A Tracking Algorithm for Sparse and Dynamic Underwater Sensor Networks

Abstract

An underwater sensor network (UWSN) has sparse and dynamic characteristics. In sparse and dynamic UWSNs, the traditional particle filter based on multi-rate consensus/fusion (CF/DPF) has the problems of a slow convergence rate and low filtering accuracy. To solve these problems, a tracking algorithm for sparse and dynamic UWSNs based on particle filter (TASD) is proposed. Firstly, the estimation results of a local particle filter are processed by a weighted average consensus filter (WACF). In this way, the reliability difference of state estimation between nodes in sparse and dynamic UWSN is reasonably eliminated. Secondly, a delayed update mechanism (DUM) is added to WACF, which effectively solves the problem of time synchronization between the two particle filters. Thirdly, under the condition of limited communication energy consumption, an alternating random scheme (ARS) is designed, which optimizes the mean square convergence rate of the fusion particle filter. Simulation results show that the proposed algorithm can be applied to maneuvering target tracking in sparse and dynamic UWSN effectively. Compared with the traditional method, it has higher tracking accuracy and faster convergence speed. The average estimation error of TASD is 91.3% lower than that of CF/DPF, and the weighted consensus tracking error of TASD is reduced by 85.6% compared with CF/DPF.

Multi-Target Tracking in Multi-Static Networks with Autonomous Underwater Vehicles Using a Robust Multi-Sensor Labeled Multi-Bernoulli Filter

“…Here, we define the function ρ(a t , X t ) to ensure that the associated variable a t is admissible, where ρ(a t , X t ) = 1 if a t ∈ A t is subject to a t ( ) ∈ Λ for ∈ L X t and a t ( ) = −1 V for ∈ L t \L X t , and ρ(a t , X t ) = 0 otherwise. As a t ( ) ∈ Λ implies that 16) is equivalent to µ ,θ t ( ) t ( x t , ) in (13) with the substitution of a t ( ) for θ t ( ) concerning ∈ L X t . Additionally, the weights w a t can be represented by (13), w a t in (17) depends on L t instead of L X t , so the weight w a t can generally be interpreted as the probability mass function (pmf) of a t .…”

“…As a t ( ) ∈ Λ implies that 16) is equivalent to µ ,θ t ( ) t ( x t , ) in (13) with the substitution of a t ( ) for θ t ( ) concerning ∈ L X t . Additionally, the weights w a t can be represented by (13), w a t in (17) depends on L t instead of L X t , so the weight w a t can generally be interpreted as the probability mass function (pmf) of a t . Therefore, we can represent the pmf of a t ∈ A t by g(a t ) = w a t , and normalize g(a t ) in that ∑ a t ∈A g(a t ) = 1.…”

“…Technological developments have made small, relatively low-cost underwater intelligent systems, especially autonomous underwater vehicles (AUVs), a reality [12][13][14]. With their mobility and ability to reconfigure locations, AUVs have gained considerable attention as "listeners" in multi-static networks [15,16].…”

DACFL: Dynamic Average Consensus-Based Federated Learning in Decentralized Sensors Network

“…In [50], the l 1 -regularized H ∞ filtering is introduced to solve the estimation problem. In [51], a tracking algorithm for sparse and dynamic underwater sensor networks based on particle filter (TASD) is proposed to improve the slow convergence rate and low filtering accuracy of the traditional particle filter. While for the dynamic average consensus problem regarding the discrete-time reference inputs, reference [12] proposes a class of discrete-time dynamic average consensus algorithms and analyzes their convergence properties.…”

An Improved EfficientNet for Rice Germ Integrity Classification and Recognition

“…Huang et al [32] built their own convolutional neural network to achieve the classification of pineapple quality. Liu et al [33,34] used an optimized neural network to identify spectral peaks and initially determine target locations. Yuan et al [35] used the Inception-v2 [36] network to detect cherries and tomatoes.…”