Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-04-01 Cooperative journals: 《计算机应用研究》
Abstract: Face recognition technology based on deep learning is one of the hot topics in the field of artificial intelligence. Considering the complexity of the angle, light and resolution of face images in real environment, this paper improves the network structure of Inception-ResNet-V1, completes the related work of dataset production and hyper-parameter adjustment, and carries out experimental research on the home service robot platform. The experimental results show that the improved network in this paper achieve 99.22% accuracy in LFW testset, which is higher than 99.05% of the original network structure. It reaches 99.20% accuracy on the Asian face dataset, which is 97.10% higher than the original network structure. The false recognition rate on self built mismatched face dataset is 3.43%, which is lower than 12.28% of the original network structure. It can be seen that compared with the original network structure, the improved network structure improves the accuracy of face recognition and reduces the false recognition rate.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-04-01 Cooperative journals: 《计算机应用研究》
Abstract: Based on matrix factorization techniques, in this paper, implemented a modified binomial matrix decomposition algorithm in order to solve the recommender system’s rating prediction problem. Suppose the user's rating of the item is based on the binomial distribution. There are differences in the user's rating habits, and there are differences in the popularity of the items, resulting in an offset in the user-item scoring matrix. So, use the maximum a posteriori estimate to design model and the model is optimized by a stochastic gradient descent algorithm. The experimental results show that the modified binomial matrix decomposition algorithm is superior to the traditional binomial matrix decomposition algorithm in terms of recommender accuracy and offline calculation time on the MovieLens 100K datasets.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-08-13 Cooperative journals: 《计算机应用研究》
Abstract: With the promotion and application of robots in the service industry, especially in the family service, the demand for information collection or target recognition for service robots is also getting stronger and stronger. Traditional commodity recognition processes typically use the more classic image recognition and machine learning algorithms such as support vector machines (SVM) , random forest or adaboost, then use the basic characteristics of the gradient, texture or color of the target image. It can be applied in a relatively simple background, but it is hard to have a more prominent performance in a complicated background environment, and it is difficult to achieve a high accuracy. At present, the convolution neural network (CNN) , which is superior in target recognition, has become the first choice in many target recognition scenarios. Considering the hardware configuration cost of service robot, Faster R-CNN, a fast algorithm of region-based convolutional neural network (R-CNN) , is introduced into the system and identified by CPU. The CNN network is used to extract image features and access to a regional proposal layer behind it. The experimental results show that it is feasible to apply the deep learning recognition method to the service robot platform. The recognition effect is accurate and the test results are good.