基于神经免疫生长可免域网络的红外光谱图像分割算法

管道运输对远距离输送石油天然气有着较大优势,而与之伴随的管道安全问题使得管道安全检测至关重要。为确保任何时间下管道状况的有效检测,红外成像技术由于其根据对象的热辐射信息反映目标特征的特殊性,能够忽视可见光的影响检测管道状态,因而在管道检测领域有重要意义。但由于户外环境的多样性,交错的管道和复杂环境使得采集的红外管道图像具有目标特征分布不均匀,目标遮挡和背景类目标干扰等问题。这些问题增加了提取管道目标的难度,不利于管道的分割和检测。生物免疫系统在抗原检测、提取和消除上表现出识别、学习、记忆、耐受和协调配合等目前复杂系统优化策略所缺乏的优异特性,借鉴生物神经系统调控免疫系统的机理,设计一种基于神经免疫网络的复杂背景下红外管道目标的检测与提取算法。根据生物神经网络在免疫系统中的调控机制,利用基础管道形状特征模型构建用于红外管道目标定位的神经网络,并将最优神经免疫可免域和区域种子生长结合,解决管道遮挡影响提取目标完整性的问题。选择三种典型的红外管道图像,将传统目标检测算法与基于神经免疫网络的算法进行了效果对比分析。结果表明,传统算法的平均真阳性率为40.56%,Jaccard相似性指数为27.18%,绝对误差率为11.75%,而基于神经免疫网络算法的真阳性率为98.05%,Jaccard相似性指数为94.44%,绝对误差率为1.18%。对比可知,神经免疫网络算法的真阳性率比传统方法高57.49%,绝对误差率则低10.57%,验证了复杂背景下,本文算法相比传统方法能够更加准确地提取完整的红外管道目标,这对管道安全检测效率的提高有着重要意义。     

基于最小平均距离免疫算法的模糊红外图像分割

由于犯罪分子利用各种方法来避开传统的刑侦图像技术,因而红外图像逐渐成为获取犯罪现场痕迹的有效手段。然而,从犯罪现场拍摄的红外图像其目标痕迹大多是弱化的,所以在这类红外图像中分割目标是一项具有挑战性的任务。已有基于生物免疫的各类算法尚未明确描述免疫分割作用领域,以及免疫网络算法模型中的免疫识别距离。为实现对目标痕迹弱化红外图像的有效分割,提出了一种新的具有免疫作用领域和最小平均免疫识别距离的人工免疫构架,设计了一种具备最小平均距离免疫域的免疫分割算法。该方法根据红外图像的特点,采用多步分类算法、免疫变异和自适应免疫最小均距识别方法,根据目标区域和背景区域的总体统计特性实现最佳分类。实验结果表明,提出的基于最小平均距离的免疫算法能够有效地分割目标弱化的红外图像。与经典的边缘模板和区域模板方法相比,该算法具有更好的分割效果,尤其是针对目标弱化红外图像的分割,该算法能够较好地给出五个手指的边界轮廓。     

Image enhancement algorithm based on improved lateral inhibition network

There is often substantial noise and blurred details in the images captured by cameras. To solve this problem, we propose a novel image enhancement algorithm combined with an improved lateral inhibition network. Firstly, we built a mathematical model of a lateral inhibition network in conjunction with biological visual perception; this model helped to realize enhanced contrast and improved edge definition in images. Secondly, we proposed that the adaptive lateral inhibition coefficient adhere to an exponential distribution thus making the model more flexible and more universal. Finally, we added median filtering and a compensation measure factor to build the framework with high pass filtering functionality thus eliminating image noise and improving edge contrast, addressing problems with blurred image edges. Our experimental results show that our algorithm is able to eliminate noise and the blurring phenomena, and enhance the details of visible and infrared images.     

An ensemble template algorithm for extracting targets from blurred infrared images

In this paper, an ensemble template algorithm is proposed to extract targets from blurred infrared images. First, the image pixels are divided according to their gray values into three pixel sets, a target set, a background set and the third set without class label. Second, the neighborhood statistical characteristics for each pixel are calculated as its template features. Third, ensemble detectors are designed using target pixels and background pixels based on their template features, and these ensemble detectors are used to detect the third pixel set. To evaluate the performance of the proposed extraction algorithm, this paper compares the ensemble template with other extraction algorithms using blurred infrared image of hand trace. Experimental results show that the ensemble template algorithm proposed in this paper exhibits better extraction performance.     

