多尺度超像素分割和奇异谱分析的高光谱影像分类

目的 高光谱影像（hyperspectral image,HSI）中“同物异谱,异物同谱”的现象普遍存在,使分类结果存在严重的椒盐噪声问题。HSI中的空间地物结构复杂多样,单一尺度的空间特征提取方法无法有效地表达地物类间差异和区分地物边界。有效解决光谱混淆和空间尺度问题是提高分类精度的关键。方法 结合多尺度超像素和奇异谱分析,提出一种新的高光谱影像分类方法,从而充分挖掘地物的局部空间特征和光谱特征,解决空间尺度和光谱混淆的问题,提高分类精度。利用多尺度超像素对影像进行分割,获取不同尺度的分割影像,同时在分割区域内进行均值滤波,减少类内的光谱差异,增强类间的光谱差异;对每个区域计算平均光谱向量,并利用奇异谱分析方法获取光谱的主要鉴别特征,同时消除噪声的影响;利用支持向量机对不同尺度超像素分割影像进行分类,并进行决策融合,得到最终的分类结果。结果 实验选取了两个标准高光谱数据集和一个真实数据集,结果表明,利用本文算法提取的光谱—空间特征进行分类,比直接在原始数据上进行分类分别提高约26.8%、9.2%和13%的精度;与先进的深度学习SSRN （spectral-spatial residual network）算法相比,本文算法在精度上分别提升约5.2%、0.7%和4%,并且运行时间仅为前者的18.3%、45.4%和62.1%,处理效率更高。此外,在训练样本有限的情况下,两个标准数据集的样本分别为1%和0.2%时,本文算法均能取得87%以上的分类精度。结论 针对高光谱影像分类中的难题,提出一种新的融合光谱和多尺度空间特征的HSI分类方法。实验结果表明,本文方法优于对比方法,可以产生更精细的分类结果。     

Image super-resolution using supervised multi-scale feature extraction network

Multimedia Tools and Applications - Image super-resolution using deep convolutional networks have recently achieved great successes. However, previous studies have failed to consider the spatial...     

Study of UAV tracking based on CNN in noisy environment

Multimedia Tools and Applications - Recently, there are lots of tracking methods proposed to improve the performance of visual tracking in videos with challenging situations, such as background...     

FAGNet:基于MAFPN和GVR的遥感图像多尺度目标检测算法

针对大场景遥感图像内容复杂,并且具有目标种类较多、尺度不一、方向多变等特点,导致遥感图像中目标多类多尺度多方向的问题,提出一种基于多尺度注意力特征金字塔网络(MAFPN)以及滑动顶点回归(GVR)机制的遥感图像目标检测方法.首先利用骨干网络提取多层特征作为MAFPN的输入,MAFPN结合特征融合和注意力机制,在融合多个尺度的特征映射的基础上使用通道域注意力和空间域注意力机制来抑制噪声,增强有效特征复用,提高网络对目标多尺度特征的自适应性;将MAFPN输出的融合特征图输入区域建议网络(RPN)生成感兴趣区域,然后将其送入分类/回归网络;在分类/回归网络中使用GVR机制在预测水平框的基础上增加4个顶点偏移比例参数和旋转因子,将水平框转换为旋转框,以减少边框中冗余区域,使预测得到的旋转边框更贴合目标.在DOTA公开数据集上与多种基于卷积神经网络的经典检测算法进行对比的实验结果表明,该方法的平均检测精度得到显著提高,能够更加准确地检测多个尺度以及多个方向的目标,实现了多尺度目标的鲁棒性检测.     

FGSC-23：面向深度学习精细识别的高分辨率光学遥感图像舰船目标数据集

目的 基于光学遥感图像的舰船目标识别研究广受关注,但是目前公开的光学遥感图像舰船目标识别数据集存在规模小、目标类别少等问题,难以训练出具有较高舰船识别精度的深度学习模型。为此,本文面向基于深度学习的舰船目标精细识别任务研究需求,搜集公开的包含舰船目标的高分辨率谷歌地球和GF-2卫星水面场景遥感图像,构建了一个高分辨率光学遥感图像舰船目标精细识别数据集（fine-grained ship collection-23,FGSC-23）。方法 将图像中的舰船目标裁剪出来,制作舰船样本切片,人工对目标类别进行标注,并在每个切片中增加舰船长宽比和分布方向两类属性标签,最终形成包含23个类别、4 052个实例的舰船目标识别数据集。结果 按1：4比例将数据集中各类别图像随机划分为测试集和训练集,并展开验证实验。实验结果表明,在通用识别模型识别效果验证中,VGG16（Visual Geometry Group 16-layer net）、ResNet50、Inception-v3、DenseNet121、MobileNet和Xception等经典卷积神经网络（convolutional neural network,CNN）模型的整体测试精度分别为79.88%、81.33%、83.88%、84.00%、84.24%和87.76%;在舰船目标精细识别的模型效果验证中,以VGG16和ResNet50模型为基准网络,改进模型在测试集上的整体测试精度分别为93.58%和93.09%。结论 构建的FGSC-23数据集能够满足舰船目标识别算法的验证任务。     

