Hierarchical Classification Approach Research Articles

A hierarchical machine learning classification approach for secondary task identification from observed driving behavior data

According to NHTSA, more than 3477 people (including 551 non-occupants) were killed and 391,000 were injured due to distraction-related crashes in 2015. The distracted driving epidemic has long been under research to identify its impact on driving behavior. There have been a few attempts to detect drivers’ engagement in secondary tasks from observed driving behavior. Yet, to the authors’ knowledge, not much effort has been directed to identify the types of secondary tasks from driving behavior parameters. This study proposes a bi-level hierarchical classification methodology using machine learning to identify the different types of secondary tasks drivers are engaged in using their driving behavior parameters. At the first level, drivers’ engagement in secondary tasks is detected, while at the second level, the distinct types of secondary tasks are identified. Comparative evaluation is performed between nine ensemble tree classification methods to identify three types of secondary tasks (hand-held cellphone calling, cellphone texting, and interaction with an adjacent passenger). The inputs to the models are five driving behavior parameters (speed, longitudinal acceleration, lateral acceleration, pedal position, and yaw rate) along with their standard deviations. The results showed that the overall secondary task detection accuracy ranged from 66% to 96%, except for the Decision Tree that was able to detect engagement in secondary tasks with a high accuracy of 99.8%. For the identification of secondary tasks types, the overall accuracy ranged from 55% to 79%, with the highest accuracy of 82.2% achieved by the Random Forest method. The findings of the paper show the proposed methodology promising to (1) characterize drivers’ engagement in unlawful secondary tasks (such as texting) as a counter measure to prevent crashes, and (2) alert drivers to pay attention back to the main driving task when risky changes to their driving behavior take place.

Automatic physical activity and in‐vehicle status classification based on GPS and accelerometer data: A hierarchical classification approach using machine learning techniques

AbstractDue to the advancement of tracking technology, a large quantity of movement data has been collected and analyzed in various research domains. In human mobility and physical activity (PA) research, GPS trajectories and the capabilities of geographic information systems (GIS) facilitate a better understanding of the associations between PA and various environmental factors taking individuals’ daily travels into account. PA research, however, needs to widen its focus from the intensity of PA to types of PA, which may provide useful clues for understanding specific health behaviors in particular geographic contexts. This study proposes and develops an algorithm to automatically classify PA types and in‐vehicle status using GPS and accelerometer data. Walking, standing, jogging, biking and sedentary/in‐vehicle statuses are identified through hierarchical classification processes based on machine learning and geospatial techniques. The proposed algorithm achieved high predictive accuracy on real‐world GPS and accelerometer data. It can greatly reduce participants’ and researchers’ burdens by automatically identifying PA types and in‐vehicle status for human mobility research, which is also known as travel mode imputation in transportation research.

A framework for retinal vessel segmentation from fundus images using hybrid feature set and hierarchical classification

Retinal blood vessels play an imperative role in detection of many ailments, such as cardiovascular diseases, hypertension, and diabetic retinopathy. The automated way of segmenting vessels from retinal images can help in early detection of many diseases. In this paper, we propose a framework based on hybrid feature set and hierarchical classification approach to segment blood vessels from digital retinal images. Firstly, we apply bidirectional histogram equalization on the inverted green channel to enhance the fundus image. Six discriminative feature extraction methods have been employed comprising of local intensities, local binary patterns, histogram of gradients, divergence of vector field, high-order local autocorrelations, and morphological transformation. The selection of feature sets has been carried out by classifying vessel and background pixels using random forests and evaluating the segmentation performance for each category of features. The selected feature sets are then used in conjunction with our proposed hierarchical classification approach to segment the vessels. The proposed framework has been tested on the DRIVE, STARE, and CHASEDB1 which are the benchmark datasets for retinal vessel segmentation methods. The results obtained from the experimental analysis show that the proposed framework can achieve better results than most state-of-the-art methods.

