自监督学习下小样本遥感图像场景分类

目的 卷积神经网络(convolutional neural network, CNN)在遥感场景图像分类中广泛应用,但缺乏训练数据依然是不容忽视的问题。小样本遥感场景分类是指模型只需利用少量样本训练即可完成遥感场景图像分类任务。虽然现有基于元学习的小样本遥感场景图像分类方法可以摆脱大数据训练的依赖,但模型的泛化能力依然较弱。为了解决这一问题,本文提出一种基于自监督学习的小样本遥感场景图像分类方法来增加模型的泛化能力。方法 本文方法分为两个阶段。首先,使用元学习训练老师网络直到收敛;然后,双学生网络和老师网络对同一个输入进行预测。老师网络的预测结果会通过蒸馏损失指导双学生网络的训练。另外,在图像特征进入分类器之前,自监督对比学习通过度量同类样本的类中心距离,使模型学习到更明确的类间边界。两种自监督机制能够使模型学习到更丰富的类间关系,从而提高模型的泛化能力。结果 本文在NWPU-RESISC45(North Western Polytechnical University-remote sensing image scene classification)、AID(aerial ima...     

基于改进胶囊神经网络的遥感影像分类

针对遥感影像卷积神经网络（CNN）分类会导致特征信息丢失及泛化能力差的问题,提出一种基于通道注意力和混合注意力改进的胶囊神经网络分类模型。首先,为了胶囊神经网络能够适应于大尺寸输入图像,在特征提取模块中使用2个最大池化层;其次,为了提高分类精度,分别将SENet注意力和CBAM注意力加在特征提取模块的最后一层去改进特征提取模块;最后,将样本集随机地划分为训练集、验证集和测试集,进一步使用训练集和验证集训练模型,测试集测试模型,使用AID数据集对模型分类的泛化能力进行验证。实验结果表明:基于SENet网络改进的胶囊神经网络的准确率与Kappa系数要高于其他模型,泛化能力也优于其他模型,本文提出的模型的总体分类精度和泛化能力有了显著性提升,从而验证了本文方法的可行性和使用性。     

Growing random forest on deep convolutional neural networks for scene categorization

Breakthrough performances have been achieved in computer vision by utilizing deep neural networks. In this paper we propose to use random forest to classify image representations obtained by concatenating multiple layers of learned features of deep convolutional neural networks for scene classification. Specifically, we first use deep convolutional neural networks pre-trained on the large-scale image database Places to extract features from scene images. Then, we concatenate multiple layers of features of the deep neural networks as image representations. After that, we use random forest as the classifier for scene classification. Moreover, to reduce feature redundancy in image representations we derived a novel feature selection method for selecting features that are suitable for random forest classification. Extensive experiments are conducted on two benchmark datasets, i.e. MIT-Indoor and UIUC-Sports. Obtained results demonstrated the effectiveness of the proposed method. The contributions of the paper are as follows. First, by extracting multiple layers of deep neural networks, we can explore more information of image contents for determining their categories. Second, we proposed a novel feature selection method that can be used to reduce redundancy in features obtained by deep neural networks for classification based on random forest. In particular, since deep learning methods can be used to augment expert systems by having the systems essentially training themselves, and the proposed framework is general, which can be easily extended to other intelligent systems that utilize deep learning methods, the proposed method provide a potential way for improving performances of other expert and intelligent systems.     

基于卷积神经网络的图像数据增强算法

提升卷积神经网络的泛化能力和降低过拟合的风险是深度卷积神经网络的研究重点。遮挡是影响卷积神经网络泛化能力的关键因素之一,通常希望经过复杂训练得到的模型能够对遮挡图像有良好的泛化性。为了降低模型过拟合的风险和提升模型对随机遮挡图像识别的鲁棒性,提出了激活区域处理算法,在训练过程中对某一卷积层的最大激活特征图进行处理后对输入图像进行遮挡,然后将被遮挡的新图像作为网络的新输入并继续训练模型。实验结果表明,提出的算法能够提高多种卷积神经网络模型在不同数据集上的分类性能,并且训练好的模型对随机遮挡图像的识别具有非常好的鲁棒性。     

