Cognitive Computation and Systems最新文献

To address the problems that the UAV (Unmanned Aerial Vehicle) is vulnerable to distance limitation and environmental interference when tracking and landing on a moving platform autonomously, the accuracy of position estimation relying only on visual odometry in the point-featureless environment is insufficient, and the traditional linear path planning solvers and controllers cannot meet the fast and safe requirements under the non-linear strong coupling characteristics of the cooperative landing system, an nonlinear model predictive control (NMPC)-based multi-sensor fusion method for autonomous landing of UAVs on motion platforms is proposed. The UAV combines the position information obtained by the RTK-GPS and the image information obtained by the camera and uses the special identification codes placed in the landing area of the UAV to carry out cooperative planning and navigation while using UKF (Unscented Kalman Filter) to estimate the position of the moving platform and using the interference-resistant NMPC algorithm to optimise the UAV tracking trajectory based on the precise positioning of the two platforms to achieve the autonomous landing control of the UAV. The simulation and practical experimental results show the feasibility and effectiveness of the proposed algorithm and the autonomous landing control method and provide an effective solution for the autonomous landing of quadrotors on arbitrarily moving platforms.

Existing out-of-distribution detection models rely on the prediction of a single classifier and are sensitive to classifier bias, making it difficult to discriminate similar feature out-of-distribution data. This article proposed a multi-classifier-based model and two strategies to enhance the performance of the model. The model first trains several different base classifiers and obtains the predictions of the test data on each base classifier, then uses cross-entropy to calculate the dispersion between these predictions, and finally uses the dispersion as a metric to identify the out-of-distribution data. A large scatter implies inconsistency in the predictions of the base classifier, and the greater the probability of belonging to the out-of-distribution data. The first strategy is applied in the training process of the model to increase the difference between base classifiers by using various scales of Label smoothing regularisation. The second strategy is applied to the inference process of the model by changing the mean and variance of the activations in the neural network to perturb the inference results of the test data. These two strategies can effectively amplify the discrepancy in the dispersion of the in-distribution and out-of-distribution data. The experimental results show that the method in this article can effectively improve the performance of the model in the detection of different types of out-of-distribution data, improve the robustness of deep neural networks (DNN) in the face of unknown classes, and promote the application of DNN in systems and engineering with high security requirements.

The popularity of smart metres has brought a huge amount of demand-side data, which provides important information for the demand response of the power sector, to guide practitioners to understand the customers' electricity usage behaviours and patterns. Clustering analysis of customers' daily load data is an important tool for mining users' consumption habits and achieve non-fixed market segmentation. Since the load data is time series, it is inappropriate to perform clustering directly without extracting targeted features. Therefore, according to the shape features of the daily load curve, a shape-based clustering algorithm called BDKM is proposed. The algorithm first uses the B-splines regression to fit the time series data to extract morphological features, and then the objects are segmented based on the dynamic time warping distance by clustering. Finally, the real world daily customers' load data is used to prove the effectiveness of the proposed algorithm based on B-splines regression.

Traumatic brain injury (TBI) can affect normal brain function and may be caused by a vehicle accident, falling, and so on. The purpose of this survey is to provide clear knowledge of TBI, the causes of TBI, the impacts of TBI, and the role of family members and friends in recovery. TBI affects the daily life of the patients, both physically and mentally. After TBI, the patients may experience many emotional and behavioural changes because of a lack of certain brain functions. These changes affect their personal and social relationships. On the other hand, these changes depend on the severity of the TBI (i.e. mild, moderate, or severe), which is measured using the Glasgow coma score. Generally, three processes are used for emotion recognition: preprocessing, feature extraction, and emotion recognition. Preprocessing is performed for landmark detection and pose normalisation, which improves the performance of emotion detection. Feature extraction and emotion recognition are performed by various deep learning techniques, such as convolution neural networks and long short-term memory. These techniques recognise the behavioural and emotional changes (depression, anxiety, anger, personality changes etc.) of TBI patients using facial expressions. Family members and friends play an important role in TBI patients' recovery, the extent of which is based on the severity of the TBI. The care of family members and friends leads to quick recovery and rehabilitation of patients from TBI. Finally, testing is performed using Computed Tomography images, Magnetic Resonance Imaging images, Electroencephalography signals, and patient demographics, which together show that the deep learning methods achieve better performance in terms of accuracy, precision, recall, and F-measure in recognising emotional and behavioural changes after TBI. The authors conclude with a summary of the future of emotional and behavioural change prediction methods for TBI patients.

Accurate detection of pedestrians and vehicles on the road is an important content in autonomous driving technology. In this article, a method to optimise the object detection network using the channel attention mechanism is proposed. In general, small object detection problems and difficult sample detection problems in object detection tasks can be solved by using feature pyramids. Different from building a feature pyramid, the authors did not make extensive changes to the network, but used the channel attention mechanism to dynamically adjust the output of a layer during the feature extraction process, allowing each neuron to adjust its receptive field size adaptively according to multiple scales of the input information, so that the network pays attention to the extraction of important features, especially the features of small objects and difficult samples. In order to evaluate the performance of the proposed method, experiments were conducted on standard benchmark data sets. It has been observed that the proposed method is superior to the original object detection network in terms of the detection accuracy of pedestrians and vehicles, especially the detection of small objects.