Deep learning-based framework for monitoring of debris-covered glacier from remotely sensed images

In recent years, deep learning (DL) methods have proven their efficiency for various computer vision (CV) tasks such as image classification, natural language processing, and object detection. However, training a DL model is expensive in terms of both complexities of the network structure and the amount of labeled data needed. In addition, the imbalance among available labeled data for different classes of interest may also adversely affect the model accuracy. This paper addresses these issues using a new convolutional neural network (CNN) based architecture. The proposed network incorporates both spatial and spectral information that combines two sub-networks: spatial-CNN and spectral-CNN. The spectral-CNN extracts spectral information, while spatial-CNN captures spatial information. Moreover, to make the features more robust, a multiscale spatial CNN architecture is introduced using different kernels. The final feature vector is formed by concatenating the outputs obtained from both spatial-CNN and spectral-CNN. To address the data imbalance problem, a generative adversarial network (GAN) was used to generate data for the underrepresented class. Finally, relatively a shallower network architecture was used to reduce the number of parameters in the network and improve the processing speed. The proposed model was trained and tested on Senitel-2 images for the classification of the debris-covered glacier. The results showed that the proposed method is well-suited for mapping and monitoring debris-covered glaciers at a large scale with high classification accuracy. In addition, we compared the proposed method with conventional machine learning approaches, support vector machine (SVM), random forest (RF) and multilayer perceptron (MLP).     

基于BoF模型的多特征融合纹理图像分类

针对特征词袋（BoF）模型缺乏空间和几何信息,对纹理图像内容表达不明显等问题,提出一种基于BoF模型的多特征融合纹理分类算法。将灰度梯度共生矩阵（GGCM）和尺度不变特征转换（SIFT）融合特征作为纹理图像的区域特征描述,通过动态权重鉴别能量分析进行最优参数特征选择,并用BoF量化纹理特征,使用支持向量机对图像进行训练和预测,得出分类结果。实验结果表明,本文算法对有旋转扭曲的纹理、边缘模糊纹理、有光照变化的纹理及杂乱纹理等均能取得较好的分类效果,相对于传统BoF模型及凹凸划分（CCP）方法等算法在UIUC纹理库上的分类正确率均有不同程度的提高,平均分类正确率分别提高12.8%和7.9%,说明本文算法针对纹理图像分类具有较高的精度和较好的鲁棒性。      

基于局部线性嵌入的高光谱影像特征提取算法

特征提取能够消除冗余信息,提高高光谱数据处理的精度和计算效率,是分类等分析必要的预处理手段.传统特征提取算法基于线性变换,无法准确描述高、低维特征空间的关系,因此采用一种新型非线性特征提取算法,即局部线性嵌入(LLE,Locally Linear Em-bedding),挖掘高光谱影像的本征信息.针对分类问题,使用训练样本类别属性修正距离矩阵,并借鉴LLE计算未知样本低维映射的方法求解测试样本的特征向量,实现监督局部线性嵌入(SLLE,Supervised Locally Linear Embedding).使用机载可见光/红外成像光谱仪数据,与3种分类算法结合进行测试,实验结果表明:SLLE优于线性特征提取算法,能够解决高光谱影像的小样本分类问题.     

一种地震动信号特征提取与分类方法

为了实时检测、识别和预警对地下基础设施的挖掘破坏活动，本文提出一种地震动信号特征提取与分类方法。通过提取小波包变换域和集合经验模态变换域的多域能量联合分布特征向量，构建改进的径向基神经网络分类模型，利用机器学习的方法提取稳定的信号多域融合特征，并实现准确的信号特征分类预测。由多类别挖掘信号的仿真实验结果可以看出，本文的算法和模型能有效提升地震动信号分类的准确率，对地震动干扰信号具有较强的鲁棒性。     

最优特征选择下多层次分割的城市道路提取

针对道路提取过程中特征维数过高的问题,提出了一种基于ReliefF过滤式和Wrapper封装式的特征选择方法.将粒子群优化算法(PSO)作为Wrapper的搜索算法,优化过的随机森林算法(OPRF)作为Wrapper的分类器构成PSO_OPRF封装式子集评估器,对ReliefF预选后的特征子集进行评估,降低特征维度,选...     

基于模糊和最小二乘的SAR图像线特征提取

针对合成孔径雷达(SAR, Synthetic Aperture Radar)图像的特点,提出一种线特征提取算法.该算法首先抑制相干斑噪声,然后对SAR图像进行模糊变换和增强,再基于假设的SAR图像灰度模型,用最小二乘法设计出低通和高通滤波器,最后检测出线特征散点,并进行连接,得到SAR图像的线特征图.实验证明本算法抗噪声干扰,边缘定位精度达到1个像素.     

