Search

The common classifiers usually used to detect fall incidents depend on building and maintaining complex feature extraction for accurate machine learning tasks. However, these efforts have not succeeded in determining an ideal classifier or feature extraction for fall detection. In this research, we address the feature extraction problem along with the choice of the most appropriate classifier by using deep learning where the most prominent features are learned over the numerous layers of the network. More specifically, a general framework that relies on a faster region-based convolutional neural network was designed and developped to recognize the fall incidents. In particular, we designed three custom architectures and exploited transfer learning by using pre-trained networks such as the VGG-16 and AlexNet to overcome the fall detection challenge. The performance of the proposed networks showed the advantage of the pre-trained networks, where VGG-16 scored highest in those measures followed by AlexNet, the custom networks showed impressive results that were close to the pre-trained networks.

Semi-supervised clustering relies on both labeled and unlabeled data to steer the clustering process towards optimal categorization and escape from local minima. In this paper, we propose a novel fuzzy relational semi-supervised clustering algorithm based on an adaptive local distance measure (SSRF-CA). The proposed clustering algorithm utilizes side-information and formulates it as a set of constraints to supervise the learning task. These constraints are expressed using reward and penalty terms, which are integrated into a novel objective function. In particular, we formulate the clustering task as an optimization problem through the minimization of the proposed objective function. Solving this optimization problem provides the optimal values of different objective function parameters and yields the proposed semi-supervised clustering algorithm. Along with its ability to perform data clustering and learn the underlying dissimilarity measure between the data instances, our algorithm determines the optimal number of clusters in an unsupervised manner. Moreover, the proposed SSRF-CA is designed to handle relational data. This makes it appropriate for applications where only pairwise similarity (or dissimilarity) information between data instances is available. In this paper, we proved the ability of the proposed algorithm to learn the appropriate local distance measures and the optimal number of clusters while partitioning the data using various synthetic and real-world benchmark datasets that contain varying numbers of clusters with diverse shapes. The experimental results revealed that the proposed SSRF-CA accomplished the best performance among other state-of-the-art algorithms and confirmed the outperformance of our clustering approach.

Deafness does not restrict its negative effect on the person’s hearing, but rather on all aspect of their daily life. Moreover, hearing people aggravated the issue through their reluctance to learn sign language. This resulted in a constant need for human translators to assist deaf person which represents a real obstacle for their social life. Therefore, automatic sign language translation emerged as an urgent need for the community. The availability and the widespread use of mobile phones equipped with digital cameras promoted the design of image-based Arabic Sign Language (ArSL) recognition systems. In this work, we introduce a new ArSL recognition system that is able to localize and recognize the alphabet of the Arabic sign language using a Faster Region-based Convolutional Neural Network (R-CNN). Specifically, faster R-CNN is designed to extract and map the image features, and learn the position of the hand in a given image. Additionally, the proposed approach alleviates both challenges; the choice of the relevant features used to encode the sign visual descriptors, and the segmentation task intended to determine the hand region. For the implementation and the assessment of the proposed Faster R-CNN based sign recognition system, we exploited VGG-16 and ResNet-18 models, and we collected a real ArSL image dataset. The proposed approach yielded 93% accuracy and confirmed the robustness of the proposed model against drastic background variations in the captured scenes.

Clustering is a fundamental problem in machine learning. To address this, a large number of algorithms have been developed. Some of these algorithms, such as K-means, handle the original data directly, while others, such as spectral clustering, apply linear transformation to the data. Still others, such as kernel-based algorithms, use nonlinear transformation. Since the performance of the clustering depends strongly on the quality of the data representation, representation learning approaches have been extensively researched. With the recent advances in deep learning, deep neural networks are being increasingly utilized to learn clustering-friendly representation. We provide here a review of existing algorithms that are being used to jointly optimize deep neural networks and clustering methods.

Recycling plays a vital role in saving the planet for future generations as it allows keeping a clean environment, reducing energy consumption, and saving materials. Of special interest is the plastic material which may take centuries to decompose. In particular, the Polyethylene Terephthalate (PET) is a widely used plastic for packaging various products that can be recycled. Sorting PET can be performed, either manually or automatically, at recycling facilities where the post-consumed objects are moving on the conveyor belt. In particular, automated sorting can process a large amount of PET objects without human intervention. In this paper, we propose a computer vision system for recognizing PET objects placed on a conveyor belt. Specifically, DeepLabv3+ is deployed to segment PET objects semantically. Such system can be exploited using an autonomous robot to compensate for human intervention and supervision. The conducted experiments showed that the proposed system outperforms the state of the art semantic segmentation approaches with weighted IoU equals to 97% and Mean BFscore equals to 89%.

This research aims to separate the hands from the background of colored images representing the Arabic Sign language alphabet gestures. This hand segmentation task is one of the main challenges of image based Sign language recognition systems due to the issue of skin tones variations and the complexity of the background. For this purpose, an efficient system that segment the hand object and separate it from the rest of the image based on deep learning is investigated. More specifically, the DeepLab v3+ network architecture that is a combination of spatial pyramid pooling module and encode-decoder structure will be trained to learn the visual characteristics of the hand and segment it with detailed boundaries. The effectiveness of the proposed solution is investigated on a large dataset of size 12000 with an accuracy of 98%, an IoU of 93% of and BF score of 87%

Meningioma is among the most common primary tumors of the brain. The firmness of Meningioma is a critical factor that influences operative strategy and patient counseling. Conventional methods to predict the tumor firmness rely on the correlation between the consistency of Meningioma and their preoperative MRI findings such as the signal intensity ratio between the tumor and the normal grey matter of the brain. Machine learning techniques have not been investigated yet to address the Meningioma firmness detection problem. The main purpose of this research is to couple supervised learning algorithms with typical descriptors for developing a computer-aided detection (CAD) of the Meningioma tumor firmness in MRI images. Specifically, Local Binary Patterns (LBP), Gray Level Co-occurrence Matrix (GLCM) and Discrete Wavelet Transform (DWT) are extracted from real labeled MRI-T2 weighted images and fed into classifiers, namely support vector machine (SVM) and k-nearest neighbor (KNN) algorithm to learn association between the visual properties of the region of interest and the pre-defined firm and soft classes. The learned model is then used to classify unlabeled MRI-T2 weighted images. This paper represents a baseline comparison of different features used in CAD system that intends to accurately recognize the Meningioma tumor firmness. The proposed system was implemented and assessed using a clinical dataset. Using LBP feature yielded the best performance with 95% of F-score, 87% of balanced accuracy and 0.87 of the area under ROC curve (AUC) when coupled with KNN classifier, respectively.