首页>
外文学位
>Using Fuzzy Inference in Psychophysical Detection Experiments to Separate Hits, False Positives and Guesses; And Using Wavelet Decomposition to Detect Periodic Signals in Head Accelerometry Measures; Both in Posturally Perturbed Standing Subjects.
【24h】
Using Fuzzy Inference in Psychophysical Detection Experiments to Separate Hits, False Positives and Guesses; And Using Wavelet Decomposition to Detect Periodic Signals in Head Accelerometry Measures; Both in Posturally Perturbed Standing Subjects.
In a 2-Alternative Forced Choice Interval task (2AFCi), a standing subject was required to press a button once or twice to signal in which of two 4 to 6s sequential intervals that (s)he thought that a short ≤16 mm postural perturbation had occurred. The perturbation might or might not result in transient changes in the subject's Anterior-Posterior Center of Pressure (APCOP) or in other measures. We used fuzzy inference to explore whether the correctness of a subject's stimulus detection could be gleaned from analyzing changes in one of more metrics related to changes in APCOP. Also, distinguishing guesses from correct responses was a critical issue in any 2AFC experiments in the SLIP-FALLS Lab. Biomechanical and psychophysical data were used to design a prediction model based on fuzzy inference that was able to discriminate correct responses from guesses.;In our second model, psychophysical movement detection strategies of standing blindfolded subjects were categorized by analyzing the changes in their head acceleration data that correlated with their ability to correctly detect small translational perturbations of the movement platform. The time-series head acceleration data provided a measure of postural stability and a clear indication of postural control responses that could be directly correlated with the stimulus. Studying the biomechanical and psychophysical responses together enabled us to discriminate correct responses from guesses. To compare the biomechanical response to psychophysical response, it was necessary to find any abnormality in the biomechanical response (head acceleration) that related to platform movement. For this purpose, a novel method based on Adaptive Neural Fuzzy Inference Systems (ANFIS) was applied to identify the abnormality present in the head acceleration data. Consequently, a fuzzy logic base model was designed to take head acceleration time series data and the subject's psychophysical responses as inputs for predicting perturbation detection and distinguishing guesses from true hits. The accuracy of the designed head-acceleration-based model (87%) was smaller than the accuracy of the APCOP-based-model (95%), but its accuracy is still remarkable. Our study revealed that a subject's APCOP data was richer input sensory in comparison to the head acceleration data.;A Matlab-based GUI (Graphical User Interface) was created to study the transition of acceleration and jerk of the platform to the subject's head in the 2AFC experiments. Different movement displacements in 2AFC experiments (1mm, 4mm, and 16mm) helped us investigate the frequency dependence of a subject's movement perception. In the 1mm experiment, there were 2 differences between movement and non-movement intervals for the head acceleration and jerk data. A signal with larger amplitude and smaller frequency component was observed in the movement interval in both head acceleration and jerk data. But at 4 and 16 mm we observed a signal with only smaller frequency component during the movement interval. In other words, at 4mm and 16 mm experiments, there is no marked difference in the amplitudes of head acceleration and jerk signals between the movement and non-movement intervals. Our study revealed that there is a positive power law relationship between the length of short anterior translations and system gains in subject's head AP and APCOP. This explains the observed negative power law relationship between the length of short anterior translations and the subjects' peak acceleration thresholds. In other words, with increasing length of the short anterior translations (or the decreasing frequency of the platform acceleration), head AP and APCOP gains increased. This could justify the low PAT values at the 16mm displacement.;The subject's sway and the periodic signal overlap in the frequency domain. Simple band-pass filtering does not highlight well this periodicity signal information. The wavelet transform removed the sway component from head RL acceleration raw data and preserved the periodic 1HZ signal. De-noising was an interesting application of a wavelet transformation. We used the wavelet transform to recover a signal (head RL) from the signal with noise (baseline wandering in the head RL).
展开▼
