Often, when classifying multispectral data, only one class or crop is of interest, such as wheat in the Large Area Crop Inventory Experiment (LACIE). Usual procedures for designing a Bayes classifier require that labeled training samples and therefore ground truth be available for the "class of interest" plus all confusion classes defined by the multispectral data. This paper will consider the problem of designing a two-class Bayes classifier which will classify data into the "class of interest" or the "other" classes but will require only labeled training samples from the "class of interest" to design the classifier. Thus, this classifier minimizes the need for ground truth. For these reasons, the classifier is referred to as a single-class classifier. A procedure for evaluating the overall performance of the single-class classifier in terms of the probability of error will be discussed.
Date of this Version