COVID-19 chest X-ray image classification in the presence of noisy labels

The Corona Virus Disease 2019 (COVID-19) has been declared a worldwide pandemic, and a key method for diagnosing COVID-19 is chest X-ray imaging. The application of convolutional neural network with medical imaging helps to diagnose the disease accurately, where the label quality plays an important role in the classification problem of COVID-19 chest X-rays. However, most of the existing classification methods ignore the problem that the labels are hardly completely true and effective, and noisy labels lead to a significant degradation in the performance of image classification frameworks. In addition, due to the wide distribution of lesions and the large number of local features of COVID-19 chest X-ray images, existing label recovery algorithms have to face the bottleneck problem of the difficult reuse of noisy samples. Therefore, this paper introduces a general classification framework for COVID-19 chest X-ray images with noisy labels and proposes a noisy label recovery algorithm based on subset label iterative propagation and replacement (SLIPR). Specifically, the proposed algorithm first obtains random subsets of the samples multiple times. Then, it integrates several techniques such as principal component analysis, low-rank representation, neighborhood graph regularization, and k-nearest neighbor for feature extraction and image classification. Finally, multi-level weight distribution and replacement are performed on the labels to cleanse the noise. In addition, for the label-recovered dataset, high confidence samples are further selected as the training set to improve the stability and accuracy of the classification framework without affecting its inherent performance. In this paper, three typical datasets are chosen to conduct extensive experiments and comparisons of existing algorithms under different metrics. Experimental results on three publicly available COVID-19 chest X-ray image datasets show that the proposed algorithm can effectively recover noisy labels and improve the accuracy of the image classification framework by 18.9% on the Tawsifur dataset, 19.92% on the Skytells dataset, and 16.72% on the CXRs dataset. Compared to the state-of-the-art algorithms, the gain of classification accuracy of SLIPR on the three datasets can reach 8.67%-19.38%, and the proposed algorithm also has certain scalability while ensuring data integrity.     

改进LeNet-5网络在图像分类中的应用

LeNet-5卷积神经网络（CNN）虽然在手写数字识别上取得很好的分类效果,但在具有复杂纹理特征的数据集上分类精度不高。为提高网络在复杂纹理特征图像上分类的正确率,提出一种改进的LeNet-5网络结构。引入跨连思想,充分利用网络提取的低层次特征;把Inception V1模块嵌入LeNet-5卷积神经网络,提取图像的多尺度特征;输出层使用softmax函数对图像进行分类。在Cifar-10和Fashion MNIST数据集上进行的实验结果表明,改进的卷积神经网络在复杂纹理特征数据集上具有很好的分类能力。     

有序判别典型相关分析

多视图学习方法通过视图间互补信息的融合,达到增强单一视图方法的鲁棒性并提升学习性能的目的.典型相关分析(canonical correlation analysis,简称CCA)是一种重要的多视图信息融合技术.其研究的是针对同一组目标两组不同观测数据间的相关性,目标是得到一组相关性最大的投影向量.但当面对标号有序的分类任务时,CCA因没有利用类信息和类间有序信息,造成了对分类性能的制约.为此,通过将有序类信息嵌入CCA进行扩展,发展出有序判别典型相关分析(ordinal discriminative canonical correlation analysis,简称OR-DisCCA).实验结果表明, OR-DisCCA的性能比相关方法更优.     

Ordinal factorization machine with hierarchical sparsity

Ordinal regression(OR)or classification is a machine learning paradigm for ordinal labels.To date,there have been a variety of methods proposed including kernel based and neural network based methods with significant performance.However,existing OR methods rarely consider latent structures of given data,particularly the interaction among covariates,thus losing interpretability to some extent.To compensate this,in this paper,we present a new OR method:ordinal factorization machine with hierarchical sparsity(OFMHS),which combines factorization machine and hierarchical sparsity together to explore the hierarchical structure behind the input variables.For the sake of optimization,we formulate OFMHS as a convex optimization problem and solve it by adopting the efficient alternating directions method of multipliers(ADMM)algorithm.Experimental results on synthetic and real datasets demonstrate the superiority of our method in both performance and significant variable selection.     

