A fast hybrid computer vision technique for real-time embedded bus passenger flow calculation through camera

Bus passenger flow calculation system is a critical part of the smart public transportation framework. Bus passenger flow information can help to make data statistics report of the passenger at a bus station which can be used by public transport operator to evaluate the quality of the transportation. Statistics report of crowded passengers in the bus station help managers to understand the bus transit operations, can provide the database for the intelligent transportation scheduling, help to provide more and better services for passengers, overall data statistics of passengers has important practical significance to improve public transport environment. This paper presents a passenger counting algorithm based on hybrid machine learning approach. In the first step, an advanced method is used to extract the Histogram of oriented gradients (HOG) feature of passenger’s heads. Classification of head features is done by using support vector machine (SVM) as a classifier for the liner model. Heads are detected successfully after performing all steps. In next step Kanade-Lucas-Tomasi (KLT) is used to reality head tracking, the multiple target tracking is achieved and the head motion trajectory of passenger target is captured stably. At last, the trajectory is analyzed and the automatic counting of bus passenger flow is realized. In the last step, the proposed algorithm is move to embedded system for practical implementation. In this paper, the algorithm intends to use ADSP-BF609 embedded platform for transplantation. The experimental results demonstrate that the statistical accuracy of the proposed algorithm is enhanced successfully; especially during the daytime with the good illustration, the effective counting of the passenger flow is achieved and the inward and outward passenger counting can be realized. In this paper three feature extraction models are used namely local binary patterns, histograms of oriented gradients and binarized statistical image in order to get accurate features. Furthermore, three common classification techniques including naïve bayes classifier, boosted tress and support vector machines are used for fine classification of extracted vectors obtained from different features extractors model. 94.50% accuracy is achieved when support vector machine (SVM) classifies the features extracted using Histogram of oriented gradients (HOG). SVM surpasses the accuracy obtained by Boosted tree namely 81.30% using Histogram of oriented gradients (HOG) features.

     

Development of Efficient Classification Systems for the Diagnosis of Melanoma

Skin cancer is usually classified as melanoma and non-melanoma. Melanoma now represents 75% of humans passing away worldwide and is one of the most brutal types of cancer. Previously, studies were not mainly focused on feature extraction of Melanoma, which caused the classification accuracy. However, in this work, Histograms of orientation gradients and local binary patterns feature extraction procedures are used to extract the important features such as asymmetry, symmetry, boundary irregularity, color, diameter, etc., and are removed from both melanoma and non-melanoma images. This proposed Efficient Classification Systems for the Diagnosis of Melanoma (ECSDM) framework consists of different schemes such as preprocessing, segmentation, feature extraction, and classification. We used Machine Learning (ML) and Deep Learning (DL) classifiers in the classification framework. The ML classifier is Naïve Bayes (NB) and Support Vector Machines (SVM). And also, DL classification framework of the Convolution Neural Network (CNN) is used to classify the melanoma and benign images. The results show that the Neural Network (NNET) classifier’ achieves 97.17% of accuracy when contrasting with ML classifiers.     

Corona-Nidaan: lightweight deep convolutional neural network for chest X-Ray based COVID-19 infection detection

The coronavirus COVID-19 pandemic is today’s major public health crisis, we have faced since the Second World War. The pandemic is spreading around the globe like a wave, and according to the World Health Organization’s recent report, the number of confirmed cases and deaths are rising rapidly. COVID-19 pandemic has created severe social, economic, and political crises, which in turn will leave long-lasting scars. One of the countermeasures against controlling coronavirus outbreak is specific, accurate, reliable, and rapid detection technique to identify infected patients. The availability and affordability of RT-PCR kits remains a major bottleneck in many countries, while handling COVID-19 outbreak effectively. Recent findings indicate that chest radiography anomalies can characterize patients with COVID-19 infection. In this study, Corona-Nidaan, a lightweight deep convolutional neural network (DCNN), is proposed to detect COVID-19, Pneumonia, and Normal cases from chest X-ray image analysis; without any human intervention. We introduce a simple minority class oversampling method for dealing with imbalanced dataset problem. The impact of transfer learning with pre-trained CNNs on chest X-ray based COVID-19 infection detection is also investigated. Experimental analysis shows that Corona-Nidaan model outperforms prior works and other pre-trained CNN based models. The model achieved 95% accuracy for three-class classification with 94% precision and recall for COVID-19 cases. While studying the performance of various pre-trained models, it is also found that VGG19 outperforms other pre-trained CNN models by achieving 93% accuracy with 87% recall and 93% precision for COVID-19 infection detection. The model is evaluated by screening the COVID-19 infected Indian Patient chest X-ray dataset with good accuracy.

