Object Classification Method Research Articles

Point-cloud based 3D object detection and classification methods for self-driving applications: A survey and taxonomy

Autonomous vehicles are becoming central for the future of mobility, supported by advances in deep learning techniques. The performance of aself-driving system is highly dependent on the quality of the perception task. Developments in sensor technologies have led to an increased availability of 3D scanners such as LiDAR, allowing for a more accurate representation of the vehicle's surroundings, leading to safer systems. The rapid development and consequent rise of research studies around self-driving systems since early 2010, resulted in a tremendous increase in the number and novelty of object detection methods. After the first wave of works that essentially tried to expand known techniques from object detection in images, more recently there has been a notable development in newer and more adapted to LiDAR data works. This paper addresses the existing literature on object detection using LiDAR data within the scope of self-driving and brings a systematic way for analysing it. Unlike general object detection surveys, we will focus on point-cloud data, which presents specific challenges, notably its high-dimensional and sparse nature. This work introduces a common object detection pipeline and taxonomy to facilitate a thorough comparison between different techniques and, departing from it, this work will critically examine the representation of data (critical for complexity reduction), feature extraction and finally the object detection models. A comparison between performance results of the different models is included, alongside with some future research challenges.

Apathy Classification Based on Doppler Radar Image for the Elderly Person.

Apathy is a disease characterized by diminished motivation not attributable to a diminished level of consciousness, cognitive impairment, or emotional distress. It is a serious problem facing the elderly in today's society. The diagnosis of apathy needs to be done at a clinic, which is particularly inconvenient and difficult for elderly patients. In this work, we examine the possibility of using doppler radar imaging for the classification of apathy in the elderly. We recruited 178 elderly participants to help create a dataset by having them fill out a questionnaire and submit to doppler radar imaging while performing a walking action. We selected walking because it is one of the most common actions in daily life and potentially contains a variety of useful health information. We used radar imaging rather than an RGB camera due to the greater privacy protection it affords. Seven machine learning models, including our proposed one, which uses a neural network, were applied to apathy classification using the walking doppler radar images of the elderly. Before classification, we perform a simple image pre-processing for feature extraction. This pre-processing separates every walking doppler radar image into four parts on the vertical and horizontal axes and the number of feature points is then counted in every separated part after binarization to create eight features. In this binarization, the optimized threshold is obtained by experimentally sliding the threshold. We found that our proposed neural network achieved an accuracy of more than 75% in apathy classification. This accuracy is not as high as that of other object classification methods in current use, but as an initial research in this area, it demonstrates the potential of apathy classification using doppler radar images for the elderly. We will examine ways of increasing the accuracy in future work.

Underwater-Sonar-Image-Based 3D Point Cloud Reconstruction for High Data Utilization and Object Classification Using a Neural Network

This paper proposes a sonar-based underwater object classification method for autonomous underwater vehicles (AUVs) by reconstructing an object’s three-dimensional (3D) geometry. The point cloud of underwater objects can be generated from sonar images captured while the AUV passes over the object. Then, a neural network can predict the class given the generated point cloud. By reconstructing the 3D shape of the object, the proposed method can classify the object accurately through a straightforward training process. We verified the proposed method by performing simulations and field experiments.

Zero shot objects classification method of side scan sonar image based on synthesis of pseudo samples

Side-scan sonar (SSS) is the most important sensor in the field of ocean exploration, auto-detection of targets in SSS images is very critical. Deep neural networks (DNNs) have exhibited impressive performance but require very large number of training samples; only a limited number of SSS images will not suffice. In more extreme cases, no appropriate SSS image is available for some specific targets that need to be identified, making it impossible to train DNNs. In this paper, we deal with such extreme situations, how can a DNN recognize targets in SSS images without any training samples? which is the “zero-shot learning problem.” Inspired by the way humans perceive the world, we develop a zero-shot SSS image classification method through synthesis of pseudo SSS images. For a given category, we use a fixed style-transfer method to synthesize pseudo samples using common optical images and any available SSS images, and train the DNN with these pseudo samples. The zero-shot learning problem thus can be transformed to a conventional supervised learning problem. Experimental results showed we can achieve excellent classification ability even if no training samples are available. The source code is available at https://github.com/guizilaile23/ZSL-SSS.

