Proposed Classification Method Research Articles

Weed Classification for Site-Specific Weed Management Using an Automated Stereo Computer-Vision Machine-Learning System in Rice Fields

Site-specific weed management and selective application of herbicides as eco-friendly techniques are still challenging tasks to perform, especially for densely cultivated crops, such as rice. This study is aimed at developing a stereo vision system for distinguishing between rice plants and weeds and further discriminating two types of weeds in a rice field by using artificial neural networks (ANNs) and two metaheuristic algorithms. For this purpose, stereo videos were recorded across the rice field and different channels were extracted and decomposed into the constituent frames. Next, upon pre-processing and segmentation of the frames, green plants were extracted out of the background. For accurate discrimination of the rice and weeds, a total of 302 color, shape, and texture features were identified. Two metaheuristic algorithms, namely particle swarm optimization (PSO) and the bee algorithm (BA), were used to optimize the neural network for selecting the most effective features and classifying different types of weeds, respectively. Comparing the proposed classification method with the K-nearest neighbors (KNN) classifier, it was found that the proposed ANN-BA classifier reached accuracies of 88.74% and 87.96% for right and left channels, respectively, over the test set. Taking into account either the arithmetic or the geometric means as the basis, the accuracies were increased up to 92.02% and 90.7%, respectively, over the test set. On the other hand, the KNN suffered from more cases of misclassification, as compared to the proposed ANN-BA classifier, generating an overall accuracy of 76.62% and 85.59% for the classification of the right and left channel data, respectively, and 85.84% and 84.07% for the arithmetic and geometric mean values, respectively.

Rolling bearing faults severity classification using a combined approach based on multi-scales principal component analysis and fuzzy technique

Safety and fault diagnosis of rotating machinery play an important role in industrial systems. The reliability of the diagnosis performances is mainly linked to the signal processing tools used in the analysis phase. In this paper, a new method is proposed for the classification of rolling defects combining the optimized wavelet multi-resolution analysis (OWMRA), principal component analysis (PCA), and neuro-fuzzy. The OWMRA is performed to decompose the measured vibration signals in different frequency bands and extract additional information about the defect. The decomposition levels obtained from optimized WMRA are then used as the input of the PCA method. The extraction of individual feature sets including time domain features is generated to disclose health conditions of the bearing. Hence, the proposed classification method of neuro-fuzzy based on multi-scale principal component analysis (MSPCA) applies to real signals to analyze several types of defects on the bearings’ ball, outer race and inner race with variable fault diameter; load; and motor speed. The obtained results prove the reliability of the proposed diagnosis method to classify three types of bearing faults, and to give a better classification with greater efficiency compared to the application of individual classifiers or the artificial neural networks (ANN) alone. Finally, The effectiveness of our approach has been proven in terms of the successful classification rate compared on the one hand with two classification algorithms and two sets of features and on the other hand with the time domain method based on different choices of features.

Classification of Lung Nodules Based on Deep Residual Networks and Migration Learning.

The classification process of lung nodule detection in a traditional computer-aided detection (CAD) system is complex, and the classification result is heavily dependent on the performance of each step in lung nodule detection, causing low classification accuracy and high false positive rate. In order to alleviate these issues, a lung nodule classification method based on a deep residual network is proposed. Abandoning traditional image processing methods and taking the 50-layer ResNet network structure as the initial model, the deep residual network is constructed by combining residual learning and migration learning. The proposed approach is verified by conducting experiments on the lung computed tomography (CT) images from the publicly available LIDC-IDRI database. An average accuracy of 98.23% and a false positive rate of 1.65% are obtained based on the ten-fold cross-validation method. Compared with the conventional support vector machine (SVM)-based CAD system, the accuracy of our method improved by 9.96% and the false positive rate decreased by 6.95%, while the accuracy improved by 1.75% and 2.42%, respectively, and the false positive rate decreased by 2.07% and 2.22%, respectively, in contrast to the VGG19 model and InceptionV3 convolutional neural networks. The experimental results demonstrate the effectiveness of our proposed method in lung nodule classification for CT images.

