Hybrid neural network classifiers for automatic target detection

Abstract

We describe a one-class classification approach to an automatic target detection problem, which involves distinguishing targets from clutter in diverse environments. We use only target statistics to construct the classifier. The classifier combines conventional and neural network methods. The classifier is a Parzen estimator, which requires storage and recall of all training points. To reduce the size of the training set, we apply two neural network learning algorithms: ( I ) we use a backpropagation network to approximate the Parzen estimator; (2) we apply the infomax learning principle to compress the size of the training set before constructing the Parzen estimator. We find that the results obtained with the infomax scheme approach those obtained with Parzen alone and are better than those obtained with backpropagation.

Robust content-based image indexing using contextual clues and automatic pseudofeedback

“…Replacing the face detector used in the system with another object detector and replacing the contextual analysis module to reflect heuristics about that object type would render a new system capable of handling a new object type. Neural networks have been shown to be very effective in detecting a multitude of objects [7,13,16]. Region-based wavelet signatures have also been shown to be effective in object detection in images [32].…”

Time Series Classification Based on Attributes Weighted Sample Reducing KNN

“…This approach includes building a new archetype or justifying some data points on the original sample [4] . Mostly the neural network is used as editing tool [5,6] . There are also some other approaches: Reference [9] proposed IMKNN approach, which is to cluster based on attributes weight, and use the cluster center as representative points [17,18] , then KNN is carried out on the new sample.…”