Development of an automated method for the identification of defective hazelnuts based on RGB image analysis and colourgrams

Over the past decades, Red-Green-Blue (RGB) image analysis has gained increasing importance in industrial applications, since it has widely proved to be a suitable tool for food quality and process control.This article describes the development of a fast and objective method for the automated identification of defective hazelnut kernels based on multivariate analysis of RGB images.To this aim, an overall sample set of 2000 half-cut hazelnut kernels, previously assigned by industrial expert assessors as sound or defective (i.e. rotten or pest-affected), was collected and imaged using a digital camera. The colour-related information of the images was converted into one-dimensional signals, named colourgrams, which were firstly explored through the Principal Component Analysis and subsequently used to build classification models, based on both Partial Least Square-Discriminant Analysis (PLS-DA) and interval-PLS-DA (iPLS-DA) algorithms.A tree-structure hierarchical classification approach has been considered, i.e. the discrimination between sound and defective kernels as a first rule, and the discrimination between the two types of defect as a second rule. The best sound vs defective classification model was able to correctly recognize approximately the 97% of the test set defective samples, while the best rotten vs pest-affected model allowed classifying correctly more than 92% of the test set samples.Moreover, the image reconstruction performed using the selected colourgram features led to an exhaustive interpretation of the decision-making criteria adopted by the classification algorithms and further confirmed the reliability of the proposed method.

Improving large-scale hierarchical classification by rewiring: a data-driven filter based approach

Hierarchical Classification (HC) is a supervised learning problem where unlabeled instances are classified into a taxonomy of classes. Several methods that utilize the hierarchical structure have been developed to improve the HC performance. However, in most cases apriori defined hierarchical structure by domain experts is inconsistent; as a consequence performance improvement is not noticeable in comparison to flat classification methods. We propose a scalable data-driven filter based rewiring approach to modify an expert-defined hierarchy. Experimental comparisons of top-down hierarchical classification with our modified hierarchy, on a wide range of datasets shows classification performance improvement over the baseline hierarchy (i.e., defined by expert), clustered hierarchy and flattening based hierarchy modification approaches. In comparison to existing rewiring approaches, our developed method (rewHier) is computationally efficient, enabling it to scale to datasets with large numbers of classes, instances and features. We also show that our modified hierarchy leads to improved classification performance for classes with few training samples in comparison to flat and state-of-the-art hierarchical classification approaches. Source Code: https://cs.gmu.edu/~mlbio/TaxMod/

Multiscale and Hierarchical Classification for Benthic Habitat Mapping

Developing quantitative and objective approaches to integrate multibeam echosounder (MBES) data with ground observations for predictive modelling is essential for ensuring repeatability and providing confidence measures for benthic habitat mapping. The scale of predictors within predictive models directly influences habitat distribution maps, therefore matching the scale of predictors to the scale of environmental drivers is key to improving model accuracy. This study uses a multi-scalar and hierarchical classification approach to improve the accuracy of benthic habitat maps. We used a 700-km2 region surrounding Cape Otway in Southeast Australia with full MBES data coverage to conduct this study. Additionally, over 180 linear kilometers of towed video data collected in this area were classified using a hierarchical classification approach. Using a machine learning approach, Random Forests, we combined MBES bathymetry, backscatter, towed video and wave exposure to model the distribution of biotic classes at three hierarchical levels. Confusion matrix results indicated that greater numbers of classes within the hierarchy led to lower model accuracy. Broader scale predictors were generally favored across all three hierarchical levels. This study demonstrates the benefits of testing predictor scales across multiple hierarchies for benthic habitat characterization.

A hierarchical classification approach for recognition of low-density (LDPE) and high-density polyethylene (HDPE) in mixed plastic waste based on short-wave infrared (SWIR) hyperspectral imaging

The aim of this work was to recognize different polymer flakes from mixed plastic waste through an innovative hierarchical classification strategy based on hyperspectral imaging, with particular reference to low density polyethylene (LDPE) and high-density polyethylene (HDPE). A plastic waste composition assessment, including also LDPE and HDPE identification, may help to define optimal recycling strategies for product quality control. Correct handling of plastic waste is essential for its further “sustainable” recovery, maximizing the sorting performance in particular for plastics with similar characteristics as LDPE and HDPE. Five different plastic waste samples were chosen for the investigation: polypropylene (PP), LDPE, HDPE, polystyrene (PS) and polyvinyl chloride (PVC). A calibration dataset was realized utilizing the corresponding virgin polymers. Hyperspectral imaging in the short-wave infrared range (1000–2500nm) was thus applied to evaluate the different plastic spectral attributes finalized to perform their recognition/classification. After exploring polymer spectral differences by principal component analysis (PCA), a hierarchical partial least squares discriminant analysis (PLS-DA) model was built allowing the five different polymers to be recognized. The proposed methodology, based on hierarchical classification, is very powerful and fast, allowing to recognize the five different polymers in a single step.

