针对当前手势图像数据集不能均匀、全面地覆盖所有手势参数空间内的各种手势的问题,提出一种基于级联卷积神经网络的手势特征提取方法.该方法通过级联式模型,分层次地对高维度、高自由度的手势参数进行特征感知和提取.首先,将手腕角度参数作为手势参数的全局参数,进行划分和特征提取;然后,将手指角度参数作为局部参数,进行特征提取.为解决局部参数特征提取网络数量过多的问题,减少神经网络的数量和节约训练网络所需的时间与内存开销,采用多分支结构的神经网络模型,将五个手指的局部特征提取网络集成为一个整体.实验结果表明,所提方法在真实训练集上平均分类准确率达到95.13％,测试集平均准确率达到54％,测试集准确率相较于全卷积神经网络的算法提高了4.76个百分点.

Based on the problem of low stability and high network latency in the traditional building energy consumption acquisition system, in this paper, a building energy consumption acquisition system based on configuration is proposed. The system adopts the embedded technology and WAN communication technology such as TCP/IP, GSM, ZigBee, NB-loT and so on. Sensor-based system, the configuration system to support, embedded MCU as the core, a variety of network communication technologies complement each other, constitute the entire building energy collection system. Through experiments, the system can stably and quickly acquire the data information of the running equipment inside the building, and at the same time it can ensure the integrity and correctness of the data information transmission process. The system has the advantages of high automation, high reliability and fast transmission speed.

For the problem of low manual detection accuracy under the conditions of illumination and occlusion, the detection of human hands based on common optical images was explored, and an accurate manual detection method under general conditions was proposed. The method based on skin color model combined with Convolutional Neural Network (CNN) was mainly used. Realize the detection of human hands. Firstly, the skin color model is obtained according to the characteristics of skin color in the HSV (Hue, Saturation and Value) space, which is used to segment skin area. On this basis, a convolutional neural network for the detection of human hand contours is constructed, which is used to extract the human hand contour features to constrain skin region to obtain the hand region. The results show that even in light and shielding, it also has adaptability, which improves the accuracy of hand detection.

Gesture image segmentation is a challenge task due to the high degree of freedom of human gestures, large differences in shape and high flexibility, traditional pattern recognition and image processing methods are not effective in gesture detection. The traditional image segmentation based on the detection of skin color and the image of the depth image are limited by the effects of ambient light, skin color difference and image depth variation, resulting in unsatisfactory results. Therefore, we propose a hand gesture depth image segmentation method based on random forest. The method learns the gesture image feature representation of the depth image by supervising learning. Experiments show that the proposed method segments the gesture s’ pixels from the backgrounds area of the depth image. The proposed method potential has widely usages in gesture tracking, gesture recognition and human computer interaction.

Currently, the hand tracking based on deep learning has made good progress, but these literatures have less influence on the tracking accuracy of Convolutional Neural Network (CNN) architecture and parameters. In this paper, we proposed a new method to analyze the influence factors of gesture tracking. Firstly, we establish the gesture image and corresponding gesture parameter database based on virtual 3D human hand, on which the convolutional neural network models are constructed, after that we research some related factors, such as network structure, iteration times, data augmentation and Dropout, etc., that affect the performance of hand tracking. Finally we evaluate the objective parameters of the virtual hand, and make the subjective evaluation of the real hand extracted in the video. The results show that, on the premise of the fixed training amount of the hand, the effect of increasing the number of convolutional cores or convolution layers on the accuracy of the real gesture is not obvious, the data augmentation is obvious. For the real gesture, when the number of iterations and the Dropout ratio is about 20

To ensure the security and reliability of the transmission of digital images in the network,an image encryption algorithm was proposed based on the sequences of Logistic map and DES (Data Encryption Standard) combined with scrambling and diffusion.Firstly,two modified chaotic sequences of Logistic maps were generated and applied to encryption processes such as iteration and bit encryption;secondly,pixels were scrambled according to the sorted order of the chaotic sequences;finally,the encrypted image was generated with more effective diffusion and more random scrambling,as the initial scrambling and cryptographic function of DES were modified.After the encryption system was analyzed in terms of entropy,correlation and key space,the result shows that this image encryption method is able to cover up plaintext effectively and safely with huge key space,high speed,good anti-attack capability.

For the problem that BP neural network is easy to fall into the local minimum,the metropolis acceptance criteria in the mechanism of simulated annealing algorithm and the taboo list in the taboo searching algorithm are introduced in the system to escape from the local minimum,and the values of minima were recorded and compared in the taboo list,which is more efficient in dealing with multiple local minima and accurate system. The modified algorithm is applied to the BP network to construct a neural network which is easier to jump off from local minima. At last, the modified neural network algorithm is used for image compression and reconstruction. The experimental results show the higher efficiency and accuracy of the modified neural network.

Challenges exist in the field of sports news generation automatically from webcast that (1) finding hot events and sentences accurately; (2) organizing the selected sentences with highly readability. This paper proposes a framework to generate sports news automatically. First, to obtain accurate hot events and sentences, we design a neural network to predict the probabilities that each statement in live webcast script appears in the writing news, where the inputs of the neural network are weighed word vectors obtained from football keywords dictionary, and the outputs the similarity of statements in training live webcast script and sentences in training news. In this way, the “good” sentences selected from webcast contribute to the semi-finished sport news. To make the generated news to be possibly similar to human writing, we adopt idioms often appeared in football game to describe or summarize the games’ development or turns between the selected sentences, and come into being the final sport news. The proposed framework are validated on the training and test data set proved by “Sports News Generation from Live Webcast scripts” task of NLPCC 2016, the experiments show that the proposed method present good performance.