A physics based generative adversarial network for single image defogging

In the field of single image defogging, there are two main methods. One is the image restoration method based on the atmospheric scattering theory which can recover the image texture details well. The other is the image enhancement method based on Retinex theory which can improve the image contrast well. In practice, however, the former can easily lead to low contrast images; the latter is prone to losing texture details. Therefore, how to effectively combine the advantages of both to remove fog is a key issue in the field. In this paper, we have developed a physics based generative adversarial network (PBGAN) to exploit the advantages between those two methods in parallel. To our knowledge, it is the first learning defogging framework that incorporates these two methods and to enable them to work together and complement each other. Our method has two generative adversarial modules, the Contrast Enhancement (CE) module and the Texture Restoration (TR) module. To improve contrast in the CE module, we introduced a novel inversion-adversarial loss and a novel inversion-cycle consistency loss for training the generator. To improve the texture in the TR module, we introduced two convolutional neural networks to learn the atmospheric light coefficient and the transmission map, respectively. Extensive experiments on both synthetic and real-world datasets demonstrate that the proposed approach performs better than several state-of-the-art methods quantitatively and qualitatively.     

结合改进卷积神经网络与通道加权的轻量级表情识别

目的 表情是人机交互过程中重要的信息传递方式,因此表情识别具有重要的研究意义。针对目前表情识别方法存在背景干扰大、网络模型参数复杂、泛化性差等问题,本文提出了一种结合改进卷积神经网络(convolutional neural network, CNN)与通道加权的轻量级表情识别方法。方法 首先,采用标准卷积和深度可分离卷积组合神经网络结构,再利用全局平均池化层作为输出层,简化网络的复杂程度,有效降低网络参数;其次,网络引入SE(squeeze-and-excitation)模块进行通道加权,通过在不同卷积层后设置不同的压缩率增强表情特征提取能力,提升网络模型精度;最后,用softmax分类函数实现各类表情的准确分类。结果 本文网络参数量为6 108 519,相较于识别性能较好的Xception神经网络参数减少了63%,并且通过对网络模型的实时性测试,平均识别速度可达128帧/s。在5个公开的表情数据集上验证网络模型对7种表情的识别效果,与7种卷积神经网络方法相比,在FER2013 (Facial Expression Recognition 2013)、CK+(the extended...     

基于多尺度特征卷积神经网络的目标定位

针对实际应用中诸多数据集标签部分缺失、无定位标注等问题，提出了基于多尺度特征卷积神经网络的弱监督定位算法。其核心思想是利用神经网络分层的特性，在多层卷积层上使用梯度加权类激活映射，生成梯度金字塔模型，并通过均值滤波计算特征质心位置，利用置信强度映射和阈值梯减模块产生连接的像素段，围绕最大边界标注进行弱监督定位。在标准测试集上的实验结果表明，该算法能够在存在大量类别、多尺度图像的情况下完成目标定位，具有较高的精确度。     

基于自注意力孪生网络的野生蛇细粒度分类

野生蛇的分类相较于其他细粒度图像分类更加困难和复杂, 这是因为蛇姿势各异、变化急促、常处于运动或盘曲状态, 很难根据蛇的局部特征去判断并分类. 为了解决这个问题, 本文将自注意力机制应用野生蛇细粒度图像分类, 从而解决卷积神经网络因层数加深造成的过于专注局部而忽略全局信息问题. 通过Swin Transformer (Swin-T)进行迁移学习获得细粒度特征提取模型. 为了进一步研究自注意力机制在元学习领域的性能, 本文改进特征提取模型搭建孪生网络并构造元学习器对少量样本进行学习和分类. 相较于其他方法, 本方法减少了元学习在特征提取时所造成的时间和空间开销, 提高了元学习分类的准确率和效率并增加了元学习的自主学习性.     