Video super-resolution reconstruction method based on deep Back projection and motion feature fusion

How to effectively utilize inter-frame redundancies is the key to improve the accuracy and speed of video super-resolution reconstruction methods. Previous methods usually process every frame in the whole video in the same way, and do not make full use of redundant information between frames, resulting in low accuracy or long reconstruction time. In this paper, we propose the idea of reconstructing key frames and non-key frames respectively, and give a video super-resolution reconstruction method based on deep back projection and motion feature fusion. Key-frame reconstruction subnet can obtain key frame features and reconstruction results with high accuracy. For non-key frames, key frame features can be reused by fusing them and motion features, so as to obtain accurate non-key frame features and reconstruction results quickly. Experiments on several public datasets show that the proposed method performs better than the state-of-the-art methods, and has good robustness.

     

A tensor decomposition based collaborative filtering algorithm for time-aware POI recommendation in LBSN

Point of interest (POI) recommendation problem in location based social network (LBSN) is of great importance and the challenge lies in the data sparsity, implicit user feedback and personalized preference. To improve the precision of recommendation, a tensor decomposition based collaborative filtering (TDCF) algorithm is proposed for POI recommendation. Tensor decomposition algorithm is utilized to fill the missing values in tensor (user-category-time). Specifically, locations are replaced by location categories to reduce dimension in the first phase, which effectively solves the problem of data sparsity. In the second phase, we get the preference rating of users to POIs based on time and user similarity computation and hypertext induced topic search (HITS) algorithm with spatial constraints, respectively. Finally the user’s preference score of locations are determined by two items with different weights, and the Top-N locations are the recommendation results for a user to visit at a given time. Experimental results on two LBSN datasets demonstrate that the proposed model gets much higher precision and recall value than the other three recommendation methods.

     

Observer‐based compensation control of servo systems with backlash

For compensating backlash phenomenon in servo systems, the authors propose an observer method in this paper to estimate both system states and vibration torque before controller design. First, a systematic scheme is given to obtain plant parameters, which is very important in observing system states. This is a parameter estimation principle that gives a crude estimation and computes the differences between the crude and true values. As a result, the precise value of the parameters is obtained by adding together the crude value and the difference. Then, based on the precise estimated parameters, an extended state observer (ESO) is designed to obtain feedback and feedforward signals. Consequently, robust compensation control is achieved by designing an output feedback controller, consisting of a feedback term and a feedforward term. Finally, in order to validate the proposed approach, extensive experiments are performed on a practical servo system with backlash nonlinearity.     

基于边缘敏感的SLIC和二次密度聚类的GGO分割

针对磨玻璃肺结节(Ground Glass Opacity,GGO)边界对比度低、大小各异和灰度不均匀等造成分割准确率低的问题,提出一种基于边缘敏感的SLIC和二次密度聚类相结合的分割算法。将图像边缘检测结果与SLIC超像素算法相结合,并将其中含有边缘的超像素块用区域质心代替其原始聚类中心,改善SLIC边界黏连性较差的问题;针对密度聚类不能完整分割GGO的问题,提出二次密度聚类的方法,对密度聚类定位到的簇及其邻域簇进行二次密度聚类。实验结果表明,该算法分割GGO的平均准确率达90.17%,灵敏度达84%。     

轻量化卷积神经网络在SAR图像语义分割中的应用

针对合成孔径雷达图像的语义分割问题,构建了一个全新的TerraSAR-X语义分割数据集GDUT-Nansha。然后,为解决传统深度学习方法模型体积大,难以在样本数量偏少的合成孔径雷达图像数据集上应用的问题,对轻量化卷积神经网络ENet模型进行了分析和改造。提出了一种改进的轻量化卷积神经网络模型(revised weighted loss eNet,RWL-ENet);针对合成孔径雷达图像数据集样本不平衡问题,使用了带有权重的损失函数。通过和其他经典卷积神经网络语义分割模型的对比实验,验证了新数据集的可靠性;同时,在参数量和模型体积远远小于其他网络模型的前提下,RWL-ENet模型在像素精度、平均像素精度、平均交并比三个定量指标上分别达到了0.884、0.804和0.645。