Spatial-spectral classification of hyperspectral images based on multiple fractal-based features

Hyperspectral images are efficient tools for discriminating different types of earth's surface materials. Spectral features traditionally perform classification of hyperspectral images, but different studies have proved the efficiency of spatial features as complementary information in increasing the classification accuracy. The fractal geometry can be regarded as a potent tool for spatial data modeling. This study proposes a new classification method based on the integration of fractal and spectral features. For this purpose, three groups of fractal features, including mono-fractal, lacunarity and multi-fractal features are generated from the first few principal components of the hyperspectral image in different window sizes. These features are later stacked with spectral features and then fed to support vector machines classifier. The experiments are conducted on two real hyperspectral images Indian Pines, Pavia University. Final classification accuracy proved the superiority of the proposed classification method against the other competitive spatial-spectral methods.

Identifying and Classifying Aberrant Response Patterns Through Functional Data Analysis

We propose new methods for identifying and classifying aberrant response patterns (ARPs) by means of functional data analysis. These methods take the person response function (PRF) of an individual and compare it with the pattern that would correspond to a generic individual of the same ability according to the item-person response surface. ARPs correspond to atypical difference functions. The ARP classification is done with functional data clustering applied to the PRFs identified as ARP. We apply these methods to two sets of simulated data (the first is used to illustrate the ARP identification methods and the second demonstrates classification of the response patterns flagged as ARP) and a real data set (a Grade 12 science assessment test, SAT, with 32 items answered by 600 examinees). For comparative purposes, ARPs are also identified with three nonparametric person-fit indices (Ht, Modified Caution Index, and ZU3). Our results indicate that the ARP detection ability of one of our proposed methods is comparable to that of person-fit indices. Moreover, the proposed classification methods enable ARP associated with either spuriously low or spuriously high scores to be distinguished.

Molecular classification of breast cancer using the mRNA expression profiles of immune-related genes

Breast cancer is the most lethal cancer in women and displaying a broad range of heterogeneity in terms of clinical, molecular behavior and response to therapy. Increasing evidence demonstrated that immune-related genes were an important source of prognostic information for several types of tumors. In this study, the k-mean clustering was applied to gene expression data from the immune-related genes, two molecular clusters were identified for 1980 breast cancer patients. The prognostic significance of the immune-related genes based classification was confirmed in the log-rank test. These clusters were also associated with immune checkpoints, immune-related features and tumor infiltrating levels. In addition, we used the shrunken centroid algorithm to predict the cluster of a given breast cancer sample, and good predictive results were obtained by this algorithm. These results indicated that the proposed classification method is a promising method, and we hope that this method may improve the treatment stratification of breast cancer in the future.

Enhancing the feature representation of multi-modal MRI data by combining multi-view information for MCI classification

The classification of mild cognitive impairment (MCI), which is a early stage of Alzheimer’s disease and is associated with brain structural and functional changes, is still a challenging task. Recent studies have shown great promise for improving the performance of MCI classification by combining multiple structural and functional features, such as grey matter volume and clustering coefficient. However, extracting which features and how to combine multiple features to improve the performance of MCI classification have always been challenging problems. To address these problems, in this study we propose a new method to enhance the feature representation of multi-modal MRI data by combining multi-view information to improve the performance of MCI classification. Firstly, we extract two structural features (including grey matter volume and cortical thickness) and two functional features (including clustering coefficient and shortest path length) of each cortical brain region based on automated anatomical labeling (AAL) atlas from both T1w MRI and rs-fMRI data of each subject. Then, in order to obtain features that are more helpful in distinguishing MCI subjects, an improved multi-task feature selection method, namely MTFS-gLASSO-TTR, is proposed. Finally, a multi-kernel learning algorithm is adopted to combine multiple features to perform the MCI classification task. Our proposed MCI classification method is evaluated on 315 subjects (including 105 LMCI subjects, 105 EMCI subjects and 105 NCs) with both T1w MRI and rs-fMRI data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. Experimental results show that our proposed method achieves an accuracy of 88.5% and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.897 for LMCI/NC classification, an accuracy of 82.7% and an AUC of 0.832 for EMCI/NC classification, and an accuracy of 79.6% and an AUC of 0.803 for LMCI/EMCI classification, respectively. In addition, by comparison, the accuracy and AUC values of our proposed method are better than those of some existing state-of-the-art methods in MCI classification. Overall, our proposed MCI classification method is effective and promising for automatic diagnosis of MCI in clinical practice.

