One-shot 3D skeleton action recognition task struggles with diverse intra-class action execution styles, causing excessive discriminative information to obstruct obtaining separable feature space. We innovatively propose leveraging shared information among intra-class action executions to mitigate the over-influence of discriminative information. To this end, we proposed dynamic information interaction module (DIIM) that enables shared information to effectively weaken excessive discriminative information. Specifically, DIIM facilitates effective information interaction by constructing a guided evolution pool to store execution-related shared information and ensure such information can be retrieved. We devise shared-discriminative projection strategy (SDPS) which adopts different feature extraction strategies for specific skeleton topologies to target mining discriminative and shared information from different views of skeleton data. In summary, our proposed Cross-View Dynamic Information Interaction (CVDII) framework integrates DIIM and SDPS, effectively tackles the problem of discriminative information redundancy caused by diverse intra-class action execution styles. Experiments conducted on NTU 60, NTU 120, PKU-MMD, and Kinetics datasets demonstrate that our proposed CVDII achieves remarkable performance.

This paper aims to address the problem of efficient extraction of complete multi-scale information for supervised monocular depth estimation. Most of the existing depth estimation methods are based on Convolutional Neural Network (CNN). By gradually exploring the contextual and semantic features, they have achieved good results in scene depth estimation. However, with the expansion of the receptive field, global information limited by the local induction bias is gradually suppressed, resulting in the performance cannot be further improved. Recently, Transformer-based methods have been widely used to model the global correlation between features. Nevertheless, since the Transformer networks are not spatially aware enough, they usually lose local details and have no clear mechanism for reusing features when processing images. The Transformer networks perform self-attention mechanism at each location and cannot directly obtain information from other locations for features. Therefore, we propose a novel dual framework called as Transformer-CNN, which includes the Transformer-branch and the CNN-branch for monocular depth estimation. Specifically, the Transformer-branch is able to model the global contextual information and the CNN-branch can capture local spatial relationships in images. However, simply fusing these two independent branches may result in insufficient feature aggregation. To this end, we design a Parallel Feature Interaction Module (PFIM), which contains a Self-Attention Module (SAM) and a Cross-Attention Module (CAM), so as to highlight features from the Transformer-branch and the CNN-branch respectively and extract complementary information between the two branches. Meanwhile, in order to make full use of the low-level features with low quality in the scene, we propose a Low-level Information Acquisition Module (LIAM) to capture texture-related information and preserve texture details in the CNN-branch. Finally, to address the lack of multi-scale contextual information in Vision Transformer (ViT), we introduce a Wide Area Multi-scale Decoder (WAMD), which incorporates the multi-scale feature representations into the decoder part via a Wide Area Attention (WAA). Extensive experiments on benchmark datasets collected in the outdoor and indoor environments demonstrate the competitive results of the proposed method, compared with the state-of-the-art monocular depth estimation methods.

Both global contextual information and local texture information are of vital importance for the semantic segmentation of remote sensing images due to the high spatial resolution of remote sensing images and large variations in intra-class object size. In this letter, we propose a novel dual-path feature fusion semantic segmentation network for remote sensing images. A pure convolutional module called dual-path feature extraction module (DPFE) is applied to model global contextual and local texture features simultaneously with low complexity. Inspired by ConvNeXt with comparable global contextual modeling capacity with Transformer, the global path of DPFE draws some successful strategies of ConvNeXt to generate powerful global feature. Meanwhile, an attention feature fusion module (AFF) is proposed, which achieves the global and local feature comprehensive fusion by exploring the correlation of channels through attention mechanism. The proposed network is evaluated on Vaihingen and Potsdam benchmarks and the quantitative results show the proposed network can achieve overall accuracy (OA) of 91.3% and 89.7%, respectively, which are better than several representative semantic segmentation approaches.