Big Data Analytics with Optimal Deep Learning Model for Medical Image Classification

In recent years, huge volumes of healthcare data are getting generated in various forms. The advancements made in medical imaging are tremendous owing to which biomedical image acquisition has become easier and quicker. Due to such massive generation of big data, the utilization of new methods based on Big Data Analytics (BDA), Machine Learning (ML), and Artificial Intelligence (AI) have become essential. In this aspect, the current research work develops a new Big Data Analytics with Cat Swarm Optimization based deep Learning (BDA-CSODL) technique for medical image classification on Apache Spark environment. The aim of the proposed BDA-CSODL technique is to classify the medical images and diagnose the disease accurately. BDA-CSODL technique involves different stages of operations such as preprocessing, segmentation, feature extraction, and classification. In addition, BDA-CSODL technique also follows multi-level thresholding-based image segmentation approach for the detection of infected regions in medical image. Moreover, a deep convolutional neural network-based Inception v3 method is utilized in this study as feature extractor. Stochastic Gradient Descent (SGD) model is used for parameter tuning process. Furthermore, CSO with Long Short-Term Memory (CSO-LSTM) model is employed as a classification model to determine the appropriate class labels to it. Both SGD and CSO design approaches help in improving the overall image classification performance of the proposed BDA-CSODL technique. A wide range of simulations was conducted on benchmark medical image datasets and the comprehensive comparative results demonstrate the supremacy of the proposed BDA-CSODL technique under different measures.     

提高小样本高光谱图像分类性能的变维卷积神经网络

目的 为了解决基于卷积神经网络的算法对高光谱图像小样本分类精度较低、模型结构复杂和计算量大的问题,提出了一种变维卷积神经网络。方法 变维卷积神经网络对高光谱分类过程可根据内部特征图维度的变化分为空—谱信息融合、降维、混合特征提取与空—谱联合分类的过程。这种变维结构通过改变特征映射的维度,简化了网络结构并减少了计算量,并通过对空—谱信息的充分提取提高了卷积神经网络对小样本高光谱图像分类的精度。结果 实验分为变维卷积神经网络的性能分析实验与分类性能对比实验,所用的数据集为Indian Pines和Pavia University Scene数据集。通过实验可知,变维卷积神经网络对高光谱小样本可取得较高的分类精度,在Indian Pines和Pavia University Scene数据集上的总体分类精度分别为87.87%和98.18%,与其他分类算法对比有较明显的性能优势。结论 实验结果表明,合理的参数优化可有效提高变维卷积神经网络的分类精度,这种变维模型可较大程度提高对高光谱图像中小样本数据的分类性能,并可进一步推广到其他与高光谱图像相关的深度学习分类模型中。     

结合自适应融合网络与哈希的图像检索算法

卷积神经网络因其对图像识别准确率高而在图像检索领域备受青睐,但处理大规模数据集时,基于卷积神经网络提取的深度特征维度高,容易引发"维度灾难".针对图像检索中深度特征维度高的问题,提出一种基于自适应融合网络特征提取与哈希特征降维的图像检索算法.由于传统哈希处理高维特征复杂度高,因此本文在卷积神经网络中加入自适应融合模块对特征进行重新整合,增强特征表征能力的同时降低特征维度;然后应用稀疏化优化算法对深度特征进行第2次降维,并通过映射获得精简的哈希码;最后,实验以Inception网络作为基础模型,在数据集CIFAR-10和ImageNet上进行了丰富的实验.实验结果表明,该算法能有效提高图像检索效率.     

高光谱图像小样本分类的卷积神经网络方法

目的 与传统分类方法相比,基于深度学习的高光谱图像分类方法能够提取出高光谱图像更深层次的特征。针对现有深度学习的分类方法网络结构简单、特征提取不够充分的问题,提出一种堆叠像元空间变换信息的数据扩充方法,用于解决训练样本不足的问题,并提出一种基于不同尺度的双通道3维卷积神经网络的高光谱图像分类模型,来提取高光谱图像的本质空谱特征。方法 通过对高光谱图像的每一像元及其邻域像元进行旋转、行列变换等操作,丰富中心像元的潜在空间信息,达到数据集扩充的作用。将扩充之后的像素块输入到不同尺度的双通道3维卷积神经网络学习训练集的深层特征,实现更高精度的分类。结果 5次重复实验后取平均的结果表明,在随机选取了10%训练样本并通过8倍数据扩充的情况下,Indian Pines数据集实现了98.34%的总体分类精度,Pavia University数据集总体分类精度达到99.63%,同时对比了不同算法的运行时间,在保证分类精度的前提下,本文算法的运行时间短于对比算法,保证了分类模型的稳定性、高效性。结论 本文提出的基于双通道卷积神经网络的高光谱图像分类模型,既解决了训练样本不足的问题,又综合了高光谱图像的光谱特征和空间特征,提高了高光谱图像的分类精度。     