     

PSO-SVM算法在肝脏B超图像识别中的应用

为提高肝脏B超图像的诊断准确率,研究了将粒子群算法(Particle Swarm Optimization,PSO)和支持向量机(Support Vec-tor Machine,SVM)相结合进行肝脏B超图像识别的方法;该方法首先提取肝脏B超图像的空域和频域的纹理特征,然后运用SVM对108幅肝脏B超图像进行分类,利用PSO算法优化SVM的模型参数,最后将该方法与基于网格搜索法优化的SVM和基于BP神经网络的分类方法进行了对比;实验结果表明,在PSO-SVM算法下,所提取的两种纹理特征相结合能够有效地描述肝脏B超图像,基于粒子群优化算法的支持向量机模型具有较高的识别精度,平均分类准确率达94.44%,这就表明PSO-SVM算法适用于对肝脏B超图像的识别。     

Deep learning based detection of COVID-19 from chest X-ray images

The whole world is facing a health crisis, that is unique in its kind, due to the COVID-19 pandemic. As the coronavirus continues spreading, researchers are concerned by providing or help provide solutions to save lives and to stop the pandemic outbreak. Among others, artificial intelligence (AI) has been adapted to address the challenges caused by pandemic. In this article, we design a deep learning system to extract features and detect COVID-19 from chest X-ray images. Three powerful networks, namely ResNet50, InceptionV3, and VGG16, have been fine-tuned on an enhanced dataset, which was constructed by collecting COVID-19 and normal chest X-ray images from different public databases. We applied data augmentation techniques to artificially generate a large number of chest X-ray images: Random Rotation with an angle between ??10 and 10 degrees, random noise, and horizontal flips. Experimental results are encouraging: the proposed models reached an accuracy of 97.20?% for Resnet50, 98.10?% for InceptionV3, and 98.30?% for VGG16 in classifying chest X-ray images as Normal or COVID-19. The results show that transfer learning is proven to be effective, showing strong performance and easy-to-deploy COVID-19 detection methods. This enables automatizing the process of analyzing X-ray images with high accuracy and it can also be used in cases where the materials and RT-PCR tests are limited.

     

基于SVM的大鱼际掌纹图像二分类法

提出一种基于支持向量机(SVM)的大鱼际掌纹图像二分类法。采用高频强调滤波,对分割得到的大鱼际掌纹图像进行图像增强,提取其灰度共生矩阵4个方向的8个特征量作为分类特征向量。对比不同核函数下的分类准确率,结果表明,组合特征向量的SVM方法对大鱼际掌纹的初步二分类效果较好。     

基于多特征融合的交通标志识别算法

综合考虑识别率、时间复杂度以及鲁棒性,提出一种边缘、纹理、颜色多特征融合和支持向量机（SVM）的交通标志识别算法。通过提取能够描述交通标志图像边缘信息的方向梯度直方图（HOG）特征并进行统计平均,与能够表示标志图像内部纹理信息的局部二值模式（LBP）特征融合得到降维后的HOG-maxLBP特征,再级联交通标志的颜色特征作为最终的特征向量,最后利用SVM进行交通标志训练和分类。实验结果表明,该算法不仅提高了交通标志的识别率,而且降低了时间复杂度,增强了系统鲁棒性。     

基于最大相关和最小冗余准则及极限学习机的癫痫发作检测方法

癫痫发作检测可以实现脑电分类和病灶定位,对癫痫的临床治疗具有重要意义。针对大数据量、高特征值空间长程脑电的快速和准确分类问题,提出一种基于最大相关和最小冗余准则及极限学习机的癫痫发作检测方法。对脑电信号进行短时傅里叶变换,并选取能量时频分布为特征,利用基于最大相关和最小冗余准则的方法进行特征选择,并使用极限学习机、支持向量机和反向传播算法对癫痫不同状态进行分类和判别。实验结果表明,极限学习机的分类准确率和训练速度两方面性能优于支持向量机和反向传播算法,发作间期和发作期的分类准确率达到98%以上,训练时间仅为0.8s,所提方法能够实时准确地检测癫痫发作。     