Spatio-temporal patterns of land use and cropping frequency in a tropical catchment of South India

India's rapid population and economic growth leads to fast changing land use and management practices that have a major impact on the environment. Therefore, this study assesses spatio-temporal dynamics of land use and cropping frequencies using moderate resolution spaceborne data (Landsat 7 and LISS III). Based on a hierarchical knowledge-based classification approach, multi-temporal satellite data from the years 2000/2001 and 2010/2011 have been used to derive land use and cropping frequency maps. The approach adopted in this study resulted in a satisfactory classification quality as indicated by overall accuracies >90% for the individual classifications. A reduced land use pressure on mountainous areas was found, indicated by an increasing development of forests within the transition zones between cultivated land and steep slopes. Furthermore, an increase of tree plantations points to a shift from drought vulnerable plants to less risk prone perennial plants. We found a higher cropping frequency in 2010/11 related to both inter-annual precipitation differences over the course of the rainy season and long-term socio-economic changes. While low yield areas are left for natural succession or switched to tree plantations, the cultivation of high yield areas was intensified.

Hierarchical Fuzzy Primitive Surface Classification From Tessellated Solids for Defining Part-to-Part Removal Directions

This paper is arranged in three main sections: the first section is a hierarchical method based on clustering and a fuzzy membership system where the tessellated three-dimensional (3D) models are classified into their containing primitives: cylinder, cone, sphere, and flat. In the second section, automated assembly planning (AAP) is considered as the main application of our novel hierarchical primitive classification approach. The classified primitives obtained from the first section are used to define the removal directions between mating parts in an assembly model. Finally, a fuzzification method is used to express the uncertainty of the detected connections between every pair of parts. The acquired uncertainties are used in a user interaction process to approve, deny, or modify the connections with higher uncertainties.

Design of continuous EMG classification approaches towards the control of a robotic exoskeleton in reaching movements.

Myoelectric control of rehabilitation devices engages active recruitment of muscles for motor task accomplishment, which has been proven to be essential in motor rehabilitation. Unfortunately, most electromyographic (EMG) activity-based controls are limited to one single degree-of-freedom (DoF), not permitting multi-joint functional tasks. On the other hand, discrete EMG-triggered approaches fail to provide continuous feedback about muscle recruitment during movement. For such purposes, myoelectric interfaces for continuous recognition of functional movements are necessary. Here we recorded EMG activity using 5 bipolar electrodes placed on the upper-arm in 8 healthy participants while they performed reaching movements in 8 different directions. A pseudo on-line system was developed to continuously predict movement intention and attempted arm direction. We evaluated two hierarchical classification approaches. Movement intention detection triggered different movement direction classifiers (4 or 8 classes) that were trained and tested over a 5-fold cross validation. We also investigated the effect of 3 different window lengths to extract EMG features on classification. We obtained classification accuracies above 70% for both hierarchical approaches. These results highlight the viability of classifying online 8 upper-arm different directions using surface EMG activity of 5 muscles and represent a first step towards an online EMG-based control for rehabilitation devices.

A Directed Acyclic Graph (DAG) Ensemble Classification Model

In this paper, a hierarchical ensemble classification approach that utilizes a Directed Acyclic Graph (DAG) structure is proposed as a solution to the multi-class classification problem. Two main DAG structures are considered: (i) rooted DAG, and (ii) non-rooted DAG. The main challenges that are considered in this paper are: (i) the successive misclassification issue associated with hierarchical classification, and (i) identification of the starting node within the non-rooted DAG approach. To address these issues the idea is to utilize Bayesian probability values to: select the best starting DAG node, and to dictate whether single or multiple paths should be followed within the DAG structure. The reported experimental results indicated that the proposed DAG structure is more effective than when using a simple binary tree structure for generating a hierarchical classification model.

Random forest wetland classification using ALOS-2 L-band, RADARSAT-2 C-band, and TerraSAR-X imagery

Wetlands are important ecosystems around the world, although they are degraded due both to anthropogenic and natural process. Newfoundland is among the richest Canadian province in terms of different wetland classes. Herbaceous wetlands cover extensive areas of the Avalon Peninsula, which are the habitat of a number of animal and plant species. In this study, a novel hierarchical object-based Random Forest (RF) classification approach is proposed for discriminating between different wetland classes in a sub-region located in the north eastern portion of the Avalon Peninsula. Particularly, multi-polarization and multi-frequency SAR data, including X-band TerraSAR-X single polarized (HH), L-band ALOS-2 dual polarized (HH/HV), and C-band RADARSAT-2 fully polarized images, were applied in different classification levels. First, a SAR backscatter analysis of different land cover types was performed by training data and used in Level-I classification to separate water from non-water classes. This was followed by Level-II classification, wherein the water class was further divided into shallow- and deep-water classes, and the non-water class was partitioned into herbaceous and non-herbaceous classes. In Level-III classification, the herbaceous class was further divided into bog, fen, and marsh classes, while the non-herbaceous class was subsequently partitioned into urban, upland, and swamp classes. In Level-II and -III classifications, different polarimetric decomposition approaches, including Cloude-Pottier, Freeman-Durden, Yamaguchi decompositions, and Kennaugh matrix elements were extracted to aid the RF classifier. The overall accuracy and kappa coefficient were determined in each classification level for evaluating the classification results. The importance of input features was also determined using the variable importance obtained by RF. It was found that the Kennaugh matrix elements, Yamaguchi, and Freeman-Durden decompositions were the most important parameters for wetland classification in this study. Using this new hierarchical RF classification approach, an overall accuracy of up to 94% was obtained for classifying different land cover types in the study area.

