Learning Background-Suppressed Dual-Regression Correlation Filters for Visual Tracking.

The discriminative correlation filter (DCF)-based tracking method has shown good accuracy and efficiency in visual tracking. However, the periodic assumption of sample space causes unwanted boundary effects, restricting the tracker's ability to distinguish between the target and background. Additionally, in the real tracking environment, interference factors such as occlusion, background clutter, and illumination changes cause response aberration and, thus, tracking failure. To address these issues, this work proposed a novel tracking method named the background-suppressed dual-regression correlation filter (BSDCF) for visual tracking. First, we utilize the background-suppressed function to crop out the target features from the global features. In the training step, while introducing the spatial regularity constraint and background response suppression regularization, we construct a dual regression structure to train the target and global filters separately. The aim is to exploit the difference between the output response maps for mutual constraint to highlight the target and suppress the background interference. Furthermore, in the detection step, the global response can be enhanced by a weighted fusion of the target response to further improve the tracking performance in complex scenes. Finally, extensive experiments are conducted on three public benchmarks (including OTB100, TC128, and UAVDT), and the experimental results indicate that the proposed BSDCF tracker achieves tracking performance comparable to many state-of-the-art (SOTA) trackers in a variety of complex situations.

Lightweight 3D Dense Autoencoder Network for Hyperspectral Remote Sensing Image Classification.

The lack of labeled training samples restricts the improvement of Hyperspectral Remote Sensing Image (HRSI) classification accuracy based on deep learning methods. In order to improve the HRSI classification accuracy when there are few training samples, a Lightweight 3D Dense Autoencoder Network (L3DDAN) is proposed. Structurally, the L3DDAN is designed as a stacked autoencoder which consists of an encoder and a decoder. The encoder is a hybrid combination of 3D convolutional operations and 3D dense block for extracting deep features from raw data. The decoder composed of 3D deconvolution operations is designed to reconstruct data. The L3DDAN is trained by unsupervised learning without labeled samples and supervised learning with a small number of labeled samples, successively. The network composed of the fine-tuned encoder and trained classifier is used for classification tasks. The extensive comparative experiments on three benchmark HRSI datasets demonstrate that the proposed framework with fewer trainable parameters can maintain superior performance to the other eight state-of-the-art algorithms when there are only a few training samples. The proposed L3DDAN can be applied to HRSI classification tasks, such as vegetation classification. Future work mainly focuses on training time reduction and applications on more real-world datasets.

A Positioning and Navigation Method Combining Multimotion Features Dead Reckoning with Acoustic Localization.

Accurate location information can offer huge commercial and social value and has become a key research topic. Acoustic-based positioning has high positioning accuracy, although some anomalies that affect the positioning performance arise. Inertia-assisted positioning has excellent autonomous characteristics, but its localization errors accumulate over time. To address these issues, we propose a novel positioning navigation system that integrates acoustic estimation and dead reckoning with a novel step-length model. First, the features that include acceleration peak-to-valley amplitude difference, walk frequency, variance of acceleration, mean acceleration, peak median, and valley median are extracted from the collected motion data. The previous three steps and the maximum and minimum values of the acceleration measurement at the current step are extracted to predict step length. Then, the LASSO regularization spatial constraint under the extracted features optimizes and solves for the accurate step length. The acoustic estimation is determined by a hybrid CHAN-Taylor algorithm. Finally, the location is determined using an extended Kalman filter (EKF) merged with the improved pedestrian dead reckoning (PDR) estimation and acoustic estimation. We conducted some comparative experiments in two different scenarios using two heterogeneous devices. The experimental results show that the proposed fusion positioning navigation method achieves 8~56.28 cm localization accuracy. The proposed method can significantly migrate the cumulative error of PDR and high-robustness localization under different experimental conditions.

Deep Neural Network-Based Fusion Localization Using Smartphones.

Indoor location-based services (LBS) have tremendous practical and social value in intelligent life due to the pervasiveness of smartphones. The magnetic field-based localization method has been an interesting research hotspot because of its temporal stability, ubiquitousness, infrastructure-free nature, and good compatibility with smartphones. However, utilizing discrete magnetic signals may result in ambiguous localization features caused by random noise and similar magnetic signals in complex symmetric and large-scale indoor environments. To address this issue, we propose a deep neural network-based fusion indoor localization system that integrates magnetic and pedestrian dead reckoning (PDR). In this system, we first propose a ResNet-GRU-LSTM neural network model to achieve magnetic localization more accurately. Afterward, we put forward a multifeatured-driven step length estimation. A hierarchy GRU (H-GRU) neural network model is proposed, and a multidimensional dataset using acceleration and a gyroscope is constructed to extract more valid characteristics. Finally, more reliable and accurate pedestrian localization can be achieved under the particle filter framework. Experiments were conducted at two trial sites with two pedestrians and four smartphones. Results demonstrate that the proposed system achieves better accuracy and robustness than other traditional localization algorithms. Moreover, the proposed system exhibits good generality and practicality in real-time localization with low cost and low computational complexity.

A Low-Cost and Efficient Indoor Fusion Localization Method.

Accurate indoor location information has considerable social and economic value in applications, such as pedestrian heatmapping and indoor navigation. Ultrasonic-based approaches have received significant attention mainly since they have advantages in terms of positioning with temporal correlation. However, it is a great challenge to gain accurate indoor localization due to complex indoor environments such as non-uniform indoor facilities. To address this problem, we propose a fusion localization method in the indoor environment that integrates the localization information of inertial sensors and acoustic signals. Meanwhile, the threshold scheme is used to eliminate outliers during the positioning process. In this paper, the estimated location is fused by the adaptive distance weight for the time difference of arrival (TDOA) estimation and improved pedestrian dead reckoning (PDR) estimation. Three experimental scenes have been developed. The experimental results demonstrate that the proposed method has higher localization accuracy in determining the pedestrian location than the state-of-the-art methods. It resolves the problem of outliers in indoor acoustic signal localization and cumulative errors in inertial sensors. The proposed method achieves better performance in the trade-off between localization accuracy and low cost.