Design of multi-scale receptive field convolutional neural network for surface inspection of hot rolled steels

Due to the large intra-class variations and unbalanced training samples, the accuracy of existing algorithms used in defect classification of hot rolled steels is unsatisfactory. In this paper, a new hierarchical learning framework is proposed based on convolutional neural networks to classify hot rolled defects. Multi-scale receptive field is introduced in the new framework to extract multi-scale features, which can better represent defects than the feature maps produced by a single convolutional layer. A group of AutoEncoders are trained to reduce the dimension of the extracted multi-scale features which improve the generalization ability under insufficient training samples. Besides, to mitigate the deviation caused by fine-tuning the pre-trained model with images of different context, we add a penalty term in the loss function, which is to reconstruct the input image from the feature maps produced by the pre-trained model, to help network encode more effective and structured information. The experiments with samples captured from two hot rolled production lines showed that the proposed framework achieved a classification rate of 97.2% and 97% respectively, which are much higher than the conventional methods.     

基于可重叠混淆树的卷积神经网络

卷积神经网络(convolutional neural network, CNN)被广泛用于图像分类任务中。大多数现有的CNN模型都按照N路分类器的形式训练。然而,不同类别之间总存在差异性限制了N路分类器的分类能力。为了解决上述问题,提出的神经网络模型将混淆树结构(confusion tree, CT)和CNN模型结合,设计了性能更强的基于混淆树的卷积神经网络模型(confusion tree CNN,CT-CNN)。该模型首先建立一个混淆树来对类别之间的混淆性进行建模;然后,将混淆树的分层结构嵌入到CNN模型中,通过这种方式可以引导CNN的训练过程更加关注混淆性强的类别集合。该模型在公共数据集上进行了评估,实验结果证明,CT-CNN能克服大规模数据类别间的分类难度分布不均匀的局限,在复杂大规模的分类任务中取得稳定的优秀表现。     

Tripartite-structure transformer for hyperspectral image classification

Hyperspectral images contain rich spatial and spectral information, which provides a strong basis for distinguishing different land-cover objects. Therefore, hyperspectral image (HSI) classification has been a hot research topic. With the advent of deep learning, convolutional neural networks (CNNs) have become a popular method for hyperspectral image classification. However, convolutional neural network (CNN) has strong local feature extraction ability but cannot deal with long-distance dependence well. Vision Transformer (ViT) is a recent development that can address this limitation, but it is not effective in extracting local features and has low computational efficiency. To overcome these drawbacks, we propose a hybrid classification network that combines the strengths of both CNN and ViT, names Spatial-Spectral Former(SSF). The shallow layer employs 3D convolution to extract local features and reduce data dimensions. The deep layer employs a spectral-spatial transformer module for global feature extraction and information enhancement in spectral and spatial dimensions. Our proposed model achieves promising results on widely used public HSI datasets compared to other deep learning methods, including CNN, ViT, and hybrid models.     

特征分组提取融合深度网络手写汉字识别

针对传统脱机手写汉字识别的过程复杂、精度低，而常用卷积神经网络的特征信息提取不充分，同时存在相同特征信息的重叠和冗余问题。设计了一个特征分组提取融合的深度卷积神经网络模型。通过多级堆叠的特征分组提取模块，提取图像的深层抽象特征信息，并进行特征信息之间的交流融合。利用设计的下采样和通道扩增模块，在降低特征维度的同时保留图像重要信息。将特征信息进行精炼和浓缩，来解决特征信息的重叠和冗余问题。最终训练出的神经网络达到top1当前先进的正确率为97.16%，同时top5正确率为99.36%，并具有很好的泛化能力。