Computer Vision – ACCV 2009, 9th Asian Conference on Computer Vision, Xi’an, September 23-27, 2009, Revised Selected Papers, Part III

Human activity recognition using MotionNode sensors is getting prominence effect in our daily life logs. Providing accurate information on human's activities and behaviors is one of the most challenging tasks in ubiquitous computing and human-Computer interaction. In this paper, we proposed an efficient model for having statistical features along SMO-based random forest. Initially, we processed a 1-D Hadamard transform wavelet and 1-D LBP based extraction algorithm to extract valuable features. For activity classification, we used sequential minimal optimization along with Random Forest over two benchmarks USC-HAD dataset and IMSB datasets. Experimental results show that our proposed model can compete with other state-of-the-art methods and can be effectively used to recognize robust human activities in terms of efficiency and accuracy.

Wearable Sensors for Activity Analysis using SMO-based Random Forest over Smart home and Sports Datasets

Clustering with similarity preserving

Odor-induced emotion recognition based on average frequency band division of EEG signals

Objective Convolutional neural networks (CNNs) have proven successful as function approximators and have therefore been used for classification problems including electroencephalography (EEG) signal decoding for brain-computer interfaces (BCI). Artificial neural networks, however, are considered black boxes, because they usually have thousands of parameters, making interpretation of their internal processes challenging. Here we systematically evaluate the use of CNNs for EEG signal decoding and investigate a method for visualizing the CNN model decision process. Approach We developed a CNN model to decode the covert focus of attention from EEG event-related potentials during object selection. We compared the CNN and the commonly used linear discriminant analysis (LDA) classifier performance, applied to datasets with different dimensionality, and analyzed transfer learning capacity. Moreover, we validated the impact of single model components by systematically altering the model. Furthermore, we investigated the use of saliency maps as a tool for visualizing the spatial and temporal features driving the model output. Main results The CNN model and the LDA classifier achieved comparable accuracy on the lower-dimensional dataset, but CNN exceeded LDA performance significantly on the higher-dimensional dataset (without hypothesis-driven preprocessing), achieving an average decoding accuracy of 90.7% (chance level = 8.3%). Parallel convolutions, tanh or ELU activation functions, and dropout regularization proved valuable for model performance, whereas the sequential convolutions, ReLU activation function, and batch normalization components reduced accuracy or yielded no significant difference. Saliency maps revealed meaningful features, displaying the typical spatial distribution and latency of the P300 component expected during this task. Significance Following systematic evaluation, we provide recommendations for when and how to use CNN models in EEG decoding. Moreover, we propose a new approach for investigating the neural correlates of a cognitive task by training CNN models on raw high-dimensional EEG data and utilizing saliency maps for relevant feature extraction.

https://iopscience.iop.org/article/10.1088/1741-2552/ab3bb4/pdf

Convolutional neural networks for decoding of covert attention focus and saliency maps for EEG feature visualization.

Breast MRI interpretation requires that radiologists examine several images, depending on the acquisition protocol that is managed in the health institution, a very subjective and time-consuming process that reports large variability, which affects the final diagnosis and prognosis of the patient. In this paper, we present a computational method for classifying lesions detected in breast MRI studies, which aims to reduce physician subjectivity. The proposed approach take advantage of the ability of the Multiple Kernel Learning (MKL) strategy for optimally fusing the features extracted from the different image sequences that compose a breast MRI Study, which describe the grey level distribution of the original image and its saliency map, computed using the Graph-based Visual Saliency (GBVS) algorithm. Breast lesions were classified as positive and negative findings with an accuracy of \(85.5\%\) and \(84.8\%\) when nine and five sequences were used, respectively.

