作物的快速检测提供方法参考。     

基于DCResNet的SAR图像车辆目标识别

合成孔径雷达（SAR）图像自动目标识别中,特征提取和目标分类是两个重要环节。残差网络（ResNet）作为一种较新的卷积神经网络,凭借其对目标特征的自适应学习能力,在SAR图像分类领域表现突出。本文在ResNet基础上,设计出了密集连接型残差网络（DCResNet）,用于SAR图像目标识别。DCResNet在残差模块中增加了跳跃性连接的密度,不仅继承了ResNet的易学习的优点,还加强了特征的传播和利用率。除此之外,DCResNet采用平均池化的方式进行下采样,抑制了SAR图像中噪声对识别精度造成的影响。关于SAR图像目标识别的实验结果证明,本文提出的DCResNet与ResNet、AlexNet相比,不仅具有更快的收敛速度和推理速度,而且目标分类的准确率更高。     

一种基于监控视频的出租车识别方法

研究了一种基于视频监控的出租车识别算法.对已经完成跟踪的车辆,通过提取车辆的方向梯度直方图(HOG)特征,作为支持向量机(SVM)分类检测的输入,进行车辆是否为出租车的分类识别.通过多窗口投票机制,增强了分类识别算法的准确性与鲁棒性.实验证明,该方法能准确进行出租车的分类识别,基于实际的标清监控视频,出租车的分类准确率达到90％左右.     

基于颜色均衡分类网络的细胞识别算法

为解决血细胞涂片染色多样性造成卷积神经网络在细胞形态识别时准确率较低的问题,提出了一种新的细胞形态分类算法。通过最优传输算法预训练颜色均衡网络与图像分类网络,保存网络权重,将二者网络结构级联,使用冻结层方法搭载网络参数并进行精细调节。实验结果表明,传统迁移方法的平均识别准确率为92.05%,而该文提出的颜色均衡分类网络U-RNET的平均识别准确率达到95.5%且图像处理速度更快。     

向量机的视频检索应用

提出一种基于支持向量机的视频检索方法,用于复杂背景灰度图像的识别。算法首先用相似度分析方法进行粗筛选,滤去大量非目标窗口,之后用支持向量机对通过的窗口进行分类。由于在通过相似度分析方法所限定的子空间内训练SVM,有效地降低了训练的难度。实验对比数据表明,该方法降低了分类器的训练难度,计算复杂度较低,大大提高了检测速度。     

（英）CCNet:面向多光谱图像的高速船只检测级联卷积神经网络

针对实现遥感图像中船只目标的快速检测,提出了一个采用多光谱图像、基于级联的卷积神经网络(CNN)船只检测方法CCNet。该方法所采用两级级联的CNN依次实现感兴趣区域(ROI)的快速搜索、基于感兴趣区域的船只目标定位和分割。同时,采用含有更多细节信息的多光谱图像作为CCNet的输入,能够提升网络提取特征鲁棒性,从而使得检测更加精确。基于SPOT 6卫星多光谱图像的实验表明：与当前主流的深度学习船只检测方法相比,该方法能够在实现高检测精准度的基础上将检测速度提高5倍以上。     

基于视频图像的含硫矿自燃探测研究

阐述了含硫矿开采过程自燃的危害,传统火灾探测方法的局限;研究了基于视频图像的含硫矿自燃探测方案;提出了一种背景自动更新目标提取法;提出了一种改进的支持向量机分类识别算法;实验数据表明,本研究提高了分类速度,提高了识别准确率。     

基于多尺度重叠滑动池化的SSD果冻杂质检测方法

提出一种基于SSD的杂质检测方法,用于检测生产线中果冻内部的杂质,并标注出杂质的类型和位置。在预处理阶段,提出滑动图像块分割方法,将整张果冻图像分割成若干图像块,避免杂质占比过小,造成准确率低的现象。使用迁移学习的方法,将神经网络在ImageNet数据库上学习到的特征迁移到果冻数据库中,加快网络收敛速度,同时,在一定程度上避免了过拟合现象。提出多尺度重叠滑动池化(SOSP)方法,取代第五层池化以取得更加鲁棒的特征池化。最后,将一幅图下的所有分割块上的检测结果进行整合,得到整张图像的检测结果。实验结果表明,本文提出的方法有效可行,对多种缺陷平均准确率达到0.7271。相比其他方法,本文的算法更具鲁棒性,可应用到果冻生产线中。      

