R. Buccheri

Use of a direct non-parametric pattern recognition method for the automatic selection of high energy gamma-ray events in spark chamber pictures

Abstract The problem of the automatic selection from background events of electron pairs due to gamma-ray conversion in a multiplate spark chamber is dealt with by means of a direct non-parametric pattern recognition method. The method is applied to the data obtained in a gamma-ray astronomy experiment using a balloon-borne optical spark chamber. A filter operation is initially introduced to exclude clearly evident spurious events; the remaining spark chamber images are then reduced to their most significant zone in order to lower time and cost of subsequent analysis. The feature selection is performed by using a “training set” of labeled events and the classification made introducing a pre-established linear threshold decision function. Results are presented giving the percent of correct classification as a function of the threshold value chosen.

Study of Pulsar Light Curves by Cluster Analysis

The distribution of the phase numbers, corresponding to the arrival times of the gamma-ray photons detected by the COS-B satellite from the directions of the Crab and Vela pulsars, is analyzed by a clustering technique with the aim to detect possible microstructures in the pulsed emission. The method is found to be promising especially in view of the future gamma-ray experiments where better photon counting statistics is expected.

Multivariate Cluster Analysis of Radio Pulsar Data

Data relating to 336 radio pulsars are analyzed by a cluster method aiming to study the relevant features of the sample.

Search of Significant Features in a Direct Non Parametric Pattern Recognition Method. Application to the Classification of a Multiwire Spark Chamber Pictures.

We give preliminary results of the application of a direct non parametric pattern recognition method to the classification of the pictures of a multiwire spark chamber. The method, developped in an earlier work, for an optical spark chamber, looks Promising. The picture sample used has, respect to the previous one, the following characteristics: a) the event pictures have a more complicated structure; b) the amount of background sparks in an event is greater; c) there exists a kind of noise which is almost always present in some structured way (double sparkling, bursts, …). New features have been used to characterize the event pictures; the results show that the method could be also used as a super filter to reduce the cost of further analysis.