基于卷积神经网络的遥感图像舰船目标检测

针对遥感图像背景复杂、受环境因素影响大的问题,提出一种将卷积神经网络(CNN)与支持向量机(SVM)相结合的舰船目标检测方法,利用卷积神经网络可自主提取图像特征并进行学习的优点,避免了复杂的特征选择和提取过程,在复杂海况背景图像的处理中体现出较优的性能;同时,由于军舰样本获取难度大,应用迁移学习的概念,利用大量民船样本辅助军舰目标的检测,取得较好的效果。通过参数调整与实验验证,此方法在自行建立的测试集上检测率达到90.59%,对光照、环境等外界因素具有一定程度的鲁棒性。     

基于NMF的SVM故障诊断方法

针对大维数系统故障诊断中存在特征提取困难和识别率低的问题,提出基于非负矩阵分解(NMF,Non-negative Matrix Factorization)的支持向量机(SVM,Support Vector Machine)诊断方法,避免了直接对故障特征的选择和提取,实现特征降维,提高故障模式分类的准确性和速度;对于NMF中的结果随机性问题,提出用前次分解所得系数矩阵求解样本降维特征矩阵的方法,保证多次NMF分解尺度一致.实验表明该方法能对故障特征有效降维,并具有较高的诊断效率和故障识别率.     

基于MSST及HOG特征提取的雷达辐射源信号识别

针对传统雷达信号识别算法在低信噪比下识别准确率低的问题,提出了基于多重同步压缩（MSST）时频变换及方向梯度直方图（HOG）特征提取的雷达辐射源信号识别算法。所提算法在雷达时域信号短时傅里叶变换（STFT）基础上进行多重同步压缩处理获得信号时频分布图,通过HOG算子对信号时频分布图进行HOG特征提取,将提取的HOG特征通过主成分分析法（PCA）进行降维,将降维后的特征参数送入支持向量机（SVM）对雷达信号进行分类与识别。实验结果表明：所提算法具有较低的复杂度,当信噪比为-8 dB时,仿真实验与半实物仿真实验针对9种典型雷达信号的识别准确率达到90%以上。     

基于多种空间信息的高光谱遥感图像分类方法

在高光谱遥感图像分类方法中,空间特征和光谱特征的融合可以有效地改善分类效果。针对单一空间特征的信息表达不充分问题,提出了一种联合多种空间特征的高光谱图像空谱分类方法。利用超像素信息对分类结果进行后处理去掉椒盐噪声,并创造性地将超像素信息应用于分类前处理,提出了一种利用超像素信息对像素点的特征向量进行线性加权融合的方法。试验结果表明,所提方法的性能优于目前的通常方法。     

基于联合正则化策略的人脸表情识别方法

针对目前人脸表情识别大多采用基于深度学习的端到端特征提取及分类方法的现象，提出了一种新的深度模型优化方法。基于ResNet18残差网络架构和正则化思想，提出了联合正则化策略，即将过滤器响应正则化和批量正则化、实例正则化和组正则化、组正则化和批量正则化分别嵌入网络之中，平衡和改善特征数据分布，弥补单一正则化的缺点，提升模型性能。在2个公开数据集FER2013和CK+进行了验证和测试，最高准确率分别达到了73.558%和94.9%，实验结果表明，联合正则化策略提高了基础网络的性能，其表现优于诸多当前较新的人脸表情识别方法。      

Multi-seasonal evaluation of hybrid poplar (P. Deltoides) plantations using Worldview-3 imagery and State-Of-The-Art ensemble learning algorithms

Forest resources are the primary components of the ecosystem environment. Poplars (Populus sp.), a member of the fast-growing trees, are one of the most productive forest tree species for industrial production thanks to their desirable traits comprising rapid growth, hybridization ability, and ease of propagation. Determining poplar cultivated areas and mapping their geographical distributions is critical for planners and decision-makers to increase the ecological and economic benefits of poplar trees. Due to the biodiversity of each geographical region and seasonal vegetation variations, classification based on remotely sensed imagery is essential for cropland monitoring. The main goal of this study is to evaluate the potential of high-resolution multi-temporal (growing season and end of the growing season) Worldview-3 imagery in mapping poplar plantations in the Akyaz? district of Sakarya, Turkey. For this purpose, pixel- and object-based image analysis with up-to-date ensemble learning algorithms, namely random forest (RF), categorical boosting (CB), and extreme gradient boosting tree (XGB), were employed for mapping poplar fields. Results indicated that the object-based classification approach provided statistically significant improvements in map-level (about 4%) and class-level accuracy (e.g., approximately 7% and %2 for poplar and young poplar classes, respectively) than pixel-based classification. While the CB performed superior classification performance for the object-based approach (92.56%), the highest classification performance was obtained with the XGB algorithm for the pixel-based approach (90.42%) for the end of the growing season data. McNemar’s statistical test also confirmed that the performances of CB and XGB algorithms were statistically similar in pixel-based classification. Finally, analysis of multi-season images revealed that sensitivity of the vegetation phenology and seasonal effects considerably affect the separability of poplar tree species.     