梯度直方图卷积特征的胶囊网络在交通监控下的车型分类

为了充分利用图像信息以提高现有交通监控下车型分类的效果,在胶囊网络的基础上增加梯度直方图卷积（HOG-C）特征提取方法,提出HOG-C特征的胶囊网络模型——HOG-C CapsNet。首先,使用梯度统计特征提取层对图像中的梯度信息进行统计,构建方向梯度直方图（HOG）特征图;其次,使用卷积层提取出图像的颜色信息,把提取出的颜色信息与HOG特征图融合构成HOG-C特征图;最后,输入卷积层提取HOG-C特征图的抽象特征,并通过胶囊网络对提取的抽象特征进行具有三维空间特征表达的胶囊封装,使用动态路由算法实现车型分类。在BIT-Vehicle数据集上对该模型和其他相关模型进行的对比实验中,该模型得到98.17%的准确率、97.98%的平均精确率均值（MAP）、98.42%的平均召回率均值（MAR）和98.20%的综合评价指标。实验结果表明,该模型在交通监控下的车型分类上具有更好的效果。     

梯度直方图卷积特征的胶囊网络在交通监控下的车型分类

为了充分利用图像信息以提高现有交通监控下车型分类的效果，在胶囊网络的基础上增加梯度直方图卷积（HOG-C）特征提取方法，提出HOG-C特征的胶囊网络模型——HOG-C CapsNet。首先，使用梯度统计特征提取层对图像中的梯度信息进行统计，构建方向梯度直方图（HOG）特征图；其次，使用卷积层提取出图像的颜色信息，把提取出的颜色信息与HOG特征图融合构成HOG-C特征图；最后，输入卷积层提取HOG-C特征图的抽象特征，并通过胶囊网络对提取的抽象特征进行具有三维空间特征表达的胶囊封装，使用动态路由算法实现车型分类。在BIT-Vehicle数据集上对该模型和其他相关模型进行的对比实验中，该模型得到98.17%的准确率、97.98%的平均精确率均值（MAP）、98.42%的平均召回率均值（MAR）和98.20%的综合评价指标。实验结果表明，该模型在交通监控下的车型分类上具有更好的效果。     

基于HOG和SVM的双眼虹膜图像的人眼定位算法

针对近红外光下现有的人眼定位算法普遍存在准确性不高、泛化能力不佳等问题，提出了一种基于方向梯度直方图（HOG）和支持向量机（SVM）相结合的双眼虹膜图像的人眼定位算法。利用HOG提取虹膜图像的人眼特征，并结合SVM分类器对HOG特征进行训练从而实现人眼的精确定位。为了减少漏检和误检，进一步提高定位准确率，又提出了多级级联SVM分类器算法；另外针对近红外光线下虹膜图像独特的灰度分布特点，设计了一种图像预处理方法，能够显著提高人眼定位速度。在MIR2016和CASIA-IRIS-Distance数据集上的实验结果表明，基于HOG和SVM的双眼虹膜图像的人眼定位算法具有高准确率、强泛化能力和高实时性。     

基于Gibbs抽样的轴承故障诊断方法

针对现有轴承故障诊断方法的不足,即诊断片面性问题,提出了一种基于Gibbs抽样的轴承故障诊断方法。首先对轴承振动信号进行局部特征尺度分解（LCD）得到内禀尺度分量（ISC）;然后对轴承振动信号和ISC分别提取时域特征,按照特征敏感度高低对时域特征排名,选择排名靠前的特征组成特征集;其次使用特征集训练产生基于Gibbs抽样的多维高斯分布模型;最后通过后验分析得到概率,实现轴承故障诊断。实验结果表明诊断正确率达到100%,与基于SVM的轴承诊断方法相比,在特征数为43个时诊断正确率提升了11.1个百分点。所提方法能够有效地对滚动轴承故障状态进行诊断,对高维复杂的轴承故障数据也有很好的诊断效果。     

基于关键帧特征库统计特征的双人交互行为识别

针对双人交互行为识别算法中普遍存在的算法计算复杂度高、识别准确性低的问题,提出一种新的基于关键帧特征库统计特征的双人交互行为识别方法。首先,对预处理后的交互视频分别提取全局GIST和分区域方向梯度直方图（HOG）特征。然后,采用k-means聚类算法对每类动作训练视频的所有帧的特征表示进行聚类,得到若干个近似描述同类动作视频的关键帧特征,构造出训练动作类别对应的关键帧特征库;同时,根据相似性度量统计出特征库中各个关键帧在交互视频中出现的频率,得到一个动作视频的统计直方图特征表示。最后,利用训练后的直方图相交核支持向量机（SVM）,对待识别视频采用决策级加权融合的方法得到交互行为的识别结果。在标准数据库测试的结果表明,该方法简单有效,对交互行为的正确识别率达到了85%。     

