A Novel Loss Recovery and Tracking Scheme for Maneuvering Target in Hybrid WSNs.

Tracking a mobile target, which aims to timely monitor the invasion of specific target, is one of the most prominent applications in wireless sensor networks (WSNs). Traditional tracking methods in WSNs only based on static sensor nodes (SNs) have several critical problems. For example, to void the loss of mobile target, many SNs must be active to track the target in all possible directions, resulting in excessive energy consumption. Additionally, when entering coverage holes in the monitoring area, the mobile target may be missing and then its state is unknown during this period. To tackle these problems, in this paper, a few mobile sensor nodes (MNs) are introduced to cooperate with SNs to form a hybrid WSN due to their stronger abilities and less constrained energy. Then, we propose a valid target tracking scheme for hybrid WSNs to dynamically schedule the MNs and SNs. Moreover, a novel loss recovery mechanism is proposed to find the lost target and recover the tracking with fewer SNs awakened. Furthermore, to improve the robustness and accuracy of the recovery mechanism, an adaptive unscented Kalman filter (AUKF) algorithm is raised to dynamically adjust the process noise covariance. Simulation results demonstrate that our tracking scheme for maneuvering target in hybrid WSNs can not only track the target effectively even if the target is lost but also maintain an excellent accuracy and robustness with fewer activated nodes.

Adaptively Customizing Activation Functions for Various Layers.

To enhance the nonlinearity of neural networks and increase their mapping abilities between the inputs and response variables, activation functions play a crucial role to model more complex relationships and patterns in the data. In this work, a novel methodology is proposed to adaptively customize activation functions only by adding very few parameters to the traditional activation functions such as Sigmoid, Tanh, and rectified linear unit (ReLU). To verify the effectiveness of the proposed methodology, some theoretical and experimental analysis on accelerating the convergence and improving the performance is presented, and a series of experiments are conducted based on various network models (such as AlexNet, VggNet, GoogLeNet, ResNet and DenseNet), and various datasets (such as CIFAR10, CIFAR100, miniImageNet, PASCAL VOC, and COCO). To further verify the validity and suitability in various optimization strategies and usage scenarios, some comparison experiments are also implemented among different optimization strategies (such as SGD, Momentum, AdaGrad, AdaDelta, and ADAM) and different recognition tasks such as classification and detection. The results show that the proposed methodology is very simple but with significant performance in convergence speed, precision, and generalization, and it can surpass other popular methods such as ReLU and adaptive functions such as Swish in almost all experiments in terms of overall performance.