RESUMO
Enormous heterogeneous sensory data are generated in the Internet of Things (IoT) for various applications. These big data are characterized by additional features related to IoT, including trustworthiness, timing and spatial features. This reveals more perspectives to consider while processing, posing vast challenges to traditional data fusion methods at different fusion levels for collection and analysis. In this paper, an IoT-based spatiotemporal data fusion (STDF) approach for low-level data in-data out fusion is proposed for real-time spatial IoT source aggregation. It grants optimum performance through leveraging traditional data fusion methods based on big data analytics while exclusively maintaining the data expiry, trustworthiness and spatial and temporal IoT data perspectives, in addition to the volume and velocity. It applies cluster sampling for data reduction upon data acquisition from all IoT sources. For each source, it utilizes a combination of k-means clustering for spatial analysis and Tiny AGgregation (TAG) for temporal aggregation to maintain spatiotemporal data fusion at the processing server. STDF is validated via a public IoT data stream simulator. The experiments examine diverse IoT processing challenges in different datasets, reducing the data size by 95% and decreasing the processing time by 80%, with an accuracy level up to 90% for the largest used dataset.
RESUMO
The most dangerous disease in recent decades is lung cancer. The most accurate method of cancer diagnosis, according to research, is through the use of histopathological images that are acquired by a biopsy. Deep learning techniques have achieved success in bioinformatics, particularly medical imaging. In this paper, we present an innovative method for rapidly identifying and classifying histopathology images of lung tissues by combining a newly proposed Convolutional Neural Networks (CNN) model with a few total parameters and the enhanced Light Gradient Boosting Model (LightGBM) classifier. After the images have been pre-processed in this study, the proposed CNN technique is provided for feature extraction. Then, the LightGBM model with multiple threads has been used for lung tissue classification. The simulation result, applied to the LC25000 dataset, demonstrated that the novel technique successfully classifies lung tissue with 99.6% accuracy and sensitivity. Furthermore, the proposed CNN model has achieved the lowest training parameters of only one million parameters, and it has also achieved the shortest processing time of just one second throughout the feature extraction process. When this result is compared with the most recent state-of-the-art approaches, the suggested approach has increased effectiveness in the areas of both disease classification accuracy and processing time.
RESUMO
Image understanding and scene classification are keystone tasks in computer vision. The development of technologies and profusion of existing datasets open a wide room for improvement in the image classification and recognition research area. Notwithstanding the optimal performance of exiting machine learning models in image understanding and scene classification, there are still obstacles to overcome. All models are data-dependent that can only classify samples close to the training set. Moreover, these models require large data for training and learning. The first problem is solved by few-shot learning, which achieves optimal performance in object detection and classification but with a lack of eligible attention in the scene classification task. Motivated by these findings, in this paper, we introduce two models for few-shot learning in scene classification. In order to trace the behavior of those models, we also introduce two datasets (MiniSun; MiniPlaces) for image scene classification. Experimental results show that the proposed models outperform the benchmark approaches in respect of classification accuracy.