A Study on the Research Trends of Degree dissertation on the Skin Beauty Superintend Program according to Ayurveda Constitution Types

A study on the research trends analysis of the degree dissertation on skin beauty program according to Ayurveda Constitution types. In terms of research subjects and data collection, data were searched with the keywords of “Ayurveda Constitution Types”, “Ayurveda Skin Beauty Superintend Program”, “Ayurveda therapy” and “Ayurveda oil” through the Research Information Search System (RISS). Then, a total of 123 master’s and Ph.D. papers were all examined. In addition, first classification was chosen as the title of the paper, and unrelated and redundant papers were filtered out. After a second classification through content examination, the final 32 papers were chosen and reviewed. For a study on Skin Beauty Superintend by Ayurveda Constitution Types, this study was mostly performed, using the papers which adopted measurement tools. Concerning research type, a literature review through a theoretical approach phrase, verification of survey questions and investigation on current constitution were most common. Recently, the number of studies targeted to identify an objective classification method has gradually increased. This study is significant in that it has attempted to analyze theses and dissertations on aesthetic superintend programs by domestic Ayurveda Constitution Types. Therefore, based on the analysis of Ayurveda s prior research and the prior study of the journal and the analysis of the situation in the conceptual theorem program of each keyword term, the development of a standardized aesthetic superintend program by Korean Ayurveda Constitution Type is needed. In addition, it is anticipated that this information would be available for future studies.

Machine learning utilising spectral derivative data improves cellular health classification through hyperspectral infra-red spectroscopy.

The objective differentiation of facets of cellular metabolism is important for several clinical applications, including accurate definition of tumour boundaries and targeted wound debridement. To this end, spectral biomarkers to differentiate live and necrotic/apoptotic cells have been defined using in vitro methods. The delineation of different cellular states using spectroscopic methods is difficult due to the complex nature of these biological processes. Sophisticated, objective classification methods will therefore be important for such differentiation. In this study, spectral data from healthy/traumatised cell samples using hyperspectral imaging between 2500–3500 nm were collected using a portable prototype device. Machine learning algorithms, in the form of clustering, have been performed on a variety of pre-processing data types including ‘raw’ unprocessed, smoothed resampling, background subtracted and spectral derivative. The resulting clusters were utilised as a diagnostic tool for the assessment of cellular health and quantified using both sensitivity and specificity to compare the different analysis methods. The raw data exhibited differences for one of the three different trauma types applied, although unable to accurately cluster all the traumatised samples due to signal contamination from the chemical insult. The background subtracted and smoothed data sets reduced the accuracy further, due to the apparent removal of key spectral features which exhibit cellular health. However, the spectral derivative data-types significantly improved the accuracy of clustering compared to other data types, with both sensitivity and specificity for the background subtracted data set being >94% highlighting its utility to account for unknown signal contamination while maintaining important cellular spectral features.

An Attention-based Bi-LSTM Method for Visual Object Classification via EEG

Background and ObjectiveDespite many models have been proposed for brain visual perception and content understanding via electroencephalograms (EEGs), due to the lack of research on the inherent temporal relationship, EEG-based visual object classification still demands the improvement on its accuracy and computation complexity. MethodsTo take full advantage of the uneven visual feature saturation between time segments, an end-to-end attention-based Bi-LSTM Method is proposed, named Bi-LSTM-AttGW. Two attention strategies are introduced to Bi-LSTM framework. The attention gate replaces the forget gate in traditional LSTM. It is only relevant to the historical cell state, and not related to the current input. Hence, the attention gate can greatly reduce the number of training parameters. Moreover, the attention weighting method is applied to Bi-LSTM output, and it can explore the most decisive information. ResultsThe best classification accuracy achieved by Bi-LSTM-AttGW model is 99.50%. Compared with the state-of-art algorithms and baseline models, the proposed method has great advantages in classification performance and computational complexity. Considering brain region level contribution on visual cognition task, we also verify our method using EEG signals collected from the frontal and occipital regions, that are highly correlated with visual perception tasks. ConclusionsThe results show promise towards the idea that human brain activity related to visual recognition can be more effectively decoded by neural networks with neural mechanism. The experimental results not only could provide strong support for the modularity theory about the brain cognitive function, but show the superiority of the proposed Bi-LSTM model with attention mechanism again.