基于改进斑点鬣狗优化算法的同步优化特征选择

针对传统支持向量机（SVM）在封装式特征选择中分类精度低、特征子集选择冗余以及计算效率差的不足,利用元启发式优化算法同步优化SVM与特征选择。为改善SVM分类效果以及选择特征子集的能力,首先,利用自适应差分进化（DE）算法、混沌初始化与锦标赛选择策略对斑点鬣狗优化（SHO）算法改进,以增强其局部搜索能力并提高其寻优效率与求解精度;其次,将改进后的算法用于特征选择与SVM参数调整的同步优化中;最后,在UCI数据集进行特征选择仿真实验,采取分类准确率、选择特征数、适应度值及运行时间来综合评估所提算法的优化性能。实验结果证明,改进算法的同步优化机制能够在高分类准确率下降低特征选择的数目,该算法比传统算法更适合解决封装式特征选择问题,具有良好的应用价值。     

利用可分性指数的极化SAR图像特征选择与多层SVM分类

可分性指数（SI）可用来选择各类地物的有效分类特征,但在多维特征以及地物可分性较好的情况下,只利用可分性指数进行特征选择不能有效去除特征之间的冗余性。基于此,提出了利用可分性指数并辅以顺序后退（SBS）算法进行特征选择与多层支持向量机（SVM）分类的方法。首先,由各类地物在所有特征下的可分性指数选择分类地物和特征;然后,以该地物的分类精度为评估依据,利用顺序后退法筛选特征;其次,由剩余地物之间的可分性指数和顺序后退法依次选择各类地物的分类特征;最后利用多层SVM进行分类。实验结果表明,与只利用可分性指数选择特征进行多层SVM分类的方法相比,所提方法的分类精度提高了2%,各类地物的分类精度均高于86%,且运行时间为原来方法的一半。     

基于SURF和全局特征融合的图像分类研究

针对SURF对图像局部特征具有极好的描述能力,但对于全局特征描述能力不强的缺点,提出将SURF和全局颜色特征相融合的图像分类算法,提取图像的SURF特征向量集,并利用随机直方图算法将该向量集进行数据归约成单一高维特征向量;提取图像HSV颜色直方图;分别利用支持向量机（SVM）对这两种特征进行分类;将两个分类结果进行高层特征融合得到最终分类结果。实验结果表明,该算法显著提高了图像分类的准确率。     

基于CNN和SVM的疲劳驾驶闭眼特征实时检测

针对现有疲劳驾驶检测方法中实时性和泛化能力不足的问题, 本文提出了一种基于卷积神经网络(Convolutional Neural Networks, CNN)和支持向量机(Support Vector Machine, SVM)的疲劳驾驶闭眼特征检测方法, 使用CNN获取人脸相关特征点的位置并定位眼部感兴趣区域(Reg...     

低小慢无人机降落野外场景识别方法

针对低小慢无人机野外飞行场景复杂自主降落场景识别问题,提出了一种融合局部金字塔特征和卷积神经网络学习特征的野外场景识别算法。首先,将场景分为4×4和8×8块的小场景,使用方向梯度直方图（HOG）算法提取所有块的场景特征,所有特征首尾连接得到具有空间金字塔特性的特征向量。其次,设计一个针对场景分类的深度卷积神经网络,采用调优训练方法得到卷积神经网络模型,并提取深度网络学习特征。最后,连接两个特征得到最终场景特征,并使用支持向量机（SVM）分类器进行分类。所提算法在Sports-8、Scene-15、Indoor-67以及自建数据集上较传统手工特征方法的识别准确率提高了4个百分点以上。实验结果表明,所提算法能有效提升降落场景识别准确率。     

多特征融合的道路车辆检测方法

通过改进基于Haar-like特征和Adaboost的级联分类器,提出一种融合Haar-like特征和HOG特征的道路车辆检测方法。在传统级联分类器的Harr-like特征基础上引入HOG特征;为Haar-like特征和HOG特征分别设计不同形式的弱分类器,对每一个特征进行弱分类器的训练,用Gentle Adaboost算法代替Discrete Adaboost算法进行强分类器的训练;在级联分类器的最后几层上使用Adaboost算法挑选出来的特征组成特征向量训练SVM分类器。实验结果表明所提出的方法能有效检测道路车辆。