Search

The automatic recognition of facial expressions is challenging for many reasons, and there is still scope for research in the control of sample quality, the construction of effective feature extraction model and multi feature fusion. This study proposes a facial expression recognition method that combines transfer learning, attention mechanism, and feature fusion technology. In this method, various features are extracted from several pre-trained deep learning networks; the recognition weight of each sample is then learned via a quality attention network. This weight can be used to distinguish the quality of samples and suppress the contribution of low-quality samples in the training process. Finally, for a variety of features, this study proposes a fusion method based on feature clustering and correction. The experimental results indicate that the proposed method can significantly improve the accuracy of expression recognition and achieve state-of-the-art recognition accuracy for RAF-DB and FER2013+ datasets.

This paper shows 100% renewable electricity by 2050 in Australia is both technically and economically feasible. Modelling the transition from today’s electricity generation mix to 2050 using a capacity expansion model with hourly variability in wind and solar potential input, detailed transmission, ramp rate and inertia constraints, we show that the least-cost power system will consist mostly of wind, solar PV, pumped hydro energy storage and upgraded transmission networks. We demonstrate that the 100% renewable system is not only cost-competitive, but also equally robust in terms of reliability and security compared to the longstanding fossil fuel dominated power system.

This paper presents an automatic clustering algorithm for polarimetric synthetic aperture radar (PolSAR) data. Based on the Wishart mixture model which describes non-Gaussian areas more accurately, the PolSAR image is clustered by variational Bayesian expectation maximization algorithm. The incomplete gamma mixture distribution is incorporated to describe the equivalent number of looks for each cluster, which stands for the homogeneity and non-Gaussian statistics of each cluster. Then, the lower bound is computed each iteration until it reaches convergence. Meanwhile, spatial information is utilized by introducing two similarity measures. Finally, the experimental results of real measured AIRSAR dataset show that the proposed algorithm can obtain comparable clustering accuracy and determine the number of clusters automatically.

Large scale ontology applications require efficient reasoning services, of which ontology classification is the fundamental reasoning task. The special EL reasoners are efficient, but they can not classify ontologies with axioms outside the OWL 2 EL profile. The general-purpose OWL 2 reasoners for expressive Description Logics are less efficient when classifying the OWL 2 EL ontologies. In this work, we propose a novel technique that combines an OWL 2 reasoner with an EL reasoner for classification of ontologies expressed in DL SROIQ. We develop an efficient task decomposition algorithm for identifying the minimal non-EL module that is assigned to the OWL 2 reasoner, and the bulk of the workload is assigned to the EL reasoner. Furthermore, this paper reports on the implementation of our approach in the ComR system which integrates the two types of reasoners in a black-box manner. The experimental results show that our method leads to a reasonable task assignment and can offer a substantial speed up (over 50%) in ontology classification.

The soul and premise of cultivation modes of compound application talents are correct attitudes of education and engineering. This paper assumes that science and engineering universities should adapt to the needs of time and substitute “the attitudes of engineering education” for traditional “attitude of technology education”. The aim is to cultivate compound application talents with solid scientific theory and technology, management skills, spirits of humanity. By discussing curriculum system reform, strengthening practice, emphasizing the guidance to major-oriented educations and innovation, this paper tends to provide an applicable systematic talents-cultivating mode for ordinary colleges of science and engineering.

According to existing problems in universities undergraduate practice teaching work, a simulation model is established on neural network, according to experiment teaching, practice teaching, graduation project and graduation thesis, laboratory management and equipment use as input layer, quality of universities undergraduate practice teaching work as output layer, establish artificial neural network model, and actual data were used for training and practice of the network. The practice showed that the model has well recognizes the precision. Finally obtains digitized the appraisal result, can be accurate, irect-viewing reflection the quality undergraduate practice teaching work the fit and unfit, thus the wide prospect of artificial neural network in the evaluation of quality of universities undergraduate practice teaching work were presented.

The evaluation of real estate investment environment is a systematic project involving multi-factor. Accordingly, select the most impact on the real estate investment environment to evaluate factors as indicators, establish unascertained measure comprehensive evaluation model. Applications used by the model shows that the evaluation result can objectively reflect the real estate investment environment, which provides a new method and way for evaluation of real estate investment environment and has theoretical and practical value. Due to the ordered division of evaluation space, the unascertained measure model improves distinguish leve. It has reliable evaluation result and is a practical model.