A method for predicting variables of interest related to a system includes collecting one or more sensor streams over a time period from sensors in the system and generating one or more anomaly streams for the time period based on the sensor streams. Values for variables of interest for the time period are determined based on the sensor streams and the anomaly streams. Next, a time-series predictive algorithm is applied to the (i) the sensor streams, (ii) the anomaly streams, and (iii) the values for the variables of interest to generate a model for predicting new values for the variables of interest. The model may then be used to predict values for the variables of interest at a time within a new time period based on one or more new sensor streams.

In a method for the recognition of an object by means of a radar sensor system, a primary radar signal is transmitted into an observation space, a secondary radar signal reflected by the object is received, a Micro-Doppler spectrogram of the secondary radar signal is generated, and at least one periodicity quantity relating to an at least essentially periodic motion of a part of the object is determined based on the Micro-Doppler spectrogram. The determining of the at least one periodicity quantity includes the following steps: (i) determining the course of at least one periodic signal component corresponding to an at least essentially periodic pattern of the Micro-Doppler spectrogram, (ii) fitting a smoothed curve to the periodic signal component, (iii) determining the positions of a plurality of peaks and/or valleys of the smoothed curve, and (iv) determining the periodicity quantity based on the determined positions of peaks and/or valleys.

A wearable device capable of recognizing doze-off stage including a processor and an electrocardiogram sensor is provided. The processor trains a neural network module. The processor is coupled to the electrocardiogram sensor. The electrocardiogram sensor is configured to generate an electrocardiogram signal. The processor performs a heart rate variability analysis operation and a R-wave amplitude analysis operation to analyze a heart beat interval variation of the electrocardiogram signal, so as to generate a plurality of characteristic values. The processor utilizes the trained neural network module to perform a doze-off stage recognition operation according to the characteristic values, so as to obtain a doze-off stage recognition result. In addition, a recognition method is also provided.

A method for suppressing sensor noise in a spatially oversampled sensor array includes receiving spatially oversampled multi-channel sensor data from a region of interest and building a spatial model from the data for essential spatial degrees of freedom. The method further includes decomposing the data into the underlying spatial model to obtain associated amplitude components containing a mixture of original temporal waveforms of the data and, for each channel of the multi-channel sensor, estimating time-domain amplitude components using cross-validation. Next, for each channel, based on the estimated time-domain amplitude components, sensor noise and/or artifacts for that channel are identified. Finally, for each channel, the identified sensor noise and/or artifacts can be suppressed from the data.

Techniques, including systems and methods for monitoring a rotating equipment, are provided. A sensor that is in proximity of the rotating equipment senses vibrations of the rotating equipment. The sensor generates a digital signal corresponding to the vibrations of the rotating equipment and transmits the digital signal over a communication network. A server receives the digital signal and pre-processes the digital signal using ensemble empirical mean decomposition (EEMD) technique. The server processes the digital signal using wavelet neural network (WNN) to detect faults in the rotating equipment. Further, the server processes the digital signal using the wavelet neural network to predict remaining useful life (RUL) of the rotating equipment.

A system and method for restricting the number of layout patterns by pattern identification, matching and classification, includes decomposing the pattern windows into a low frequency component and a high frequency component using a wavelet analysis for an integrated circuit layout having a plurality of pattern windows. Using the low frequency component as an approximation, a plurality of moments is computed for each pattern window. The pattern windows are classified using a distance computation for respective moments of the pattern windows by comparing the distance computation to an error value to determine similarities between the pattern windows.

For a signal waveform to be processed, the continuous wavelet transform is performed with various scale factors, and a wavelet coefficient at each point in time is calculated. On an image showing the strength of the wavelet coefficient with respect to the scale factor and time, ridge lines are detected, and based on these ridge lines, positive and negative peak candidates are extracted, after which an error in the position and width of the peak due to the influence of a neighboring peak is corrected. Subsequently, the degree of non-symmetry of the peak shape or other features are examined to remove false negative peaks due to negative peak artifacts. Subsequently, a true peak cluster, a false peak cluster resulting from the removal of high-frequency components of a high-frequency noise or other causes, and other kinds of peaks are identified, and the obtained result is used to remove false peaks.

An apparatus and method for detecting a bio-signal feature are provided. The apparatus according to one aspect may include: a bio-signal acquirer configured to acquire a bio-signal; and a processor configured to generate an envelope signal of the bio-signal, and detect at least one feature of the bio-signal based on a difference between the envelope signal and the bio-signal.

An apparatus and methods for processing monitored biosignals are provided that are particularly suited for reducing noise and artifacts in continuously monitored quasi-periodic biosignals without prior knowledge of the noise distribution. The framework trains a subspace manifold with reference signals. Subsequent signals are successively projected onto the trained manifold and adjusted based on the nearest neighbors of the state of the sample being projected as well as the state of the sample at the previous time point. A denoised or modified output is obtained with inverse mapping. The reference signals may optionally be labeled during manifold training with clinical events/variables or measurable diseases/injuries from a library of relevant labels. During reconstruction, the label of the estimated state in the manifold can be obtained from the label corresponding to the estimated state.

A method for detecting and predicting an epileptic seizure. The method includes preparing a plurality of electrical signals, extracting a plurality of patterns from the plurality of electrical signals, extracting a plurality of features from the plurality of electrical signals by applying the plurality of patterns on the plurality of electrical signals, optimizing the plurality of patterns and the plurality of features, and classifying each of the plurality of electrical signals in a plurality of classes by applying a plurality of classifiers on the plurality of features. The plurality of electrical signals include a plurality of samples. The plurality of classes include a seizure class and a non-seizure class, and the plurality of classifiers include a plurality of cascaded AdaBoost classifiers.