Mesh Convolution: A Novel Feature Extraction Method for 3D Nonrigid Object Classification

Applying convolution methods to domains that lack regular underlying structures is a challenging task for 3D vision. Existing methods require the manual design of feature representations suitable for the task or full-voxel-level analysis, which is memory intensive. In this paper, we propose a novel feature extraction method to facilitate 3D nonrigid shape analysis. Our approach, called 3D-MConv, extends convolution operations from regular grids to irregular mesh sets by parametrizing a series of convolutional templates and adopts a novel local perspective to ensure that the algorithm is more invariant against global isometric deformation and articulation. We carefully design the convolutional template as a polynomial function that flexibly represents the local shape. An unsupervised learning method is adopted to learn the convolutional template function. By using the convolution operation and the movement of the template on the model surface, we can obtain the distribution of the typical template shapes. We combine this distribution feature with the spatial co-occurrence information of typical template shapes modelled by Markov chains to form a high-level descriptor of a 3D model. The support vector machine method is used to classify the nonrigid 3D objects. Experiments on SHREC10 and SHREC15 demonstrate that 3D-MConv achieves state-of-the-art accuracy on standard benchmarks.

IMCLASS – A USER-TAILORED MACHINE LEARNING IMAGE CLASSIFICATION CHAIN FOR CHANGE DETECTION OR LANDCOVER MAPPING

Abstract. With the increasing availability of satellite imagery at several spatial, spectral and temporal resolutions, the choice of the best image and the most appropriate method for object detection and classification of a broad range of land surface classes or processes is still a difficult task for many users. In order to guide the users, we proposed a user-tailored machine learning method (IMage CLASSification - ImCLASS) to detect and classifiy specific landcover classes.The method assumes a mono-class approach taking several ill-posed problems (e.g. class imbalance, high diversity inside the studied class, similarities with the adjacent samples…) as use cases (landslides, construction works in urban areas, burnt areas, vegetation classes…). It is a generalization of the ALADIM processor already validated in the context of landslide mapping and available as a service on the ESA GeoHazards Exploitation Platform (GEP). The proposed chain is able to combine optical and radar images, uses open source libraries, and is optimized for rapid calculation on HPC environments. The ImCLASS processor is presented and its performance is evaluated on three use cases: landslide detection and mapping after disasters in different regions of the World, urban classes change detection with a focus on construction works in Strasbourg, and crop mapping (vineyard) in the Grand-Est region. First results using either bi-dates or mono-date imagery are presented.

Android skin cancer detection and classification based on MobileNet v2 model

The latest developments in the smartphone-based skin cancer diagnosis application allow simple ways for portable melanoma risk assessment and diagnosis for early skin cancer detection. Due to the trade-off problem (time complexity and error rate) on using a smartphone to run a machine learning algorithm for image analysis, most of the skin cancer diagnosis apps execute the image analysis on the server. In this study, we investigate the performance of skin cancer images detection and classification on android devices using the MobileNet v2 deep learning model. We compare the performance of several aspects; object detection and classification method, computer and android based image analysis, image acquisition method, and setting parameter. Skin cancer actinic Keratosis and Melanoma are used to test the performance of the proposed method. Accuracy, sensitivity, specificity, and running time of the testing methods are used for the measurement. Based on the experiment results, the best parameter for the MobileNet v2 model on android using images from the smartphone camera produces 95% accuracy for object detection and 70% accuracy for classification. The performance of the android app for object detection and classification model was feasible for the skin cancer analysis. Android-based image analysis remains within the threshold of computing time that denotes convenience for the user and has the same performance accuracy with the computer for the high-quality images. These findings motivated the development of disease detection processing on android using a smartphone camera, which aims to achieve real-time detection and classification with high accuracy.