Brain MR Image Classification for Glioma Tumor detection using Deep Convolutional Neural Network Features.

Detection of brain tumor is a complicated task, which requires specialized skills and interpretation techniques. Accurate brain tumor classification and segmentation from MR images provide an essential choice for medical treatments. Different objects within an MR image have similar size, shape, and density, which makes the tumor classification and segmentation even more complex. Classification of the brain MR images into tumorous and non-tumorous using deep features and different classifiers to get higher accuracy. In this study, a novel four-step process is proposed; pre-processing for image enhancement and compression, feature extraction using convolutional neural networks (CNN), classification using the multilayer perceptron and finally, tumor segmentation using enhanced fuzzy cmeans method. The system is tested on 65 cases in four modalities consisting of 40,300 MR Images obtained from the BRATS-2015 dataset. These include images of 26 Low-Grade Glioma (LGG) tumor cases and 39 High-Grade Glioma (HGG) tumor cases. The proposed CNN feature-based classification technique outperforms the existing methods by achieving an average accuracy of 98.77% and a noticeable improvement in the segmentation results are measured. The proposed method for brain MR image classification to detect Glioma Tumor detection can be adopted as it gives better results with high accuracies.

Semi-supervised hyperspectral image classification algorithm based on graph embedding and discriminative spatial information

Image classification is one of the important techniques in computer vision. Due to the limited access of labeled samples in hyperspectral images, semi-supervised learning (SSL) methods have been widely applied in hyperspectral image classification. Graph based semi-supervised learning provides an effective solution to model data in classification problems, of which graph construction is the critical step. In this paper we employ the graphs constructed with a typical manifold learning method-locally linear embedding (LLE), based on which semi-supervised classification is then conducted. To exploit the valuable spatial information contained in hyperspectral images, discriminative spatial information (DSI) is then extracted. The proposed classification method is evaluated using three real hyperspectral data sets, revealing state-of-art performance when compared with different classification methods.

A Novel Classification Method for a Driver's Cognitive Stress Level by Transferring Interbeat Intervals of the ECG Signal to Pictures.

In this study, a novel classification method for a driver’s cognitive stress level was proposed, whereby the interbeat intervals extracted from an electrocardiogram (ECG) signal were transferred to pictures, and a convolution neural network (CNN) was used to train the pictures to classify a driver’s cognitive stress level. First, we defined three levels of tasks and collected the ECG signal of the driver at different cognitive stress levels by designing and performing a driving simulation experiment. We extracted the interbeat intervals and converted them to pictures according to the number of consecutive interbeat intervals in each picture. Second, the CNN model was used to train the data set to recognize the cognitive stress levels. Classification accuracies of 100%, 91.6% and 92.8% were obtained for the training set, validation set and test set, respectively, and were compared with those the BP neural network. Last, we discussed the influence of the number of interbeat intervals in each picture on the performance of the proposed classification method. The results showed that the performance initially improved with an increase in the number of interbeat intervals. A downward trend was observed when the number exceeded 40, and when the number was 40, the model performed best with the highest accuracy (98.79%) and a relatively low relative standard deviation (0.019).

Deep Spatial-Temporal Feature Fusion From Adaptive Dynamic Functional Connectivity for MCI Identification.