Breast Lesion Discrimination Using Saliency Features from MRI Sequences and MKL-Based Classification

Multi-label learning has been becoming an increasingly active area into the machine learning community due to a wide variety of real world problems. However, only over the past few years class balancing for these kind of problems became a topic of interest. In this paper, we present a novel method named hyperparameter calibration to treat class imbalance in a multi-label problem, to this aim we develop an extensive analysis over four real-world databases and two own synthetic databases exhibiting different ratios of imbalance. The empirical analysis shows that the proposed method is able to improve the classification performance when it is combined with three of the most widely used strategies for treating multi-label classification problems.

Multi-label Learning by Hyperparameters Calibration for Treating Class Imbalance

This work proposes two novel alternatives for dealing with the highly important issue of the clustering performance estimation. One of the measures is the cluster coherence aimed to quantifying the normalized ratio of cuts within a graph-partitioning framework, and therefore it uses a graph-driven approach to explore the nature of data regarding the cluster assignment. The another one is the probability-based-performance quantifier, which calculates a probability value for each cluster through relative frequencies. Proposed measures are tested on some clustering representative techniques applied to real and artificial data sets. Experimental results probe the readability and robustness to noisy labels of our measures.

Two Novel Clustering Performance Measures Based on Coherence and Relative Assignments of Clusters

Biodiversity loss in tropical ecosystems is advancing due to factors such as human pressure and climate change, and determining their ecological transformation level can help to monitor climate change trends. Soundscapes can indicate the health of an ecosystem; therefore, passive acoustic monitoring is interesting because of its non-invasive nature, low cost, and allows recording over long periods. However, soundscapes generate a large amount of data that must be analyzed using automatic techniques such as machine learning to reduce manual labor. This process is generally based on acoustic indices, which present limitations such as sensitivity to background noise or biased results to the study sites under evaluation. Therefore, in this document, we propose training supervised models to identify landscape transformation from studied site soundscape at three stages: low, medium, and high. These models are trained on data collected in 24 tropical sites in the Colombian Caribbean region from 2015 to 2017. To achieve this, we extracted high-level features from a VGGish architecture from each recording to assess the performance of features obtained from the raw time series with neural networks. Furthermore, we compared the features from VGGish with 60 acoustic indices as features to train Random Forest, Neural Network, and XGBoost classifiers. F1-score for test data accomplishes a performance of 90% with the XGBoost classifier on the acoustic indices set, and the neural network achieves 95% for the VGGish feature set. Through supervised models, it is possible to identify the level of transformation of an ecosystem from its soundscape. 

Identification of Tropical Dry Forest Transformation from Soundscapes Using Supervised Learning

Training of deep learning models requires large and properly labeled datasets, which make unfeasible using it for developing computer-aided diagnosis systems in medical imaging. As an alternative, transfer learning has shown to be useful to extract deep features using architectures previously trained. In this paper, a new method for classification of breast lesions in magnetic resonance imaging is proposed, which uses the pre-trained ResNet-50 architecture for extracting a set of image features that are then used by an SVM model for differentiating between positive and negative findings. We take advantage of the ResNet-50 architecture for introducing volumetric lesion information by including three consecutive slices per lesion. Filters used as feature descriptors were selected using a multiple kernel learning method, which allows identifying those filters that provide the most relevant information for the classification task. Additionally, instead of using raw filters as features, we propose to characterize it using statistical moments, which improves the classification performance. The evaluation was conducted using a set of 146 ROIs extracted from three sequences proposed for designing abbreviated breast MRI protocols (DCE, ADC, and T2-Vista). Positive findings were identified with an AUC of 82.4 using a DCE image, and 81.08 fusing features from the three sequences.

Andrés Eduardo Castro-Ospina

Papers

Breast Lesion Discrimination Using Saliency Features from MRI Sequences and MKL-Based Classification

Multi-label Learning by Hyperparameters Calibration for Treating Class Imbalance

Two Novel Clustering Performance Measures Based on Coherence and Relative Assignments of Clusters

Identification of Tropical Dry Forest Transformation from Soundscapes Using Supervised Learning

Characterizing ResNet Filters to Identify Positive and Negative Findings in Breast MRI Sequences.