While convolutional neural networks (CNNs) have been successfully applied to many challenging classification applications, they typically require large datasets for training. When the availability of labeled data is limited, data augmentation is a critical preprocessing step for CNNs. However, data augmentation for wearable sensor data has not been deeply investigated yet.In this paper, various data augmentation methods for wearable sensor data are proposed. The proposed methods and CNNs are applied to the classification of the motor state of Parkinson's Disease patients, which is challenging due to small dataset size, noisy labels, and large intra-class variability. Appropriate augmentation improves the classification performance from 77.54% to 86.88%.

Contact CEH NORA team at noraceh@ceh.ac.ukThe NERC and CEH trademarks and logos ('the Trademarks') are registered trademarks of NERC in the UK and other countries, and may not be used without the prior written consent of the Trademark owner.

patients with advanced parkinson's disease regularly experience unstable motor states. objective and reliable monitoring of these fluctuations is an unmet need. We used deep learning to classify motion data from a single wrist-worn iMU sensor recording in unscripted environments. for validation purposes, patients were accompanied by a movement disorder expert, and their motor state was passively evaluated every minute. We acquired a dataset of 8,661 minutes of IMU data from 30 patients, with annotations about the motor state (off,on, DYSKinetic) based on MDS-UpDRS global bradykinesia item and the AIMS upper limb dyskinesia item. Using a 1-minute window size as an input for a convolutional neural network trained on data from a subset of patients, we achieved a three-class balanced accuracy of 0.654 on data from previously unseen subjects. This corresponds to detecting the OFF, ON, or DYSKINETIC motor state at a sensitivity/specificity of 0.64/0.89, 0.67/0.67 and 0.64/0.89, respectively. on average, the model outputs were highly correlated with the annotation on a per subject scale (r = 0.83/0.84; p < 0.0001), and sustained so for the highly resolved time windows of 1 minute (r = 0.64/0.70; p < 0.0001). Thus, we demonstrate the feasibility of long-term motor-state detection in a free-living setting with deep learning using motion data from a single iMU. Parkinson's disease (PD) is characterized by slowness of movement, decremented small amplitude, and loss of movement spontaneity that are dramatically relieved when dopamine is orally restituted 1. Due to the pharmacokinetic properties of the main medication, i.e. L-DOPA, motor fluctuations may occur and complicate the symptomatic treatment 2-4. Troughs in dopaminergic therapy are accompanied by parkinsonistic phases, so-called OFF-states, while peaks can lead to phases with excessive (hyperkinetic) spontaneous movements, the dyskinetic (DYS), or ON + motor state 5. Ideally, patients with PD (PwP) experience neither OFF nor dyskinetic motor states but maintain a state resembling normal motor function, i.e. the ON state. These motor fluctuations are a major limiting factor for patients' quality of life, especially in later disease stages 6. Consequently, therapeutic innovations have to demonstrate superiority in terms of their ability to reduce motor fluctuations in order to be licensed by health agencies e.g. 7-9. The current standard for assessing motor fluctuations relies on patient self-reporting in the form of diaries (e.g. 10), or expert ratings using standardized scales (e.g. 11 , see 12 for a review). Both approaches have their merits. But they are prone to rater bias and placebo effects, and they can capture the motor state only with coarse temporal resolutions 13,14. In the past, clinically relevant features in motion data has been extracted to quantify motor states of PwP over long periods of time in free-living setups 15-17. Those approaches were not capable of a dynamic detection of typical motion patterns and failed, for example, when the sensor data we...

Neuromelanin (NM) is a black pigment located in the brain in substantia nigra pars compacta (SN) and locus coeruleus. Its loss is directly connected to the loss of nerve cells in this part of the brain, which plays a role in Parkinson’s Disease. Magnetic resonance imaging (MRI) is an ideal tool to monitor the amount of NM in the brain in vivo. The aim of the study was the development of tools and methodology for the quantification of NM in a special neuromelanin-sensitive MRI images. The first approach was done by creating regions of interest, corresponding to the anatomical position of SN based on an anatomical atlas and determining signal intensity threshold. By linking the anatomical and signal intensity information, we were able to segment the SN. As a second approach, the neural network U-Net was used for the segmentation of SN. Subsequently, the volume characterizing the amount of NM in the SN region was calculated. To verify the method and the assumptions, data available from various patient groups were correlated. The main benefit of this approach is the observer-independency of quantification and facilitation of the image processing process and subsequent quantification compared to the manual approach. It is ideal for automatic processing many image sets in one batch.