Extracting salient region for pornographic image detection

Content-based pornographic image detection, in which region-of-interest (ROI) plays an important role, is effective to filter pornography. Traditionally, skin-color regions are extracted as ROI. However, skin-color regions are always larger than the subareas containing pornographic parts, and the approach is difficult to differentiate between human skins and other objects with the skin-colors. In this paper, a novel approach of extracting salient region is presented for pornographic image detection. At first, a novel saliency map model is constructed. Then it is integrated with a skin-color model and a face detection model to capture ROI in pornographic images. Next, a ROI-based codebook algorithm is proposed to enhance the representative power of visual-words. Taking into account both the speed and the accuracy, we fuse speed up robust features (SURF) with color moments (CM). Experimental results show that the precision of our ROI extraction method averagely achieves 91.33%, more precisely than that of using the skin-color model alone. Besides, the comparison with the state-of-the-art methods of pornographic image detection shows that our approach is able to remarkably improve the performance.     

AFLNet: Adversarial focal loss network for RGB-D salient object detection

Because salient objects usually have fewer data in a scene, the problem of class imbalance is often encountered in salient object detection (SOD). In order to address this issue and achieve the consistent salient objects, we propose an adversarial focal loss network with improving generative adversarial networks for RGB-D SOD (called AFLNet), in which color and depth branches constitute the generator to achieve the saliency map, and adversarial branch with high-order potentials, instead of pixel-wise loss function, refines the output of the generator to obtain contextual information of objects. We infer the adversarial focal loss function to solve the problem of foreground–background class imbalance. To sufficiently fuse the high-level features of color and depth cues, an inception model is adopted in deep layers. We conduct a large number of experiments using our proposed model and its variants, and compare them with state-of-the-art methods. Quantitative and qualitative experimental results exhibit that our proposed approach can improve the accuracy of salient object detection and achieve the consistent objects.     

Layered moving-object segmentation for stereoscopic video using motion and depth information

A novel layered stereoscopic moving-object segmentation method is proposed in this paper by exploiting both motion information and depth information to extract moving objects for each depth layer with high accuracy on their shape boundary. By taking a higher-order statistics on two frame-difference fields across three adjacent frames, the computed motion information are used to conduct change detection and generate one motion mask that consists of all the moving objects from all the depth layers involved at each view. It would be highly desirable, and challenging, to further differentiate them according to their residing depth layer to achieve layered segmentation. For that, multiple depth-layer masks are generated using our proposed disparity estimation method, one for each depth layer. By intersecting the motion mask and one depth-layer mask at any given layer-of-interest, the moving objects associated with the corresponding layer are then extracted. All the above-mentioned processes are repeatedly performed along the video sequence with a sliding window of three frames at a time. For demonstration, only the foreground and the background layers are considered in this paper, while the proposed method is generic and can be straightforwardly extended to more layers, once the corresponding depth-layer masks are made available. Experimental results have shown that the proposed layered moving-object segmentation method is able to segment the foreground and background moving objects separately, with high accuracy on their shape boundary. In addition, the required computational load is considered fairly inexpensive, since our design methodology is to generate masks and perform intersections for extracting the moving objects for each depth layer.     

WeBox:locating small objects from weak edges

In the object detection task, how to better deal with small objects is a great challenge. The detection accuracy of small objects greatly affects the final detection performance. Our propose a detection framework WeBox based on weak edges for small object detection in dense scenes, and proposes to train the richer convolutional features (RCF) edges detection network in a weakly supervised way to generate multi-instance proposals. Then through the region proposal network (RPN) network to locate each object in the multi-instance proposals, in order to ensure the effectiveness of the multi-instance proposals, we correspondingly proposed a multi-instance proposals evaluation criterion. Finally, we use faster region-based convolutional neural network (R-CNN) to process WeBox single-instance proposals and fine-tune the final results at the pixel level. The experiments have been carried out on BDCI and TT100K proves that our method maintains high computational efficiency while effectively improving the accuracy of small objects detection.     

