Automatic classification of environmental noise events by hidden markov models

The classification and detection of event-related brain potentials was investigated using signal processing and statistical pattern recognition techniques. Amplitudes at sampled time points and frequency quantities have previously been used as features. Improvements to these procedures were obtained by using features from the time-frequency plane to utilize the geometric relationship between time and frequency, capitalizing on the nonstationarity of the evoked potential signals. These features were transformed from the original data sets based upon a two-step classification/feature selection procedure which uses selected frequencies from step 1 as parameters for data filtering in step 2. Features were selected from the filtered data, classifiers were designed, and the estimated classification accuracies were computed.

Classification and Detection of Single Evoked Brain Potentials Using Time-Frequency Amplitude Features

Experiments were carried out to demonstrate the ability of an artificial neural network (ANN) system to distinguish between the noise of two helicopters. The ANN is taught to identify helicopters by using two types of features: one that is associated with the ratio of the main-rotor to tail-rotor blade passage frequency (BPF), and the ohter that describes the distribution of peaks in the main-rotor spectrum, which is independent of the tail-rotor. It is shown that the ability of the ANN to identify helicopters is comparable to that of a conventional recognition system using the ratio of the main-rotor BPF to the tail-rotor BPF (when both the main- and the tail-rotor noise are present), but the performoance of ANN exceeds the conventional-method performance when the tail-rotor noise is absent. In addition, the results of ANN can be obtained as a function of propagation distance.

Identification of Helicopter Noise Using a Neural Network

The identification of helicopter noise using a neural network

A pattern recognition system has been developed to classify five different aerospace acoustic sources. The system consists of one microphone for data acquisition, a preprocessor, a feature selector, and a classifier. In this paper the performances of an associative memory classifier and a neural network classifier are compared with the performance of a previously designed system. Source noises are classified using features calculated from the time and frequency domain. Each classifier is trained to classify source noises correctly using a set of known sources. After training, the classifier is tested with unknown sources. Results show that over 96% of the sources were identified correctly with the new associative memory classifier. The neural network classifier identified over 81% of the sources correctly.

Sparse distributed associative memory for the identification of aerospace acoustic sources

Instrumentation currently available for the automatic monitoring of noise nuisance has the shortcoming that although the intensity, duration, and time of occurrence of noises may be recorded, their source often cannot be identified. Research directed towards providing improved instrumentation which can identify sound sources is described. Our results suggest that application of statistical pattern recognition to recorded sounds can differentiate sources which are structurally dissimilar (e.g. trains, fixed-wing aircraft, helicopters) with an accuracy of better than 95 percent. The work is continuing to discriminate sounds which are structurally similar (e.g. different types of aircraft), and to produce hardware capable of field application.

Pantelis Moukas

Papers

Automatic Identification of Noise Pollution Sources