Abstract-Gaussian Processes (GPs) are gaining increasing popularity due to their expressive power for learning the dynamics of non-linear time series data, e.g. for human motion prediction. However, so far they are restricted to Euclidean space: input data such as position and velocity need to be Euclidean. In this paper, we examine GPs over time series of 6D rigid body motions including large rotations. As the use of Euler angles with large rotations results in inaccurate predictions, we present an extension of the valid input data to quaternions H and dual quaternions H D . The quality of a GP prediction over unit quaternions is compared with GP prediction over Euler angles. The results are evaluated based on experimental data from a quadrotor and in a learning task of a collision free 6D motion trajectory incorporating large rotations based on artificial data from a motion planner.

We report extent and rate of land use/land cover change in a forest–grassland mosaic of Rio Grande do Sul, Brazil, during a recent period of increasing conflicts between native habitat protection and conversion. The area is part of the Atlantic rain forest biome, a Global Biodiversity Hotspot. Analyzing Landsat and Google Earth imagery, and calculating an effective conservation risk index (ECRI) as ratio of converted to remnant area, we specifically compared the effectiveness of designated fully protected areas (FP‐PAs) and Sustainable Use areas (SU‐PAs) in preventing conversion of native forest and grassland habitats for agri‐ and silviculture, relative to areas outside. Grassland area decreased by 17%, corresponding to a net loss of 59,671 ha, in the entire area. Forest gains exceeded losses, and ECRI was zero inside Full Protection PAs. Non‐native tree plantation area increased by 94% over the entire study area; cropland increased by 7%. Conversion for silviculture predominated outside the designated PAs and conversion for agriculture predominated inside the designated PAs. ECRI was generally higher for grassland than forest, and in SU‐PAs, grassland ECRI was several times higher than in areas without any protection status. These developments are in stark contrast to the high standards of the Brazilian protected area system and corresponding International Union for Conservation of Nature and Natural Resources categories. They are due to protracted regularization of land conversion and establishment of designated protection areas. Furthermore, they reveal the dilemma of previously managed grasslands in strictly protected areas being eventually succeeded by forest, and the hazards of broad interpretation of the term “sustainable development”.

A new real time 3D-dat,i and color acquisition system for the mid-distance range (0.5-3m) employing color-encoded structured light is presented. The system is integrated using low-cost components and allows the combination of 2D and 3D image processing algorithms since it provides a 2D-color image of the scene in addition to the range data. Its design is focused on enabling the system to operate reliably in real-world scenarios, i.e. in uncontrolled environments and with arbitrary scenes. To that end novel approaches for encoding and recognizing the projected light are usecl which make the system practically independent of intrinsic object colors and minimize the influence of the ambient light conditions. Experimental results obtained for the first applications of the system, 3D face and gesture recognition tasks, are presented.

During the past decades, agro‐biodiversity has markedly declined and some species are close to extinction in large parts of Europe. Reintroduction of rare arable plant species in suitable habitats could counteract this negative trend. The study investigates optimal sowing rates of three endangered species (Legousia speculum‐veneris (L.) Chaix, Consolida regalis Gray, and Lithospermum arvense L.), in terms of establishment success, seed production, and crop yield losses.A field experiment with partial additive design was performed in an organically managed winter rye stand with study species added in ten sowing rates of 5–10,000 seeds m−2. They were sown as a single species or as a three‐species mixture (pure vs. mixed sowing) and with vs. without removal of spontaneous weeds. Winter rye was sown at a fixed rate of 350 grains m−2. Performance of the study species was assessed as plant establishment and seed production. Crop response was determined as grain yield.Plant numbers and seed production were significantly affected by the sowing rate, but not by sowing type (pure vs. mixed sowing of the three study species), and weed removal. All rare arable plant species established and reproduced at sowing rates >25 seeds m−2, with best performance of L. speculum‐veneris. Negative density effects occurred to some extent for plant establishment and more markedly for seed production.The impact of the three study species on crop yield followed sigmoidal functions. Depending on the species, a yield loss of 10% occurred at >100 seeds m−2. Synthesis and applications: The study shows that reintroduction of rare arable plants by seed transfer is a suitable method to establish them on extensively managed fields, for example, in organic farms with low nutrient level and without mechanical weed control. Sowing rates of 100 seeds m−2 for C. regalis and L. arvense, and 50 seeds m−2 for L. speculum‐veneris are recommended, to achieve successful establishment with negligible crop yield losses.