Leandro Bezerra Marinho

Analysis of human tissue densities

A new algorithm (AHTD) is proposed to extract image features based on human tissue densities in medical images.AHTD is used to extract features from lung and brain CT images.AHTD is compared against three traditional feature extraction methods.The influence of the extraction method on the classification accuracy was assessed using four machine learning techniques.The results confirm the superiority and suitability of AHTD for use with medical images. Identification of diseases based on processing and analysis of medical images is of great importance for medical doctors to assist them in their decision making. In this work, a new feature extraction method based on human tissue density patterns, named Analysis of Human Tissue Densities (AHTD) is presented. The proposed method uses radiological densities of human tissues in Hounsfield Units to tackle the extraction of suitable features from medical images. This new method was compared against: the Gray Level Co-occurrence Matrix, Hus moments, Statistical moments, Zernikes moments, Elliptic Fourier features, Tamuras features and the Statistical Co-occurrence Matrix. Four machine learning classifiers were applied to each feature extractor for two CT image datasets:, one to classify lung disease in CT images of the thorax and the other to classify stroke in CT images of the brain. The attributes were extracted from the lung images in 5.2ms and obtained an accuracy of 99.01% for the detection and classification of lung diseases, while the attributes from the brain images were extracted in 3.8ms and obtained an accuracy of 98.81% for the detection and classification of stroke. These results show that the proposed method can be used to classify diseases in medical images, and can be used in real-time applications due to its fast extraction time of suitable attributes.

A novel mobile robot localization approach based on topological maps using classification with reject option in omnidirectional images

Novel method for localization via classification with reject option using omnidirectional images.Evaluation of feature extraction and machine learning techniques in omnidirectional images.Autonomous system based on mobile robot topological map localization.Validation in real and virtual environment.Novel virtual simulation environment and two data sets (virtual and real data). Mobile robot localization, which allows a robot to identify its position, is one of main challenges in the field of Robotics. In this work, we provide an evaluation of consolidated feature extractions and machine learning techniques from omnidirectional images focusing on topological map and localization tasks. The main contributions of this work are a novel method for localization via classification with reject option using omnidirectional images, as well as two novel omnidirectional image data sets. The localization system was analyzed in both virtual and real environments. Based on the experiments performed, the Minimal Learning Machine with Nearest Neighbors classifier and Local Binary Patterns feature extraction proved to be the best combination for mobile robot localization with accuracy of 96.7% and an Fscore of 96.6%.

A Novel Approach for Mobile Robot Localization in Topological Maps Using Classification with Reject Option from Structural Co-occurrence Matrix

Location is an elemental problem for mobile robotics due the importance of determining a position of the robot in space. This knowledge along with the environment map are basic information for robot mobility. In this paper, a new approach for navigation and location of mobile robots on topological maps using classification with reject option in attributes obtained from a Structural Co-occurrence Matrix (SCM) is proposed. Furthermore, we compare our approach with others state-of-the-art extractors, such as Statistical Moments, Gray-Level Co-occurrence Matrix (GLCM) and Local Binary Patterns (LBP). Structural Co-Occurrence Matrix was evaluated with the Average, Gaussian, Laplacian and Sobel filters. Regarding to classifiers, Bayesian classifier, Multilayer Perceptron (MLP) and Support Vector Machines (SVM) were analyzed. The descriptors Scale Invariant Feature Transform (SIFT) and Speed Up Robust Features (SURF) were also used. According to results, SCM was the fastest feature extractor with 0.117 s and accuracy of 100% in navigation test, showing the relevance of our approach in the mobile robot localization.

Lung Segmentation in Chest Computerized Tomography Images Using the Border Following Algorithm

This paper proposes a new method of lung segmentation in chest Computerized Tomography (CT) images called Follower of Lung Contour (FLC). This method works as follows: firstly, the image pixels are classified as pulmonary or not through an Artificial Neural Network (ANN) Multilayer Perceptron (MLP) based on pulmonary radiologic densities. After this, the lung detection is made based on achieved through the Border Following Algorithm together with predetermined rules that consider the detected objects area and positioning on the image. The proposed method validation is performed considering as Gold Standard a manual segmentation realized by a pulmonologist at Walter Cantidio Hospital of Federal University of Ceara. Moreover, 30 chest CT images were used, in which 10 are from patients diagnosed with Fibrosis, 10 are from patients with Chronic Obstructive Pulmonary Disease (COPD) and 10 are from healthy patients. The FLC results are compared with six other segmentation methods results using the Gold Standard as reference. Thus, the FLC algorithm shows good results with an average accuracy of 98% and average harmonic means of 98%. Furthermore, it can be concluded that this method may be part of a system to aid in medical diagnosis on Pulmonology.

A New Approach to Human Activity Recognition Using Machine Learning Techniques

Recognition of human activities aims a wide diversity of applications. However, identifying complicated activities continues a challenging and active research area. In this work, we assess a new approach of feature selection for human activity recognition. For the task, we also compare state-of-the-art classifiers, e.g., Bayes classifier, kNN, MLP, SVM, MLM and MLM-NN. Based on the experiments, the MLM-NN is able to speed up the original MLM while holding equivalent accuracy. MLM and SVM achieved accuracy of more than 99.2% in the original data set and 98.1% using new feature selection method. Results show that the proposed feature selection approach is a promising alternative to activity recognition on smartphones.

Mechanical Traction Machine Retrofitting via Digital Data Acquisition and Processing

This paper presents the development of a retrofit system in the mechanical tensile testing machine. The data acquisition system was developed using the pressure sensor, encoder and electronic circuits. The software developed automatically calculates the properties involved, generating technical report based on the specific standard ABNT. After all, the system was validated by the results of mechanical tensile tests on specimens. The results show that the retrofit was effective and that can be used commercially.