Methods for object recognition and classification for tele-controlled service robots

The need of personal service robots has been increased in the past several years. The tasks assigned to the robots vary in complexity and degree of freedom. In order to assist senior citizens and people with disabilities with their day to day tasks, a greater degree of freedom is required – more autonomy, ability to fulfil more complex tasks and easier for use human-machine interface. In this paper we show some of the methods that can be implemented in personal service robots for object recognition and classification in order to achieve greater autonomy, hence greater degree of freedom.

OP0060 MACHINE LEARNING BASED BERLIN SCORING OF MAGNETIC RESONANCE IMAGES OF THE SPINE IN PATIENTS WITH ANKYLOSING SPONDYLITIS FROM THE MEASURE 1 STUDY

Background:Magnetic resonance imaging (MRI) offers a non-invasive and objective method of early diagnosis and classification, monitoring disease burden and treatment response for patients (pts) with axial spondyloarthritis (axSpA) including ankylosing spondylitis (AS).1Numerous scoring schemes such as the AS Spine MRI Activity (ASspiMRIa) score are available for the quantitative assessment of MRI, but are subject to intra- and inter-rater variability, labor intensive and costly. Nevertheless, quantification of MRI changes has become an important tool to demonstrate treatment success of biologic drugs in axSpA.Objectives:To evaluate the performance of machine learning (ML) based software for automated Berlin grading of spinal MRI bone marrow oedema in pts with AS and compare with expert scoring.Methods:Fully automated ML software (Figure) was developed to detect and label 23 vertebrae, define vertebral units (VU) as per the Berlin modification of the ASspiMRIa score, and score each VU as either 0 (score of 0) or 1 (score of 1, 2 or 3). The ML algorithm was based on the previously developed SpineNet software.2Analysis included 108 pts from the secukinumab MEASURE 1 study3, in which imaging was done using T1 and STIR sagittal MRI at baseline and Weeks 16, 52, 104, 156 and 208. Two expert readers, blinded to treatment and visit, evaluated all images by ASspiMRIa score. The scores from Reader 2 (R2) were binned into two groups: 0 vs 1, 2, or 3. As a result of multiple pt time points and expert reading sessions, the complete dataset comprised of 10,988 VU. Ten-way cross-validation at per-VU was used to train and validate the ML software. The dataset was split into 10 randomly selected subsets, ensuring that each pt appears in only one subset, after which 8 subsets were used for training the ML software, 1 was used to check for correct training and 1 was used for validation. The process was repeated ten times such that all 10 subsets were used for validation. Accuracy weighted for the frequency of each category, sensitivity and specificity were calculated using scores from R2 as reference. Intra-reader accuracy was also calculated.Results:Accuracy of the software in relation to expert reader scores was 67% with a sensitivity of 0.63 and specificity of 0.70. The intra-reader accuracy was 71% and 77% for R1 and R2, respectively. Individual VU scoring of the Software vs. R2 are presented in the Table as a confusion matrix.Conclusion:Automated scoring of MR images in AS pts provided moderate agreement to that of expert reader-based assessments. ML software has potential to provide an automated guided-reading approach to scoring MR images, which may enable further clinical insights.

Tillage boundary detection based on RGB imagery classification for an autonomous tractor

In this study, a deep learning-based tillage boundary detection method for autonomous tillage by a tractor was developed, which consisted of image cropping, object classification, area segmentation, and boundary detection methods. Full HD (1920 × 1080) images were obtained using a RGB camera installed on the hood of a tractor and were cropped to 112 × 112 size images to generate a dataset for training the classification model. The classification model was constructed based on convolutional neural networks, and the path boundary was detected using a probability map, which was generated by the integration of softmax outputs. The results show that the F1-score of the classification was approximately 0.91, and it had a similar performance as the deep learning-based classification task in the agriculture field. The path boundary was determined with edge detection and the Hough transform, and it was compared to the actual path boundary. The average lateral error was approximately 11.4 cm, and the average angle error was approximately 8.9o. The proposed technique can perform as well as other approaches; however, it only needs low cost memory to execute the process unlike other deep learning-based approaches. It is possible that an autonomous farm robot can be easily developed with this proposed technique using a simple hardware configuration.

