Hyperspectral image (HSI) based detection has attracted considerable attention recently in agriculture, environmental protection and military applications as different wavelengths of light can be advantageously used to discriminate different types of objects. Unfortunately, estimating the background distribution and the detection of interesting local objects is not straightforward, and anomaly detectors may give false alarms. In this paper, a Deep Belief Network (DBN) based anomaly detector is proposed. The high-level features and reconstruction errors are learned through the network in a manner which is not affected by previous background distribution assumption. To reduce contamination by local anomalies, adaptive weights are constructed from reconstruction errors and statistical information. By using the code image which is generated during the inference of DBN and modified by adaptively updated weights, a local Euclidean distance between under test pixels and their neighboring pixels is used to determine the anomaly targets. Experimental results on synthetic and recorded HSI datasets show the performance of proposed method outperforms the classic global Reed-Xiaoli detector (RXD), local RX detector (LRXD) and the-state-of-the-art Collaborative Representation detector (CRD).

Accurate remaining useful life (RUL) prediction of lithium-ion battery is very important for the safety of power system of mobile equipment. At present, most of the currently proposed RUL prediction methods for lithium-ion batteries are limited by the length of the training data, and the RUL is also difficult to predict when the predicted dataset is different from the training dataset in terms of the mapping of operating conditions, and the results obtained are difficult to achieve the desired accuracy. To solve this problem, in this work, a new model is proposed. It is a fusion of the long short-term memory (LSTM) network model based on transfer learning and particle filter (PF) model. LSTM based on transfer learning (TL-LSTM) is used for long-term prediction of the lithium-ion battery capacity, which makes the model have the generalization ability to deal with the RUL prediction under different stress. Since LSTM predicts RUL as a single-point prediction, we introduce a PF model to increase the uncertainty representation of RUL prediction. The predicted value from TL-LSTM model is used as observation for the PF prediction model, and the predicted value is continuously adjusted and updated during the iterative process to achieve the uncertainty representation and improve accuracy. The model is validated with the actual tested battery cycle dataset. The experimental results show that the method is effective for the RUL prediction problem of the mapping under different working conditions, and the relative error is less than 5% when the data length is 50%, which proves the application potential of the method.

Since the control laws require sensor feedback to set the current dynamic state of the unmanned aerial vehicle (UAV), incorrect readings may lead to potentially catastrophic conditions. Thus, automated detection and mitigation of UAV flight data anomaly is an important problem in the aviation domain. However, the conventional anomaly detection algorithms simply detect the outlier points and cannot provide estimated values. To address this challenge, anomaly detection and mitigation algorithm based on online subspace tracking, an online algorithm for flight data anomaly detection and mitigation is proposed. At every time instant, data subspace matrix is used as a meaningful data representation of raw multivariate heterogeneous flight data. Besides, in terms of outlier points, the subspace matrix is tracked with incomplete partial observation. Anomaly score is calculated based on the identification of a change in the underlying subspace. Utilizing the tracked subspace matrix, the detected outlier points are replaced with reasonable recovered estimations, which will help mitigate the influence of anomaly to UAV system control. Experimental results on real UAV flight data demonstrate its ability to maintain high accuracy for anomaly detection and low error for data recovery.

Health status monitoring of flight-critical sensors is crucial to the flight safety of unmanned aerial vehicles (UAVs). While many flight data anomaly detection algorithms have been proposed, most do not consider data source information and cannot identify which data sources contribute most to the anomaly, hindering proper fault mitigation. To address this challenge, a structured sparse subspace learning (SSL) anomaly detection (SSSLAD) algorithm, which reformulates anomaly detection as a structured SSL problem, is proposed. A structured norm is imposed on the projection coefficients matrix to achieve structured sparsity and help identify anomaly sources. Utilizing an efficient optimization method based on Nesterov's method and a subspace tracking approach considering temporal dependence, the computation is efficient. Experiments on real UAV flight data sets illustrate that the proposed SSSLAD algorithm can accurately and quickly detect and identify anomalous sources in flight data, outperforming state of art algorithms, both in terms of accuracy and speed.