In perfixming estimation of the state of a nonlinear system, a standard approach that is usually adequate is to llmarizc th algebraic nonlinear measurement equation (that describes th operation of the measurement sensor) and to linearize the nonlinear differential -tion ((het describes th system or plant), es expressed in state variable notation This linearization (het is performed as part of an extended Knlman filter (EKF) algorithmic iqkmentation corresponds to use of the first two terns of t h Taylor series expcusion of th system and measurement nonlinearity as a reasonable approximation It is well documented in the estimation theory literature that just a minor change in mechanization beyond this standard EKF inpkmentation CM result in a significant improvement in EKF performance. Frequently, with a few further iterations (of the masuremnt relinearization process), the resulting intermdiate linearization is greatly improved, resulting in a payoff of offering significantly improved EKF performance while t h penalty incurred is merely the cost of mechanizing a variation of only slightly greater complexity or slightly higher operatiom counts.This work provides the details of how to implement the measurement iteration process, described above, as a software subroutine module for an exoatmospheric random variable (RV) target tracking application. The primary contribution is in providing a new mre computationaliy efficient general method for perfwmiqg measurement iteration (or rellmarization) within the inpkmntation of an EKE The results are illustrated, as used in a radar application of tracking exoatmosphcric RV targets.

Failure detection and redundancy management is discussed for avionics applications of integrated navigation involving coordinated use of multiple simultaneous sensor subsystems such as GPS, JTIDS, TACAN, VOR/DME, ILS, an inertial navigation system (INS), and possibly even Doppler AHRS. A brief high level survey is provided to assess the status of those techniques and methodologies advertized as already available for handling the challenging real-time failure detection, redundancy management, and Kalman filtering aspects of these systems with differing availabilities, differing reliabilities, differing accuracies, and differing information content/sampling rates.Following the status review, a new failure detection/redundancy management approach is developed based on voter/monitoring at both the raw data and at the filtered-data level, as well as using additional inputs from hardware built-in-testing (BIT) and from specialized tests for subsequent failure isolation in the case of ambiguous indications. The technique developed involves use of Gaussian confidence regions to reasonably account for the inherent differences in accuracy between the various sensor subsystems. Online estimates of covariances from the Kalman filter are to be used for this purpose (when available). A technique is provided for quantitatively evaluating both the probability of detecting failed component subsystems and the probability of false alarm to be incurred, which is then to be traded off as the basis for rational selection of the thresholds used in the automated decision process. Moreover, the redundancy management procedure is demonstrated to be amenable to pilot or navigation operator prompting and override, if necessary.A structure to accommodate differing rates of subsystem assessment and tally is developed and alternative designs for navigation architectures are offered based on likely subsystem utilization and newly emerging concepts in decentralized Kalman filtering. Many of the decentralized filtering concepts are only now economically feasible for real-time implementation due to recent availability of commercial parallel processing chips and/or VHSIC compatible systolic array versions of all the requisite algorithms and transformations necessary to support such Kalman filter mechanizations in a few chips.

Using automated imaging technologies, it is now possible to generate previously unprecedented volumes of plankton image data which can be used to study the composition of plankton assemblages. However, the current need to manually classify individual images introduces a bottleneck into processing chains. Although Machine Learning techniques have been used to try and address this issue, past efforts have suffered from accuracy limitations, especially in minority classes. Here we use state-of-the-art methods in Deep Learning to investigate suitable architectures for training an automated plankton classification system which achieves high efficacy for both abundant and rare taxa. We collected live plankton from Station L4 in the Western English Channel and imaged 11,371 particles covering 104 taxonomic groups using the automated plankton imaging system FlowCam. The image set contained a severe class imbalance, with some taxa represented by > 600 images while other, rarer taxa were represented by just 14. We demonstrate that by allowing multiple Deep Learning models to collaborate in a single classification system, classification accuracy improves for minority classes when compared with the best individual model. The top collaborative model achieved a 6% improvement in F1 accuracy over the best individual model, while overall accuracy improved by 3.2%. This resulted in a 97.4% overall accuracy score and a 96.2% F1 macro score on a separate holdout test set containing 104 taxonomic groups. Based on a survey of similar studies in the literature, we believe collaborative deep learning models can significantly improve the accuracy of existing automated plankton classification systems.

No abstract

No abstract