Dynamic functional connectivity (dFC) analysis using resting-state functional Magnetic Resonance Imaging (rs-fMRI) is currently an advanced technique for capturing the dynamic changes of neural activities in brain disease identification. Most existing dFC modeling methods extract dynamic interaction information by using the sliding window-based correlation, whose performance is very sensitive to window parameters. Because few studies can convincingly identify the optimal combination of window parameters, sliding window-based correlation may not be the optimal way to capture the temporal variability of brain activity. In this paper, we propose a novel adaptive dFC model, aided by a deep spatial-temporal feature fusion method, for mild cognitive impairment (MCI) identification. Specifically, we adopt an adaptive Ultra-weighted-lasso recursive least squares algorithm to estimate the adaptive dFC, which effectively alleviates the problem of parameter optimization. Then, we extract temporal and spatial features from the adaptive dFC. In order to generate coarser multi-domain representations for subsequent classification, the temporal and spatial features are further mapped into comprehensive fused features with a deep feature fusion method. Experimental results show that the classification accuracy of our proposed method is reached to 87.7%, which is at least 5.5% improvement than the state-of-the-art methods. These results elucidate the superiority of the proposed method for MCI classification, indicating its effectiveness in the early identification of brain abnormalities.

Three-dimensional quantification and classification approach for angularity and surface texture based on surface triangulation of reconstructed aggregates

Angularity and surface texture of aggregates are important morphological characteristics, which have significant effects on the micromechanical responses of asphalt composites. A novel approach to quantify and classify the complexity of aggregate angularity and surface texture is proposed from a statistical perspective in this study. The methodology consists of three main steps, as follows: 1) the aggregate is three-dimensional (3D) reconstructed, and the model surface is triangulated into facets to conduct clustering for aggregate angles evaluation; 2) consequently, a reference plane is determined for each facet cluster to quantify the surface texture of a surface area which overlaps the cluster, and thus the surface texture is quantified; and 3) aggregates are classified by using the distribution characteristics of the angularity and surface texture respectively. Based on the quantification with the presented approach, 275 grains were classified by the angularity and surface texture. Results indicate the benefit of the proposed method in accurate quantification and classification of aggregate angularity and surface texture, which facilitate the bridging of the gap between aggregate morphology and micromechanical performance of asphalt composites.

Research on the Key Technologies of Motor Imagery EEG Signal Based on Deep Learning

Brain-computer interface (BCI) is an emerging area of research that establishes a connection between the brain and external devices in a completely new way. It provides a new idea about the rehabilitation of brain diseases, human-computer interaction and augmented reality. One of the main problems of implementing BCI is to recognize and classify the motor imagery Electroencephalography(EEG) signals effectively. This paper takes the characteristic data of motor imagery of EEG as the research object to conduct the research of multi-classification method. In this study, we use the Emotiv helmet with 16 biomedical sensors to obtain EEG signal, adopt the fast independent component analysis and the fast Fourier transform to realize signal preprocessing and select the common spatial pattern algorithm to extract the features of the motor imagery EEG signal. In order to improve the accuracy of recognition of EEG signal, a new deep learning network is designed for multi-channel self-acquired EEG data set which is named as min-VGG-LSTMnet in this paper. This network combines Long Short-Term Memory Network with convolutional neural network VGG and achieves the four-classification task of the left-hand, right-hand, left-foot and right-foot lifting movements based on motor imagery. The results show that the accuracy of the proposed classification method is at least 8.18% higher than other mainstream deep-learning methods.