多局部残差连接注意网络的图像去模糊

针对现有的基于卷积神经网络的图像去模糊算法存在图像纹理细节恢复不清晰的问题,提出了一种基于多局部残差连接注意网络的图像去模糊算法。首先,采用一个卷积层进行浅层特征提取;其次,设计了一种新的基于残差连接和并行注意机制的多局部残差连接注意模块,用于消除图像模糊并提取上下文信息;再次,采用一个基于扩张卷积的成对连接模块进行细节恢复;最后,利用一个卷积层重建清晰图像。实验结果表明：在GoPro数据集上的PSNR (peak signal to noise ratio)和SSIM (structure similarity)分别为31.83 dB、0.927 5,在定性和定量两方面都表明所提方法能够有效地恢复模糊图像的纹理细节,网络性能优于对比方法。     

An Adaptive Deblurring Vehicle Detection Method for High-Speed Moving Drones: Resistance to Shake

Vehicle detection is an essential part of an intelligent traffic system, which is an important research field in drone application. Because unmanned aerial vehicles (UAVs) are rarely configured with stable camera platforms, aerial images are easily blurred. There is a challenge for detectors to accurately locate vehicles in blurred images in the target detection process. To improve the detection performance of blurred images, an end-to-end adaptive vehicle detection algorithm (DCNet) for drones is proposed in this article. First, the clarity evaluation module is used to determine adaptively whether the input image is a blurred image using improved information entropy. An improved GAN called Drone-GAN is proposed to enhance the vehicle features of blurred images. Extensive experiments were performed, the results of which show that the proposed method can detect both blurred and clear images well in poor environments (complex illumination and occlusion). The detector proposed achieves larger gains compared with SOTA detectors. The proposed method can enhance the vehicle feature details in blurred images effectively and improve the detection accuracy of blurred aerial images, which shows good performance with regard to resistance to shake.     

基于PDE的前视红外成像舰船目标检测

针对复杂环境下的海面目标提取问题,提出一种基于偏微分方程（PDE）的红外舰船检测算法。首先采用基于PDE理论的滤波模型对初始图像滤噪并进行背景估计,然后结合邻域差分进行背景抑制后分割提取目标。实验表明,该算法能够有效检测强杂波背景中的目标,方法适应性强。     

Target extraction from blurred trace infrared images with a superstring galaxy template algorithm

Accurate and efficient targets extraction from blurred trace infrared images has very important meaning for latent trace evidence collection in crime scene. Based on the superstring theory, a superstring galaxy template extraction algorithm for infrared trace target is presented. First, all of the pixels are divided into three classes: target pixels, background pixels and blurred pixels. Next, the superstring template characteristics for every pixel in a blurred infrared image are calculated as the features of each pixel. Finally, a galaxy covering algorithm is proposed, target pixels and background pixels are used for training the galaxy covering domain of every galaxy classifiers, and these classifiers will divide each blurred pixel into two classes: a target pixel or a background pixel. Experimental results indicate that the superstring galaxy template algorithm can improve the target extraction rate and reduce the extraction error rate.     

基于红外图像的GVF Snake轮廓提取算法的研究

针对红外图像目标具有边界模糊不清,区分效果较差的缺点,结合Ostu阈值法和梯度矢量流主动轮廓模型（GVF Snake）,提出一种目标轮廓自动提取方法。采用Ostu法先对图像进行分割,然后将得到的边界作为Snake模型的初始边缘轮廓,利用 GVF Snake特性将初始轮廓准确地收敛到目标边界。由于Ostu算法具有将目标物体从复杂背景中分割开来的优点,使得在应用 GVF Snake模型对复杂图像进行分割时减少了人工的干预。实验证明：该方法运算速度快,能够快速地收敛到目标轮廓,并准确地跟踪目标,具有一定的抗噪能力。     

Variational infrared image enhancement based on adaptive dual-threshold gradient field equalization

Infrared images are characterized by low signal to noise ratio (SNR) and fuzzy texture edges. This article introduces the variational infrared image enhancement algorithm based on gradient field equalization with adaptive dual thresholds. Firstly, we transform the image into gradient domain and get the gradient histogram. Then, we do the gradient histogram equalization. By setting adaptive dual thresholds to qualify the gradients, the image is prevented from over enhancement. The total variation (TV) model is adopted in the reconstruction of the enhanced image to suppress noise. It is shown from experimental results that the image edge details are significantly enhanced, and therefore the algorithm is qualified for enhancement of infrared images in different applications.     