CNN-based 3D object classification using Hough space of LiDAR point clouds

With the wide application of Light Detection and Ranging (LiDAR) in the collection of high-precision environmental point cloud information, three-dimensional (3D) object classification from point clouds has become an important research topic. However, the characteristics of LiDAR point clouds, such as unstructured distribution, disordered arrangement, and large amounts of data, typically result in high computational complexity and make it very difficult to classify 3D objects. Thus, this paper proposes a Convolutional Neural Network (CNN)-based 3D object classification method using the Hough space of LiDAR point clouds to overcome these problems. First, object point clouds are transformed into Hough space using a Hough transform algorithm, and then the Hough space is rasterized into a series of uniformly sized grids. The accumulator count in each grid is then computed and input to a CNN model to classify 3D objects. In addition, a semi-automatic 3D object labeling tool is developed to build a LiDAR point clouds object labeling library for four types of objects (wall, bush, pedestrian, and tree). After initializing the CNN model, we apply a dataset from the above object labeling library to train the neural network model offline through a large number of iterations. Experimental results demonstrate that the proposed method achieves object classification accuracy of up to 93.3% on average.

Tropospheric Dust and Associated Atmospheric Circulations over the Mediterranean Region with Focus on Romania’s Territory

The aim of this study is to assess the distribution of dust over the Mediterranean region, with a special focus on the territory of Romania. Two parameters are analyzed—Dust Load (DL) and Aerosol Optical Depth (AOD), the data is obtained from the dust forecast model BSC-DREAM8b v2.0, for the period between December 2015 and February 2019. The main geographical features of dust occurrence in the Mediterranean region are presented at the monthly and annual scale. The results show that, for Romania, the dust load is high from February to June, when it reaches its annual maximum. The atmospheric circulation inducing intense dust events over Romania have also been assessed using an objective classification method. A key element for the dust transport from the Sahara toward South-Eastern Europe is represented by the development of a deep cyclone South of Italy, following thereafter a North-East path towards the Balkan peninsula. The results at the regional scale are analyzed in connection with the aerosol optical properties at the local scale (e.g., aerosol optical depth at 440 nm, Absorption Ångström Exponent and Scattering Ångström Exponent at 440 nm and 675 nm, respectively) retrieved from the Aerosol Robotic Network (AERONET-NASA) for Romania, using data from ACTRIS-RO monitoring sites from Iași, Cluj–Napoca, and Bucharest. The differences between the forecast model and the observational data are also explored. Our results also show that the contribution of the natural mineral dust to air pollution in Romania is small, representing not more than 10% of all kinds of aerosols detected over the observation points from the ACTRIS-RO network.

A deep neural network and classical features based scheme for objects recognition: an application for machine inspection

Computer Vision (CV) domain is widely used in the current era of automation and visual surveillance for the detection and classification of different objects in a diverse environment. The automatic machine inspection of different objects in the scenes is based on internal and external parameters like features that provide a huge amount of information related to the nature of an object in the scene. In this work, we propose a new automated method based on classical and deep learning feature selection. The proposed object classification method follows three steps. The data augmentation is performed in the first step to make the balance database. Later, Pyramid HOG (PHOG) and Central Symmetric LBP (CS-LBP) features are serially fused along with deep learning-based extracted features. The deep learning features are extracted from the pre-trained CNN model name Inception V3. In the third step, a new technique name Joint Entropy along with KNN (JEKNN) is employed to select the best features. The best-selected features are finally classified by well-known supervised learning methods and choose the best one based on higher accuracy. The proposed method is evaluated on Caltech101 balanced dataset and achieved maximum accuracy of 90.4% on Ensemble classifier which outperforms as compare to existing techniques.