Hierarchical Terrain Classification Based on Multilayer Bayesian Network and Conditional Random Field

This paper presents a hierarchical classification approach for Synthetic Aperture Radar (SAR) images. The Conditional Random Field (CRF) and Bayesian Network (BN) are employed to incorporate prior knowledge into this approach for facilitating SAR image classification. (1) A multilayer region pyramid is constructed based on multiscale oversegmentation, and then, CRF is used to model the spatial relationships among those extracted regions within each layer of the region pyramid; the boundary prior knowledge is exploited and integrated into the CRF model as a strengthened constraint to improve classification performance near the boundaries. (2) Multilayer BN is applied to establish the causal connections between adjacent layers of the constructed region pyramid, where the classification probabilities of those sub-regions in the lower layer, conditioned on their parents’ regions in the upper layer, are used as adjacent links. More contextual information is taken into account in this framework, which is a benefit to the performance improvement. Several experiments are conducted on real ESAR and TerraSAR data, and the results show that the proposed method achieves better classification accuracy.

Hierarchical classification approach for mapping rubber tree growth using per-pixel and object-oriented classifiers with SPOT-5 imagery

There has been growing interest in Malaysia to increase the productivity of latex. This made accurate knowledge of rubber tree growth and age distribution a helpful decision making tool for the government, rubber plantation managers, and harvesters. Gathering this information using conventional methods is difficult, time consuming, and limited in spatial coverage. This paper presents hierarchical classification approach to obtain accurate map of rubber tree growth age distribution using SPOT-5 satellite imagery. The objective of the study is to evaluate the performance of pixel-based and object-oriented classifiers for rubber growth classification. At the first level, the general land cover was classified into eight land cover classes (soil, water body, rubber, mature oil palm, young oil palm, forest, urban area, and other vegetation) using Mahalanobis distance (MD), k-nearest neighbor (k-NN), and Support Vector Machine (SVM) classifiers. Thereafter, the best classification map, k-NN output, was used to select only pixels that belong to the rubber class from the SPOT-5 image. The extracted pixels served as input into the next classification hierarchy where four classifiers, MD, k-NN, SVM, and decision tree (DT), were implemented to map rubber trees into three intra-classes (mature, middle-aged, and young rubbers). The result produced overall accuracy of 97.48%, 96.90%, 96.25%, and 80.80% for k-NN, SVM, MD, and DT respectively. The result indicates that object-oriented classifiers are better than pixel-based methods mapping rubber tree growth.

Improving Performance of Multiclass Classification by Inducing Class Hierarchies

In the last decades, one issue that has received a lot of attention in classification problems is how to obtain better classifications. This problem becomes even more complicated when the number of classes is high. In this multiclass scenario, it is assumed that the class labels are independent of each other, and thus, most techniques and methods proposed to improve the performance of the classifiers rely on it. An alternative way to address the multiclass problem is to hierarchically distribute the classes in a collection of multiclass subproblems by reducing the number of classes involved in each local subproblem. In this paper, we propose a new method for inducing a class hierarchy from the confusion matrix of a multiclass classifier. Then, we use the class hierarchy to learn a tree-like hierarchy of classifiers for solving the original multiclass problem in a similar way as the top-down hierarchical classification approach does for working with hierarchical domains. We experimentally evaluate the proposal on a collection of multiclass datasets showing that, in general, the generated hierarchies not only outperforms the original (flat) classification but also hierarchical approaches based on other alternative ways of constructing the class hierarchy.

Evaluating Functional Annotations of Enzymes Using the Gene Ontology.

