An interactive hybrid expert system for polar cloud and surface classification

Abstract

An interactive hybrid expert system is developed to classify polar scenes using AVHRR LAC imagery. A total of 183 spectral and textural signatures are generated from which the 20 "best" are chosen using the Sequential Forward Selection procedure. These 20 features are used to populate the working memory of the expert system. A probabilistic neural network is used as the inference engine to make probabilistic estimates of class membership. As part of the inference engine, a sophisticated outlier test is performed to provide a measure of classification confidence.During a session, the user is provided with an extensive set of on-screen aids to assist in labelling. The user may modify the knowledge base by adding new samples to existing classes or by including new classes. The expert system provides confidence measures and a distance measure from the proposed class cluster centre. The interactive environment allows the user to test the impact of class labelling upon the knowledge base before new data is entered. For users working with very large datasets and very complex scenes, the integrity of the knowledge base is the primary concern.A bootstrap method is used to validate classification accuracy. On the basis of 100 bootstrap samples, an overall classification accuracy of 87% is achieved, with a standard deviation of 1%. The result is that much more

Automated Image Data Exploitation Final Report

“…8,9 For classification purposes, the use of textural features based on the intensity in a single channel is abundant in the literature. 10,11,12,13,14 We then mosaic-ed the tiles from class 1 into a mask (Mask 1), which we then use to demarcate areas of "sufficient" spectral mixture that they are likely to contain built structures [ Figure 3]. Obviously, Mask 1 needs refinement before we can consider it to indicate areas containing built structure, as distinct from areas that are spectrally mixed for some other reason.…”

Convective cloud identification and classification in daytime satellite imagery using standard deviation limited adaptive clustering

“…As a basis for improved efficiency, studies that capitalize on the textural and spectral signatures of clouds have been used to classify cloud high‐resolution Landsat multispectral imagery [see, e.g., Chen et al , 1989; Wielicki and Welch , 1986; Welch and Wielicki , 1989]. However, the classification schemes presented in these studies suffer from the limitation that they are trained for a specific environment, such as for maritime [ Bankert , 1994], tropical [ Shenk et al , 1976; Inoue , 1987], or polar [e.g., Ebert ,1987, 1989; Key and Barry , 1989; Key , 1990; Welch et al , 1989, 1992; Rabindra et al , 1992; Tovinkere et al , 1993] regions. Peak and Tag [1992] developed a method of cloud classification based on a method using a cloud classification database, image segmentation and neural network identification.…”

A comparison of paired histogram, maximum likelihood, class elimination, and neural network approaches for daylight global cloud classification using AVHRR imagery

“…First, does the use of more advanced features (other than just the five AVHRR channels and a sixth channel derived from the reflectance of channel 3) significantly improve classification accuracy? It has been shown that spectral, textural, or combinations of spectral and textural features can be effective for cloud identification, particularly in the polar regions [Baurn et al, 1995;Ebert, 1987Ebert, , 1989Welch et al, 1990Welch et al, , 1992Key and Barry, 1989;Key 1990;Rabindra et al, 1992;Tovinkere et al, 1993]. These categories represent a large number of potential features; thus various approaches for feature reduction are utilized in order to select a small subset that best separates the classes [Richards, 1993].…”