Moving Object Classification Using 3D Point Cloud in Urban Traffic Environment

Moving object classification is essential for autonomous vehicle to complete high-level tasks like scene understanding and motion planning. In this paper, we propose a novel approach for classifying moving objects into four classes of interest using 3D point cloud in urban traffic environment. Unlike most existing work on object recognition which involves dense point cloud, our approach combines extensive feature extraction with the multiframe classification optimization to solve the classification task when partial occlusion occurs. First, the point cloud of moving object is segmented by a data preprocessing procedure. Then, the efficient features are selected via Gini index criterion applied to the extended feature set. Next, Bayes Decision Theory (BDT) is employed to incorporate the preliminary results from posterior probability Support Vector Machine (SVM) classifier at consecutive frames. The point cloud data acquired from our own LIDAR as well as public KITTI dataset is used to validate the proposed moving object classification method in the experiments. The results show that the proposed SVM-BDT classifier based on 18 selected features can effectively recognize the moving objects.

Analysis of different post mortem assessment methods for cerebral edema

While cerebral edema is a live-threatening condition in living persons, also an edema-like fluid redistribution can occur post mortem. In deceased, usually macroscopic signs are evaluated during autopsy in order to determine the presence or absence of cerebral edema.As a quantitative and objective classification is beneficial, an already existing method (Radojevic et al., 2017), which is based on a mathematical formula using the intracranial dimensions and the cerebral weight, was compared to the evaluation of macroscopic signs in 31 cases. The results showed an excellent agreement for the comparison between the raters as well as the measurement methods (at opened skull or in CT images). However, both measurement methods only poorly agree with the macroscopic edema evaluation.In order to find a more concordant method, the normalized cerebral weight, which puts the cerebral weight in relation to the intracranial volume, was calculated for 115 cases. This method resulted in an excellent agreement with the macroscopic rating and showed a clear numerical difference between the edematous and nonedematous group. While the influence of the post mortem time and the cooling time was found to be negligible, the age at death might confound the edema classification due to pre-existing cerebral atrophy leading to lower cerebral weights.In summary, the present study compared different assessment methods to classify cerebral edema and developed a rater independent, objective and quantitative classification method, which was as reliable as the rating of the forensic pathologists.

PixelHop: A successive subspace learning (SSL) method for object recognition

A new machine learning methodology, called successive subspace learning (SSL), is introduced in this work. SSL contains four key ingredients: (1) successive near-to-far neighborhood expansion; (2) unsupervised dimension reduction via subspace approximation; (3) supervised dimension reduction via label-assisted regression (LAG); and (4) feature concatenation and decision making. An image-based object classification method, called PixelHop, is proposed to illustrate the SSL design. It is shown by experimental results that the PixelHop method outperforms the classic CNN model of similar model complexity in three benchmarking datasets (MNIST, Fashion MNIST and CIFAR-10). Although SSL and deep learning (DL) have some high-level concept in common, they are fundamentally different in model formulation, the training process and training complexity. Extensive discussion on the comparison of SSL and DL is made to provide further insights into the potential of SSL.

Angle histogram of Hough transform as shape signature for visual object classification - (AHOC)

This work presents a new method for object classification using Hough transform (HT) and angle histogram as an object signature. Several methods are reported in the literature that exploit HT and other techniques as a pre-processing step to characterise objects to be used in detection, recognition, classification, etc. HT is a powerful technique to extract shape features from 2D objects; it has been used in many studies and implemented successfully in many applications. Our study is unique by post processing HT voting space using a binary threshold then computing an angle histogram of the resulting angle space as a shape signature of objects. Our image set consisted of 25 simple geometric shapes and six complex natural object classes of: trees, people, cars, airplanes, houses and horses. The method was trained and tested using 225 images from six different classes and found to be robust with a classification accuracy of 95.83%.