Research on the Key Technologies of Motor Imagery EEG Signal Based on Deep Learning

Brain-computer interface (BCI) is an emerging area of research that establishes a connection between the brain and external devices in a completely new way. It provides a new idea about the rehabilitation of brain diseases, human-computer interaction and augmented reality. One of the main problems of implementing BCI is to recognize and classify the motor imagery Electroencephalography(EEG) signals effectively. This paper takes the characteristic data of motor imagery of EEG as the research object to conduct the research of multi-classification method. In this study, we use the Emotiv helmet with 16 biomedical sensors to obtain EEG signal, adopt the fast independent component analysis and the fast Fourier transform to realize signal preprocessing and select the common spatial pattern algorithm to extract the features of the motor imagery EEG signal. In order to improve the accuracy of recognition of EEG signal, a new deep learning network is designed for multi-channel self-acquired EEG data set which is named as min-VGG-LSTMnet in this paper. This network combines Long Short-Term Memory Network with convolutional neural network VGG and achieves the four-classification task of the left-hand, right-hand, left-foot and right-foot lifting movements based on motor imagery. The results show that the accuracy of the proposed classification method is at least 8.18% higher than other mainstream deep-learning methods.

Intelligent object recognition in underwater images using evolutionary-based Gaussian mixture model and shape matching

Object recognition in underwater images becomes a challenging task because of its poor visibility conditions. Marine scientists often prefer automation tools for object recognition as large amount of data is captured everyday with the help of autonomous underwater vehicles. The challenge for classification in such underwater images is the limited color information. An attempt is made to recognize objects in underwater images using an adaptive Gaussian mixture model. The Gaussian mixture model performs accurate object segmentation provided the number of clusters is predefined. Optimization techniques like genetic algorithm, particle swarm optimization and differential evolution were analyzed for initializing the parameter set. Differential evolution is known for its accurate decision making in fewer iterations and proved to be better for initializing the number of clusters for the Gaussian mixture model. Further for object recognition, inner distance shape matching technique was applied. The proposed classification method achieved a maximum accuracy of 99%.

Object-Oriented Mangrove Species Classification Using Hyperspectral Data and 3-D Siamese Residual Network

Mangrove species classification is of particular importance for coastal conservation and restoration. However, it is challenging to distinguish species-level differences with limited training data. In this letter, we propose an object-oriented classification method for mangrove forests by using the hyperspectral image (HSI) and the 3-D Siamese residual network. First, superpixel segmentation is utilized to obtain objects with various shapes and scales. Second, 3-D patches of each object are extracted from the original HSI, and those patches containing training samples are adopted to pairwise train the network. The 3-D spatial pyramid pooling (3-D-SPP) is added in the network to extract features in multiple scales. Finally, the abstract features of test samples are learned by the trained network, and the labels are determined by the nearest neighbor classifier within the metric space. Experiments on real mangrove hyperspectral data demonstrate the effectiveness of the proposed method in species classification of mangroves.

Performance Improvement of Feature-Based Fault Classification for Rotor System

For the management of rotating machines, machine learning (ML) has been researched with the use of feature parameters that have physical and statistical meanings of vibration signals. Genetic algorithm (GA) and principal component analysis (PCA) are the algorithms used for the selection or extraction process of the features; equipment condition. This study proposes a new method to maximize the advantages of the extraction and selection algorithms, thereby improving the fault classification performance. The proposed method is estimated in a variety of equipment conditions by selecting and extracting the effective features for status classification. To evaluate the performance of the fault classification through feature selection and extraction of the ML, a comparative analysis with the proposed method and the original method is also performed. With Lab-scale gearbox, several types of fault tests are conducted, and seven different fault types of equipment conditions, including the normal status, are simulated. The results of the experiments show that, the performance of classification of GA for feature selection is 85%, while PCA for feature extraction is 53%. The performance result of the proposed method for fault classification is 95%, meaning that the performance of fault diagnosis is more efficient in terms of discriminative learning than the original method. Therefore, the proposed method with feature extraction and selection algorithm can improve the fault classification performance by 10% and more for fault diagnosis through ML.