嵌入注意力机制的多尺度深度可分离表情识别

针对面部表情识别中，传统机器学习方法特征提取较为复杂，浅层卷积神经网络识别率不高，以及深度卷积神经网络易带来梯度爆炸或弥散的问题，构建了残差网络嵌入注意力机制的多尺度深度可分离表情识别网络。通过多层多尺度深度可分离残差单元的叠加进行不同尺度的表情特征提取，使用CBAM注意力机制进行表情特征的筛选，提升有效表情特征权重的表达，削弱训练数据的噪声影响。所提网络模型在Fer-2103和CK+表情数据集分别取得了73.89%和97.47%的准确度，表明所提网络具有较强的泛化性。      

Development and testing of a statistical texture model for land cover classification of the Black Sea region with MODIS imagery

A statistical model is proposed for analysis of the texture of land cover types for global and regional land cover classification by using texture features extracted by multiresolution image analysis techniques. It consists of four novel indices representing second-order texture, which are calculated after wavelet decomposition of an image and after texture extraction by a new approach that makes use of a four-pixel texture unit. The model was applied to four satellite images of the Black Sea region, obtained by Terra/MODIS and Aqua/MODIS at different spatial resolution. In single texture classification experiments, we used 15 subimages (50 × 50 pixels) of the selected classes of land covers that are present in the satellite images studied. These subimages were subjected to one-level and two-level decompositions by using orthonormal spline and Gabor-like spline wavelets. The texture indices were calculated and used as feature vectors in the supervised classification system with neural networks. The testing of the model was based on the use of two kinds of widely accepted statistical texture quantities: five texture features determined by the co-occurrence matrix (angular second moment, contrast, correlation, inverse difference moment, entropy), and four statistical texture features determined after the wavelet transformation (mean, standard deviation, energy, entropy). The supervised neural network classification was performed and the discrimination ability of the proposed texture indices was found comparable with that for the sets of five GLCM texture features and four wavelet-based texture features. The results obtained from the neural network classifier showed that the proposed texture model yielded an accuracy of 92.86% on average after orthonormal wavelet decomposition and 100% after Gabor-like wavelet decomposition for texture classification of the examined land cover types on satellite images.     

基于随机森林的航天器电信号多分类识别方法

针对航天器电特性信号数据存在数据量大、特征维数高、计算复杂度大和识别率低等问题,提出基于主成分分析(PCA)的特征提取方法和随机森林(RF)算法,对原始数据进行降维,提高计算效率和识别率,实现对航天器电信号数据的快速、准确识别分类。随机森林算法在处理高维数据上具有优越的性能,但是考虑到时间复杂度问题,利用主成分分析方法对数据进行压缩和降维,在保证准确率的同时提高了计算效率。实验结果表明:与其他算法相比,针对航天器电特性信号数据,本文方法在准确率、计算效率和稳定性等方面均显示出优异的性能。     

结合自组织映射网络及三角形算法的星图识别方法

三角形方法是最经典且应用最广的星图识别方法之一,但是存在搜索范围大、匹配冗余、抗噪能力弱等问题。将神经网络技术应用到星图识别过程中,结合自组织映射网络（SOM）优秀的分类能力和三角形算法可靠的角距匹配能力,提出了一种新的识别方法。该方法基于邻近星的分布来构建每颗导航星的特征向量,将其作为SOM网络的输入向量,通过训练得到具有分类识别功能的网络及相应的三角形库。识别阶段,输入待识别星的特征向量,网络输出识别类,在该类对应的三角形库中应用三角形算法查找匹配三角形,完成星图识别。试验发现该方法减小三角形搜索范围、实现快速匹配的同时,提高了识别系统的抗噪能力,在全天识别过程中平均识别时间低于5ms,识别率在噪声标准差为0.025时仍高达99%。     

基于多尺度残差卷积网络的HEVC视频隐写分析

图片、语音、视频等多媒体形式的信息交流在网络通信中占有重要地位,同时也有很多非法信息的传播隐匿于此。隐写分析是甄别隐秘信息是否存在的有效手段,提出了一种通用的基于多尺度残差卷积网络的HEVC视频隐写分析算法。网络主体由残差计算、特征提取和二分类3部分构成,其中在特征提取部分针对性地提出了残差卷积层、多尺度残差卷积模块及隐写分析残差块。实验结果表明:所提算法基于视频像素域分析网络的检测率高达99.75%,比传统的手工提取特征方法具有更大的优势。      

一种改进的ORB特征匹配算法

针对现有的ORB特征匹配算法在图像模糊、光照变化、图像压缩、噪声条件下,匹配准确率下降问题,提出了一种改进的ORB特征匹配算法。首先,在提取特征点过程中,对图像进行网格化处理,并引入四叉树结构,使提取的特征点在图像中均匀分布,解决传统的特征提取方法遇到的特征点集中问题。然后,利用暴力匹配进行初步匹配,并采用交叉验证的方式,剔除部分误匹配,改善暴力匹配的结果。最后,利用高斯核对网格运动统计的结果做加权处理,优化统计结果,进一步剔除误匹配,得到准确率更高的匹配集合。实验结果表明:改进后的算法在图像模糊、光照变化、图像压缩和噪声条件下,平均准确率分别提高了3.5%、4.2%、2.2%和6%。