极值温差红外图像的自适应双局部增强算法

针对目前红外焦平面成像系统在观察目标、特别是极值温差目标时,各温度段灰度描述不均匀和细节不够的问题,提出了一种自适应红外图像双局部增强算法。详细介绍了通过空间分布和灰度统计特性两个方向实现对极值温差图像自适应增强的方法,该方法首先从红外图像的空间分布特性出发,将图像切割成多个局部图像,然后再从直方图灰度分布出发,将局部图像的直方图进行聚类分段,并对分段直方图均衡增强,最后对生成的每个局部图像增强结果进行线性插值拼接完成增强算法。通过在红外焦平面系统中实验证明了极值温差自适应的红外图像双局部增强算法的可行性,并获得了很好的效果,成像质量有明显提高。     

基于热红外图像处理技术的农作物冠层识别方法研究

为解决农作物冠层热红外图像边缘灰度级分布不均且噪声较大,而传统图像分割方法难以实现其目标区域有效识别的难题,以苗期红小豆冠层热红外图像为研究对象,将模糊神经网络和仿射变换有机结合,提出了基于热红外图像处理技术的农作物冠层识别模型。首先利用五层线性归一化模糊神经网络的自适应特性,选取高斯隶属度函数,自动计算冠层可见光图像识别的推理规则,有效地分割了可见光图像中的冠层区域。通过分析3种分割指标和熵,定量评价可见光图像冠层分割质量。网络迭代38次时,误差精度为0.000 952,该算法平均有效识别率为96.13%,获取可见光冠层图像的像元信息熵值范围为2.454 4~5.198 7,与标准算法所得冠层图像的像元信息熵仅相差0.245 9。然后以取得可见光图像的冠层有效区域为参考图像,采用仿射变换算法,调整优选平移、旋转、缩放等图像变换因子,配准原始热红外图像,提出了基于仿射变换的冠层热红外图像识别方法。对于初始温度范围值在16.35~19.92 ℃的农作物热红外图像,计算选取旋转幅度为1.0和缩放因子为0.9时,作为异源图像的最优配准参数,获取目标图像的最大温差为3.17 ℃,相对于原图像的平均温度值由18.711 ℃下降至17.790 ℃,进而实现了基于热红外图像处理技术的农作物冠层识别。最后以熵的互信息作为监督指标,对农作物冠层热红外图像识别方法进行评价。提出的冠层热红外图像识别方法,所获取的目标图像与初始热红外图像的平均互信息为4.368 7,标准目标图像和初始热红外图像的平均互信息为3.981 8,二者仅相差0.486 9。同时,两种冠层热红外图像的平均温度差值为0.25 ℃,高效消除了原始热红外图像的背景噪声。结果表明本研究方法的有效性和实用性,能够为应用热红外图像反映农作物生理生态信息特征指标参数提供技术借鉴。     

纹理分析在红外目标识别中的应用

若背景有雾时，红外图像中的目标会受到影响而变得比较模糊。雾的纹理是一种比较典型的自然纹理，利用灰度共生矩阵提取纹理特征图像的方法，对雾天获取的红外图像进行分析。通过实验证明了该方法对红外图像中（背景有雾时）目标的提取和识别有积极的作用。     

层次卷积滤波红外弱小目标检测方法

针对单帧复杂背景红外图像点目标检测算法存在复杂背景下处理效果不理想、处理时间长的问题,提出了一种层次卷积滤波检测算法。主要分为两个部分：第一,根据红外小目标特性,设计一种层次卷积滤波的算子,对图像进行滤波处理,实现图像中小目标的增效和背景抑制的效果;第二,采用基于最大值的自适应阈值方法,对图像进行二值化操作,过滤背景杂波,最终提取到待检测的目标。在大量不同背景红外图像中进行实验,论文算法在背景抑制因子和信噪比增益的性能量化结果上优于现有5种典型红外弱小目标检测算法的性能结果,且平均处理时间仅为高斯拉普拉斯（Laplacian of Gaussian,LoG）滤波算法的30.42%。通过实验对比,表明该层次卷积滤波算法可以有效解决在不同复杂背景下的红外图像中对小目标检测的问题。     

几种直方图变换法在输电接头红外热像增强中的对比研究

由红外热像仪直接得到的热像会受很多外来因素的干扰而致使图像模糊不清、对比度差,由此在故障诊断过程中,需要对热像进行后续增强处理。总结了影响热故障红外检测准确性的外部因素,分析了输电接头热像的基本特征,采用了基于平方根灰度变换直方图修正和全局均衡化两种方法对热像进行了增强处理。在分析上述两种方法优劣的基础上,提出了一种基于线性平滑滤波的局部均衡化方法。结果表明,基于线性平滑滤波的局部均衡化方法对于处理输电接头热像更有效。     