Competition between high-speed trains and air travel in China: From a spatial to spatiotemporal perspective

Abstract Scheduling strategies are important to improve network efficiency and optimize time slots for both high-speed train (HST) and air travel. Although many scholars have studied HST and air travel competition from the perspectives of the spatial patterns and travel costs of the two transportation modes, few studies have investigated their potential temporal competition. To fill this research gap, this paper uses a time window method to examine the relationship between HSTs and flights operating between city-pairs. From the perspective of departure time, an analysis was conducted in the overlapping network of HSTs and flights at the city-pair level in China, based on one-day schedules. The results show that (1) HSTs and flights were temporally scheduled differently according to their within-day dynamics, and (2) HSTs and flights experienced fierce competition from 7 am to 9 am on average. Furthermore, the proposed classification method identifies city-pairs in a single-mode dominant market, full competition market, partial competition market, and complementary coordination market. Spatially, 24 city-pairs operating 343 flights with 533 train competitors in one day formed the full competition market with high-frequency schedules, including such city-pairs as Beijing-Shanghai, Guangzhou-Shanghai, and Shanghai-Xi’an. The method to classify city-pairs according to the similarity and diversity of HST and flight schedules may be generalized for allocating time slots for two modes. This work may help to schedule HSTs and flights in future transportation management.

A parallel classification framework for protein fold recognition

Proteins’ tertiary structure, which is determined by its amino acid sequence via the protein folding process, have essential role in the function of protein. Protein fold recognition is one of the interesting studies in bioinformatics. In this paper, to address this issue, we propose a Feature Selection (FS) method based on Map_Reduce framework and Vortex Search Algorithm (VSA). FS is one of the most important steps of pre-processing data, which aims to select a variable subset of relevant features. In unparalleled mode and typical data, over hundreds of feature selection and dimension reduction algorithms have been provided such as Principle Component Analysis, Linear Discriminant Analysis, and so on. Nevertheless, these algorithms are not implemented for real-world applications when data instances increasing in three-dimensional: volume, velocity and variety that called Big Data, actually if we want to use previous feature selection methods on Big Data, volume of large and complex computing will be required. VSA was inspired from the vortex pattern created by the vortical flow of the stirred fluids. In Map_Reduce framework, Map and Reduce functions executed in parallel mode. In the proposed method, in each step of Map function, a VSA is employed to find an optimized subset of features and decrease feature search space. In the light of the above consideration, we evaluate the proposed method in classification of a benchmark dataset for protein fold recognition. The experimental results indicate that the proposed method improves prediction accuracy considerably.

A computer-aided diagnosis system for differentiation and delineation of malignant regions on whole-slide prostate histopathology image using spatial statistics and multidimensional DenseNet.

Prostate cancer (PCa) is a major health concern in aging males, and proper management of the disease depends on accurately interpreting pathology specimens. However, reading prostatectomy histopathology slides, which is basically for staging, is usually time consuming and differs from reading small biopsy specimens, which is mainly used for diagnosis. Generally, each prostatectomy specimen generates tens of large tissue sections and for each section, the malignant region needs to be delineated to assess the amount of tumor and its burden. With the aim of reducing the workload of pathologists, in this study, we focus on developing a computer-aided diagnosis (CAD) system based on a densely connected convolutional neural network (DenseNet) for whole-slide histopathology images to outline the malignant regions. We use an efficient color normalization process based on ranklet transformation to automatically correct the intensity of the images. Additionally, we use spatial probability to segment the tissue structure regions for different tissue recognition patterns. Based on the segmentation, we incorporate a multidimensional structure into DenseNet to determine if a particular prostatic region is benign or malignant. As demonstrated by the experimental results with a test set of 2,663 images from 32 whole-slide prostate histopathology images, our proposed system achieved 0.726, 0.6306, and 0.5209 in the average of the Dice coefficient, Jaccard similarity coefficient, and Boundary F1 score measures, respectively. Then, the accuracy, sensitivity, specificity, and the area under the ROC curve (AUC) of the proposed classification method were observed to be 95.0% (2544/2663), 96.7% (1210/1251), 93.9% (1334/1412), and 0.9831, respectively. We provide a detailed discussion on how our proposed system demonstrates considerable improvement compared with similar methods considered in previous researches as well as how it can be used for delineating malignant regions.