High-density surface electromyography (HD-sEMG) and deep learning technology are becoming increasingly used in gesture recognition. Based on electrode grid data, information can be extracted in the form of images that are generated with instant values of multi-channel sEMG signals. In previous studies, image-based, two-dimensional convolutional neural networks (2D CNNs) have been applied in order to recognize patterns in the electrical activity of muscles from an instantaneous image. However, 2D CNNs with 2D kernels are unable to handle a sequence of images that carry information concerning how the instantaneous image evolves with time. This paper presents a 3D CNN with 3D kernels to capture both spatial and temporal structures from sequential sEMG images and investigates its performance on HD-sEMG-based gesture recognition in comparison to the 2D CNN. Extensive experiments were carried out on two benchmark datasets (i.e., CapgMyo DB-a and CSL-HDEMG). The results show that, where the same network architecture is used, 3D CNN can achieve a better performance than 2D CNN, especially for CSL-HDEMG, which contains the dynamic part of finger movement. For CapgMyo DB-a, the accuracy of 3D CNN was 1% higher than 2D CNN when the recognition window length was equal to 40 ms, and was 1.5% higher when equal to 150 ms. For CSL-HDEMG, the accuracies of 3D CNN were 15.3% and 18.6% higher than 2D CNN when the window length was equal to 40 ms and 150 ms, respectively. Furthermore, 3D CNN achieves a competitive performance in comparison to the baseline methods.

At present, in the field of video-based human action recognition, deep neural networks are mainly divided into two branches: the 2D convolutional neural network (CNN) and 3D CNN. However, 2D CNN’s temporal and spatial feature extraction processes are independent of each other, which means that it is easy to ignore the internal connection, affecting the performance of recognition. Although 3D CNN can extract the temporal and spatial features of the video sequence at the same time, the parameters of the 3D model increase exponentially, resulting in the model being difficult to train and transfer. To solve this problem, this article is based on 3D CNN combined with a residual structure and attention mechanism to improve the existing 3D CNN model, and we propose two types of human action recognition models (the Residual 3D Network (R3D) and Attention Residual 3D Network (AR3D)). Firstly, in this article, we propose a shallow feature extraction module and improve the ordinary 3D residual structure, which reduces the parameters and strengthens the extraction of temporal features. Secondly, we explore the application of the attention mechanism in human action recognition and design a 3D spatio-temporal attention mechanism module to strengthen the extraction of global features of human action. Finally, in order to make full use of the residual structure and attention mechanism, an Attention Residual 3D Network (AR3D) is proposed, and its two fusion strategies and corresponding model structure (AR3D_V1, AR3D_V2) are introduced in detail. Experiments show that the fused structure shows different degrees of performance improvement compared to a single structure.

With the popularity of the internet, the expression of emotions and methods of communication are becoming increasingly abundant, and most of these emotions are transmitted in text form. Text sentiment classification research mainly includes three methods based on sentiment dictionaries, machine learning and deep learning. In recent years, many deep learning-based works have used TextCNN (text convolution neural network) to extract text semantic information for text sentiment analysis. However, TextCNN only considers the length of the sentence when extracting semantic information. It ignores the semantic features between word vectors and only considers the maximum feature value of the feature image in the pooling layer without considering other information. Therefore, in this paper, we propose a convolutional neural network based on multiple convolutions and pooling for text sentiment classification (variable convolution and pooling convolution neural network, VCPCNN). There are three contributions in this paper. First, a multiconvolution and pooling neural network is proposed for the TextCNN network structure. Second, four convolution operations are introduced in the word embedding dimension or direction, which are helpful for mining the local features on the semantic dimensions of word vectors. Finally, average pooling is introduced in the pooling layer, which is beneficial for saving the important feature information of the extracted features. The verification test was carried out on four emotional datasets, including English emotional polarity, Chinese emotional polarity, Chinese subjective and objective emotion and Chinese multicategory. Our apporach is effective in that its result was up to 1.97% higher than that of the TextCNN network.

Compressive sensing (CS) theory has opened up new paths for the development of signal processing applications. Based on this theory, a novel single pixel camera architecture has been introduced to overcome the current limitations and challenges of traditional focal plane arrays. However, video quality based on this method is limited by existing acquisition and recovery methods, and the method also suffers from being time-consuming. In this paper, a multi-frame motion estimation algorithm is proposed in CS video to enhance the video quality. The proposed algorithm uses multiple frames to implement motion estimation. Experimental results show that using multi-frame motion estimation can improve the quality of recovered videos. To further reduce the motion estimation time, a block match algorithm is used to process motion estimation. Experiments demonstrate that using the block match algorithm can reduce motion estimation time by 30%.

This paper presents a systematic trajectory generation method for a bipedal robot walking on slopes. A suitable offline walking pattern based on an inverted pendulum model is designed and the planning is parameterized by step lengths, step period and other walking parameters. Meanwhile, a slight unevenness of a slope can cause serious instability for bipedal walking robots. Therefore, this paper also proposes an online control algorithm for a bipedal robot even walking on the unevenness slope, and the robot can adapt to the floor conditions. The control algorithm includes a landing time controller, landing direction controller, zero moment point (ZMP) regulation controller and attitude correction controller. During the process, accurate attitude information for these controllers is achieved through an adaptive filtering method and the ZMP position is measured through force sensing register sensors, which are attached to the robot’s feet. Finally, the experiment is carried out on a SCUT-I humanoid robot. The result proves that the method described in this paper can successfully control a robot walking on slopes.

With the rapid development of deep convolutional neural networks, more and more computer vision tasks have been well resolved. These convolutional neural network solutions rely heavily on the performance of the hardware. However, due to privacy issues or the network instability, we need to run convolutional neural networks on embedded platforms. Critical challenges will be raised by limited hardware resources on the embedded platform. In this paper, we design and implement an embedded inference framework to accelerate the inference of the convolutional neural network on the embedded platform. For this, we first analyzed the time-consuming layers in the inference process of the network, and then we design optimization methods for these layers. Also, we design a memory pool specifically for neural networks. Our experimental results show that our embedded inference framework can run a classification model MobileNet in 80ms and a detection model MobileNet-SSD in 155ms on Firefly-RK3399 development board.