Abstract The proposed fall detection approach is aimed at building a support system for the elders. In this work, a method based on human pose estimation and lightweight neural network is used to detect falls. First, the OpenPose is used to extract human keypoints and label them in the images. After that, the modified MobileNetV2 network is used to detect falls by integrating both human keypoint information and pose information in the original images. The above operation can use the original image information to correct the deviation in the keypoint labeling process. Through experiments, the accuracy of the proposed method is 98.6% and 99.75% on the UR and Le2i datasets, which is higher than the listed comparison methods.

In this article, an optimized backstepping (OB) control scheme is proposed for a class of stochastic nonlinear strict-feedback systems with unknown dynamics by using reinforcement learning (RL) strategy of identifier-critic-actor architecture, where the identifier aims to compensate the unknown dynamic, the critic aims to evaluate the control performance and to give the feedback to the actor, and the actor aims to perform the control action. The basic control idea is that all virtual controls and the actual control of backstepping are designed as the optimized solution of corresponding subsystems so that the entire backstepping control is optimized. Different from the deterministic system, stochastic system control needs to consider not only the stochastic disturbance depicted by the Wiener process but also the Hessian term in stability analysis. If the backstepping control is developed on the basis of the published RL optimization methods, it will be difficult to be achieved because, on the one hand, RL of these methods are very complex in the algorithm thanks to their critic and actor updating laws deriving from the negative gradient of the square of approximation of Hamilton-Jacobi-Bellman (HJB) equation; on the other hand, these methods require persistence excitation and known dynamic, where persistence excitation is for training adaptive parameters sufficiently. In this research, both critic and actor updating laws are derived from the negative gradient of a simple positive function, which is yielded on the basis of a partial derivative of the HJB equation. As a result, the RL algorithm can be significantly simplified, meanwhile, two requirements of persistence excitation and known dynamic can be released. Therefore, it can be a natural selection for stochastic optimization control. Finally, from two aspects of theory and simulation, it is demonstrated that the proposed control can arrive at the desired system performance.

In this article, a fuzzy logic system (FLS)-based adaptive optimized backstepping control is developed by employing reinforcement learning (RL) strategy for a class of nonlinear strict feedback systems with unmeasured states. For making the virtual and actual controls are optimized solution of the corresponding subsystem, RL of observer-critic-actor architecture based on FLS approximations is constructed in every backstepping step, where the observer aims to estimate the unmeasurable states, and the critic and actor aim to evaluate control performance and perform control behavior, respectively. In the proposed optimized control, on the one hand, the state observer method can avoid to require the design constants making its characteristic polynomial Hurwitz, which is universally demanded in the existing observer methods; on the other hand, the RL is significantly simple in algorithm, because the critic and actor training laws are derived from the negative gradient of a simple positive function, which is produced from the partial derivative of Hamilton–Jacobi–Bellman (HJB) equation, instead of the square of approximated HJB equation. Therefore this optimized scheme can be more easily applied and widely extended. Finally, from two aspects of theory and simulation, it is demonstrated that the desired objective can be fulfilled.

In this article, an optimized leader-follower consensus control is proposed for a class of second-order unknown nonlinear dynamical multiagent system. Different with the first-order multiagent consensus, the second-order case needs to achieve the agreement not only on position but also on velocity, therefore this optimized control is more challenging and interesting. To derive the control, reinforcement learning (RL) can be a natural consideration because it can overcome the difficulty of solving the Hamilton–Jacobi–Bellman (HJB) equation. To implement RL, it needs to iterate both adaptive critic and actor networks each other. However, if this optimized control learns RL from most existing optimal methods that derives the critic and actor adaptive laws from the negative gradient of square of the approximating function of the HJB equation, this control algorithm will be very intricate, because the HJB equation correlated to a second-order nonlinear multiagent system will become very complex due to strong state coupling and nonlinearity. In this work, since the two RL adaptive laws are derived via implementing the gradient descent method to a simple positive function, which is obtained on the basis of a partial derivative of the HJB equation, this optimized control is significantly simple. Meanwhile, it not merely can avoid the requirement of known dynamic acknowledge, but also can release the condition of persistent excitation, which is demanded in most RL optimization methods for training the adaptive parameter more sufficiently. Finally, the proposed control is demonstrated by both theory and computer simulation.