一种基于GAN和自适应迁移学习的样本生成方法

研究了基于生成式对抗网络(GAN)和跨域自适应迁移学习的样本生成和自动标注方法。该方法利用自适应迁移学习网络,基于已有的少量可见光图像样本集,挖掘目标在红外和可见光图像中特征内在相关性,构建自适应的转换迁移学习网络模型,生成标注好的目标图像。提出的方法解决了红外图像样本数量少且标注费时的问题,为后续多频段协同目标检测和识别获得了足够的样本数据。实验结果表明:自动标注算法对实际采集的装甲目标图像和生成的装甲目标图像各1 000张进行自动标注测试,对实际装甲目标图像的标注准确率达到95%以上,对生成的装甲目标标注准确率达到83%以上;利用真实图像和生成图像的混合数据集训练的分类器的性能和使用纯真实图像时基本一致。     

结合小波融合和深度学习的脑胶质瘤自动分割

针对水肿区域边界模糊和瘤内结构复杂多变导致的脑胶质瘤分割不精确问题,本文提出了一种基于小波融合和3D-UNet网络的脑胶质瘤磁共振图像自动分割算法.首先,对脑胶质瘤磁共振图像的T1、T1ce、T2、Flair四种模态进行小波融合以及偏置场校正;然后,提取待分类的图像块;再利用提取的图像块训练3D-UNet网络以对图像块中的像素进行分类;最后加载损失率较小的网络模型进行分割,并采用基于连通区域的轮廓提取方法,以降低假阳性率.对57组Brats2018（Brain Tumor Segmentation 2018）磁共振图像测试集进行分割的结果显示,肿瘤的整体、核心和水肿部分的平均分割准确率（DSC）分别达到90.64%、80.74%和86.37%,这表明该算法分割脑胶质瘤准确率较高,与金标准相近.相比多模态图像融合前,该算法在减少输入网络数据量和图像冗余信息的同时,还一定程度上解决了胶质瘤边界模糊、分割不精确的问题,提高了分割的准确度和鲁棒性.     

红外运动小目标的检测

通过分析天空背景下红外运动小目标、噪音以及背景的特征,提出一种检测方法·首先利用向量小波变换对运动图像进行预处理;其次采用图像差分进行目标的粗检测,提取出候选目标;最后可根据运动目标和噪音的特征对候选目标进行识别,检测出真实的运动小目标·实验证明,该方法可有效检测天空背景下红外运动小目标·     

循环侧抑制网络的红外图像预处理FPGA实现研究

在红外成像预处理中,循环侧抑制网络具有很好的增强图像边缘反差、突出边框的功能。由于循环侧抑制网络处理运算的复杂性,如何实时实现循环侧抑制网络的红外图像预处理成为关键问题。根据简化的循环侧抑制网络处理算法,并结合硬件计算特点,提出一种适合于硬件实现循环侧抑制网络的并行处理结构,采用流水线设计方式在FPGA中实时实现。试验证明：该算法在FPGA内具有与视频流同步的实时性,达到了增强反差、突出边框的效果。用于图像目标实时跟踪系统时,极大地提高了目标的跟踪精度和稳定性。     

Blind motion deblurring using multiple images

Recovery of degraded images due to motion blurring is a challenging problem in digital imaging. Most existing techniques on blind deblurring are not capable of removing complex motion blurring from the blurred images of complex structures. One promising approach is to recover the clear image using multiple images captured for the scene. However, in practice it is observed that such a multi-frame approach can recover a high-quality clear image of the scene only after multiple blurred image frames are accurately aligned during pre-processing, which is a very challenging task even with user interactions. In this paper, by exploring the sparsity of the motion blur kernel and the clear image under certain domains, we propose an alternative iteration approach to simultaneously identify the blur kernels of given blurred images and restore a clear image. Our proposed approach is not only robust to image formation noises, but is also robust to the alignment errors among multiple images. A modified version of linearized Bregman iteration is then developed to efficiently solve the resulting minimization problem. Experiments show that our proposed algorithm is capable of accurately estimating the blur kernels of complex camera motions with minimal requirements on the accuracy of image alignment. As a result, our method is capable of automatically recovering a high-quality clear image from multiple blurred images.