As a hot issue of visual recognition and the premise of human-computer intelligent interaction, emotion recognition from facial expression is highly recognized and widely used. However, face images collected in the natural environment tend to have different head postures and intensities of illumination, which interfere with the discrimination effect of the facial expression. In this paper, an enhanced facial expression recognition model based on sparse representation using data normalization is designed, which can boost the recognition performance in an unconstrained environment compared with the original model. Multi-task Convolutional Neural Networks (MTCNN) is used for facial landmark detection along with affine transformation for aligning all the faces. Meanwhile, histogram equalization is employed to normalize the gray-scale. Finally, the test image is classified by sparse coding. The real effect of the proposed model is verified by experiments on the CK+ and Oulu-CASIA database, and the accuracy is improved to 51.33% and 60.76% respectively. Besides, the amount of data is greatly reduced by sparse coding.
In order to reduce the injection cost of large-scale manufacturing industry and solve the problem of uncontrollable quality in the production process, this paper studies the quality control of an intelligent injection molding system based on process monitoring. The system can reliably and objectively monitor the changes of cavity pressure and temperature at different positions in the mold forming process. When the product quality is abnormal, the cavity pressure curve will change greatly. Through the system, the abnormal products can be automatically screened out, the automation of quality detection can be realized, and the qualified rate of quality products can be improved. Based on the data of cavity pressure and temperature collected by the system, a BP neural network pressure prediction model is established. The model takes some important process parameters, the pressure value at the beginning of filling as the input, and the cavity pressure at the remaining stages as the output. Through the model, the expected prediction accuracy is obtained. Finally, it can be implemented on the injection molding machine, and the injection molding process can be adjusted online Some input values in the process provide data support for further quality control.
The research of this paper applied data mining technologies to examine one of the potential causes of income inequality: inflation. In order to better understand the link between inflation rates and the growth of income inequality, we obtained and analyzed historical data sets of inflation records such as the Consumer Price Index, Gini Coefficient, Producer Price Indices, Unemployment Rate, Minimum Wage, Gross Domestic Product, Federal Tax Rate, etc. to draw empirical conclusions. This study utilized data mining technologies, to develop a data model that analyzes substantial amount of data which contain information on the income, expenses and the financial footprints of families in the United States. The deep learning intelligent system learned from the detailed historical data of the trends of the income inequality, found the major elements that govern inflation, and predicted inflation rates in the future. The public and the economic research society can use the information for their decision making.
Video captioning plays an important role in video retrieval and personalized recommendations with the rapid rise of short videos. In the process of video captioning, the accuracy of caption generation may be affected by the existence of irrelevant visual content. To address this problem, a two-stage attention model is proposed, including object-level attention and frame-level attention. The object-level attention first focuses on the salient objects in each frame and then the frame-level attention adaptively attends to the frame most relevant to the caption. In addition, because of the exposure bias problem in recurrent neural network (RNN)-based sequence generation, generative adversarial network (GAN) and reinforcement learning (RL) algorithm are combined to make the generated caption more accurate and natural. For the discriminator in the GAN framework, we take both the generated caption and the video features as input, then a new reward function is proposed to update the parameters of generator. The experiments on two benchmark datasets, MSVD and MSR-VTT, have shown the effectiveness of our proposed method.
The decision-making of taxi drivers in the airport is one of the important factors to maintain the balance of taxi supply and demand. This paper uses the large taxi trip data set to model the decision-making process of taxi drivers in Zhengzhou airport, and puts forward reasonable suggestions for improving the ground traffic and passenger satisfaction of Xinzheng airport.
