A suboptimum detector structure is developed for the detection of weak signals in non-Gaussian noise. In contrast to locally optimum detectors, the suboptimum detector structure is relatively easy to implement and is shown to perform well for a wide range of underlying noise distributions. The robustness of simple limiter detectors is discussed. The extension of the concept of a suboptimum detector to an adaptive detector for operating in an unknown noise environment is also discussed.

A method for the extraction of features for pattern recognition by system identification is presented. A test waveform is associated with a parameterized process model (PM) which is an inverse filter. The structure of the PM corresponds to the redundant information in a waveform, and the parameter values correspond to the discriminatory information. The PM used in this research is a linear predictive system whose parameters are the linear predictive coefficients (LPC's). This technique is applied to feature extraction of electrocardiograms (ECG's) for differential diagnosis. The LPC's are calculated for each ECG and used as a feature vector in a hypergeometric affine N-space spanned by the LPC's. The efficacy of this feature extraction technique is tested by three different perturbation methods, namely noise, matrix distortion, and a newly developed method called directed distortion. Both the Euclidean and Itakura distances between feature vectors in N-space are shown in increase with increasing perturbation of the template waveform. The monotonic behavior of a distance measure is a necessary attribute of a valid feature space. Thus the perturbation analyses done in this research verify the viability of using the parameters of a process model as a feature vector in a pattern recognition scheme.

A locally optimum detector structure is derived for the detection of weak signals in non-Gaussian environments. Optimum performance is obtained by employing a zero-memory nonlinearity prior to the matched filter that would be optimum for detecting the signal were the noise Gaussian. The asymptotic detection performance of the locally optimum detector under non-Gaussian conditions is derived and compared with that for the corresponding detector optimized for operations in Gaussian noise. Numerical results for the asymptotic detection performance are shown for signal detection in noise environments of practical interest.

A feature extraction technique based on a new criterion for "declustering" is presented. Declustering occurs when sample vectors from one pattern class form a densely packed point constellation, or cluster, in feature space while vectors from another class do not form a cluster but instead array themselves as outliers. Features chosen to optimize the declustering criterion enhance class separation and are robust over a wide range of measurement statistics.

A new algorithm is presented for pattern recognition by clustering. The algorithm is called structural editing by a point density function, or STEP. STEP uses a minimum spanning tree to retain the interpoint structure among the elements of an unclassified training set. The tree is pruned or edited to form clusters based on information provided by a point density function (PDF) estimate. STEP has the capability of detecting clusters of arbitrary shape in the presence of intercluster stray points or outliers. A cluster is not required to correspond to a unimodal PDF estimate. Monte Carlo simulations indicate that STEP performs as well as, or better than, a nearest neighbor classifier which requires a classified training set. A new algorithm for recursively constructing the minimum spanning tree is presented which is computationally simpler than conventional algorithms in many practical applications. Results from applying STEP to the mass screening of breast thermograms are discussed.

To reduce distortion, a technique is presented for deconvolving the effect of the distorting system on a desired signal. The technique does not require any prior assumptions about the distorting system other than that it is linear and time-invariant. However, we must assume that the desired signal is a sample function from a cyelostationary (CS) random process with known statistics. Many signals of practical importance can be modelled as CS including PAM, FSK, PWM, etc. The technique utilizes inverse filtering with an estimate of the distorting system function which is obtained exclusively from the observed signal and the known and computed properties of CS processes.

A new feature extraction technique designed specifically for the mass screening problem is presented. It is based on the hypothesis that feature values from the normal class tend to form a cluster while those from the abnormal class fall outside this cluster, or decluster. While some feature extraction techniques attempt to retain the fidelity of the data from which the features arose, the new technique is designed to enhance the discrimination between the normal and abnormal classes. Results from applying the new algorithm to the mass screening of breast thermograms to detect breast cancer are discussed.

A heuristic technique is presented for detecting the presence of an unknown signal in noise and estimating its waveshape. The heuristic utilizes the most recent estimate of the waveshape as a template against which new observations are matched. If the match is "good," the observation is used to update the estimate, and the procedure is repeated. If the match is "poor," the observation is discarded. It is the decisions of the detector which direct the evolution of the estimate while simultaneously the current estimate directs the detector's performance. The mutual interactions of the estimator/detector are conducted without external supervision and without the benefit of training samples, and thus the system falls into the category of a self-learning decision directed adaptive system.