一种联合文本和图像信息的行人检测方法

针对纯视觉行人检测方法存在的误检、漏检率高,遮挡目标以及小尺度目标检测精度低等问题,提出一种联合文本和图像信息的行人检测方法.该方法首先利用图像分析的方法初步获取图像目标的候选框,其次通过文本分析的方法获取文本中有关图像目标的实体表达,并提出一种基于马尔科夫随机场的模型用于推断图像候选框与文本实体表达之间的共指关系（Coreference Relation）,以此达到联合图像和文本信息以辅助机器视觉提高交通场景下行人检测精度的目的.在增加了图像文本描述的加州理工大学行人检测数据集上进行的测评结果表明,该方法不仅可以在图像信息的基础上联合文本信息提高交通场景中的行人检测精度,也能在文本信息的基础上联合图像信息提高文本中的指代消解（Anaphora Resolution）精度.     

一种基于深度学习的公路巡检无人机实时目标跟踪算法

公路中心标线的实时跟踪是公路巡检无人机视觉飞行中关键的一环.针对目前主流目标跟踪算法实时性差的问题,提出一种基于改进YOLO(you only look once)v3和Deep-SORT(deep simple online real-time tracking)的目标跟踪模型用于公路巡检无人机自主视觉飞行.通过引入并改进跨阶段局部网络,优化网络层级结构,使用泛化能力更好的激活函数,提升了公路道路标线的检测准确率和无人机平台的检测速度.对检测到的公路标线信息使用Deep-SORT算法进行公路中心标线跟踪.实验结果表明,与几类典型目标跟踪模型相比,在跟踪准确度基本不变的情况下,处理速度提升了数倍.     

基于激光雷达的多帧联合目标跟踪系统

本文讨论了利用激光雷达与检测前跟踪算法进行室内行人跟踪,实现监测室内人群社交距离的探测系统。激光雷达距离分辨率高,定位误差小,目标探测时间间隔小,很适用于对轨迹不确定的行人进行跟踪。本文所用检测前跟踪算法分为三步,首先利用历史数据所建杂波图滤除固定物杂波。其次利用多帧检测算法得到时间窗内的目标轨迹段,以此滤除随机噪声产生的虚警点,提升检测效率。第三步利用轨迹关联算法,将各个时间窗内的轨迹片段相互关联,得到完整的行人轨迹,从而增强机动目标与相互临近目标的跟踪效果。实测实验表明在激光雷达跟踪多个目标时,其跟踪精度在10~15cm,多次实验均未出现目标丢失或虚假轨迹,较好地完成了室内行人的检测与跟踪。     

Hierarchical Implicit Shape Modeling

In this paper, a new hierarchical approach for object detection is proposed. Object detection methods based on Implicit Shape Model (ISM) efficiently handle deformable objects, occlusions and clutters. The structure of each object in ISM is defined by a spring like graph. We introduce hierarchical ISM in which structure of each object is defined by a hierarchical star graph. Hierarchical ISM has two layers. In the first layer, a set of local ISMs are used to model object parts. In the second layer, structure of parts with respect to the object center is modeled by global ISM. In the proposed approach, the obtained parts for each object category have high discriminative ability. Therefore, our approach does not require a verification stage. We applied the proposed approach to some datasets and compared the performance of our algorithm to comparable methods. The results show that our method has a superior performance.     

传感器网络中基于分层架构的音频场景分析方法

A hierarchical method for scene analysis in audio sensor networks is proposed. This method consists of two stages: element detection stage and audio scene analysis stage. In the former stage, the basic audio elements are modeled by the HMM models and trained by enough samples off-line, and we adaptively add or remove basic element from the targeted element pool according to the time, place and other environment parameters. In the latter stage, a data fusion algorithm is used to combine the sensory information of the same area, and then, a rule-based method is employed to analyze the audio scene based on the fused data. We conduct some experiments to evaluate the performance of the proposed method that about 70% audio scenes can be detected correctly by this method. The experiment evaluations demonstrate that our method can achieve satisfactory results.