An Unambiguous Synchronization Scheme for GNSS BOC Signals Based on Reconstructed Correlation Function.

Binary offset carrier (BOC) modulation is a new modulation method that has been gradually applied to the Global Satellite Navigation System (GNSS) in recent years. However, due to the multi-peaks in its auto-correlation function (ACF), it will incur a false lock and generate synchronization ambiguous potentially. In this paper, an unambiguous synchronization method based on a reconstructed correlation function is proposed to solve the ambiguity problem. First, through the shape code vector constructed in this paper, the general cross-correlation function (CCF) expression of the BOC modulated signal will be obtained. Based on the features of the signal correlation function, it is decomposed into a matrix form of trigonometric functions. Then, it generates two local signal waves using a specific method, then the proposed method is implemented to obtain a no-side-peak correlation function by reconstructing the cross-correlation between the received signal and the two local signals. Simulations showed that it fully eliminates the side-peak threat and significantly removes the ambiguity during the synchronization of the BOC signals. This paper also gives the improved structure of acquisition and tracking. The detailed theoretical deduction of detection probability and code tracking error is demonstrated, and the corresponding phase discrimination function is given. In terms of de-blurring ability and detection probability performance, the proposed method outperformed other conventional approaches. The tracking performance was superior to the comparison methods and the phase discrimination curve only had a zero-crossing, which successfully removed the false lock points. In addition, in multipath mitigation, it outperformed the ACF of the BOC signal, and performs as well as the autocorrelation side-peak cancellation technique (ASPeCT) for BOC(kn,n) signals.

Frequency-Locked Detector Threshold Setting Criteria Based on Mean-Time-To-Lose-Lock (MTLL) for GPS Receivers.

Frequency-locked detector (FLD) has been widely utilized in tracking loops of Global Positioning System (GPS) receivers to indicate their locking status. The relation between FLD and lock status has been seldom discussed. The traditional PLL experience is not suitable for FLL. In this paper, the threshold setting criteria for frequency-locked detector in the GPS receiver has been proposed by analyzing statistical characteristic of FLD output. The approximate probability distribution of frequency-locked detector is theoretically derived by using a statistical approach, which reveals the relationship between probabilities of frequency-locked detector and the carrier-to-noise ratio (C/N0) of the received GPS signal. The relationship among mean-time-to-lose-lock (MTLL), detection threshold and lock probability related to C/N0 can be further discovered by utilizing this probability. Therefore, a theoretical basis for threshold setting criteria in frequency locked loops for GPS receivers is provided based on mean-time-to-lose-lock analysis.

Landslide Displacement Prediction Based on Time Series Analysis and Double-BiLSTM Model.

In recent years, machine learning models facilitated notable performance improvement in landslide displacement prediction. However, most existing prediction models which ignore landslide data at each time can provide a different value and meaning. To analyze and predict landslide displacement better, we propose a dynamic landslide displacement prediction model based on time series analysis and a double-bidirectional long short term memory (Double-BiLSTM) model. First, the cumulative landslide displacement is decomposed into trend and periodic displacement components according to time series analysis via the exponentially weighted moving average (EWMA) method. We consider that trend displacement is mainly influenced by landslide factors, and we apply a BiLSTM model to predict landslide trend displacement. This paper analyzes the internal relationship between rainfall, reservoir level and landslide periodic displacement. We adopt the maximum information coefficient (MIC) method to calculate the correlation between influencing factors and periodic displacement. We employ the BiLSTM model for periodic displacement prediction. Finally, the model is validated against data pertaining to the Baishuihe landslide in the Three Gorges, China. The experimental results and evaluation indicators demonstrate that this method achieves a better prediction performance than the classical prediction methods, and landslide displacement can be effectively predicted.

A study on skin tumor classification based on dense convolutional networks with fused metadata.

Skin cancer is the most common cause of death in humans. Statistics show that competent dermatologists have a diagnostic accuracy rate of less than 80%, while inexperienced dermatologists have a diagnostic accuracy rate of less than 60%. The higher rate of misdiagnosis will cause many patients to miss the most effective treatment window, risking the patients' life safety. However, the majority of the current study of neural network-based skin cancer diagnosis remains at the image level without patient clinical data. A deep convolutional network incorporating clinical patient metadata of skin cancer is presented to realize the classification model of skin cancer in order to further increase the accuracy of skin cancer diagnosis. There are three basic steps in the approach. First, the high-level features (edge features, color features, texture features, form features, etc.). Implied by the image were retrieved using the pre-trained DenseNet-169 model on the ImageNet dataset. Second, the MetaNet module is introduced, which uses metadata to control a certain portion of each feature channel in the DenseNet-169 network in order to produce weighted features. The MetaBlock module was added at the same time to improve the features retrieved from photos using metadata, choosing the most pertinent characteristics in accordance with the metadata data. The features of the MetaNet and MetaBlock modules were finally combined to create the MD-Net module, which was then used as input into the classifier to get the classification results for skin cancers. On the PAD-UFES-20 and ISIC 2019 datasets, the suggested methodology was assessed. The DenseNet-169 network model combined with this module, according to experimental data, obtains 81.4% in the balancing accuracy index, and its diagnostic accuracy is up between 8% and 15.6% compared to earlier efforts. Additionally, it solves the problem of actinic keratosis and poorly classified skin fibromas.