The Gene Ontology (GO) (Ashburner et al., Nat Genet 25(1):25–29, 2000) is a powerful tool in the informatics arsenal of methods for evaluating annotations in a protein dataset. From identifying the nearest well annotated homologue of a protein of interest to predicting where misannotation has occurred to knowing how confident you can be in the annotations assigned to those proteins is critical. In this chapter we explore what makes an enzyme unique and how we can use GO to infer aspects of protein function based on sequence similarity. These can range from identification of misannotation or other errors in a predicted function to accurate function prediction for an enzyme of entirely unknown function. Although GO annotation applies to any gene products, we focus here a describing our approach for hierarchical classification of enzymes in the Structure-Function Linkage Database (SFLD) (Akiva et al., Nucleic Acids Res 42(Database issue):D521–530, 2014) as a guide for informed utilisation of annotation transfer based on GO terms.

Can computer vision problems benefit from structured hierarchical classification?

Research in the field of supervised classification has mostly focused on the standard, so-called classification approach, where the problem classes live in a trivial, one-level semantic space. There is however an increasing interest in the hierarchical classification approach, where a performance gain is expected by incorporating prior taxonomic knowledge about the classes into the learning process. Intuitively, the hierarchical approach should be beneficial in general for the classification of visual content, as suggested by the fact that humans seem to organize objects into hierarchies based on visually perceived similarities. In this paper, we provide an analysis that aims to determine the conditions under which the hierarchical approach can consistently give better performances than the flat approach for the classification of visual content. In particular, we (1) show how hierarchical methods can fail to outperform flat methods when applied to real vision-based classification problems, and (2) investigate the underlying reasons for the lack of improvement, by applying the same methods to synthetic datasets in a simulation. Our conclusion is that the use of high-level hierarchical feature representations is crucial for obtaining a performance gain with the hierarchical approach, and that poorly chosen prior taxonomies hinder this gain even though proper high-level features are used.

Video-Based Classification of Driving Behavior Using a Hierarchical Classification System with Multiple Features

Driver fatigue and inattention have long been recognized as one of the main contributing factors in traffic accidents. Therefore, the development of intelligent driver assistance systems, which provides automatic monitoring of driver's vigilance, is an urgent and challenging task. This paper presents a novel system for video-based driving behavior recognition. The fundamental idea is to monitor driver's hand movements and to use these as predictors for safe/unsafe driving behavior. In comparison to previous work, the proposed method utilizes hierarchical classification and treats driving behavior in terms of a spatio-temporal reference framework as opposed to a static image. The approach was verified using the Southeast University Driving-Posture Dataset, a dataset comprised of video clips covering aspects of driving such as: normal driving, responding to a cell phone call, eating and smoking. After pre-processing for illumination variations and motion sequence segmentation, eight classes of behavior were identified. The overall prediction accuracy obtained using the proposed approach was [Formula: see text] when using a hierarchical classification approach. The proposed approach was able to clearly identify two dangerous driving behaviors, Responding to a cellphone call and Eating, with recognition rates of 92.39% and 92.29% respectively.

A hierarchical knowledge-based classification for glacier terrain mapping: a case study from Kolahoi Glacier, Kashmir Himalaya

AbstractA glacierized terrain comprises different land covers, and their mapping using satellite data is challenged by their spectral similarity. We propose a hierarchical knowledge-based classification (HKBC) approach for differentiation of glacier terrain classes and mapping of glacier boundaries, using Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) imagery and Global Digital Elevation Model (GDEM). The methodology was tested over Kolahoi Glacier, Kashmir Himalaya. For the sequential extraction of various glacier terrain classes, several input layers were generated from the primary datasets by applying image-processing techniques. Noticeable differences in temperature and spectral response between supraglacial debris and periglacial debris facilitated the development of a thermal glacier mask and normalized-difference debris index, which together with slope enabled their differentiation. These and the other layers were then used in several discrete tests in HKBC, to map various glacier terrain classes. An ASTER visible near-infrared image and 42 field points were used to validate results. The proposed approach satisfactorily classified all the glacier terrain classes with an overall accuracy of 89%. The Z-test reveals that results obtained from HKBC are significantly (at 95% confidence level) better than those from a maximum likelihood classifier (MLC). Glacier boundaries obtained from HKBC were found to be plausibly better than those obtained from MLC and visual interpretation.

Hierarchical Discrimination of Bamboo Species by near Infrared Spectroscopy

A hierarchical classification approach based on the K nearest neighbour algorithm is proposed for effective discrimination of bamboo species. The use of generalised least squares weighting pre-processing proved essential to achieve optimum discriminant accuracy and allowed near infrared spectra of both leaves and stem to be merged and used together. This work can have significant implications for the conservation of Giant Pandas.