Deepak Khemani

This paper focuses on the application of CBR to soil analysis from chromatograms. The shape, size and colour of the chromatogram image are hypothesized to contain important information of the mineral content in the soil. Since chromotogram preparation is cheaper than chemical analysis the goal is to predict the nutrients from the chromatogram image features in the future rather than by direct chemical analysis. The method proposed in this paper will be new, as the current process of chemical analysis of soil is done manually, which is an expensive, time consuming and laborious process. This method of analysis will benefit farmers all across the globe, who are looking for innovative means to obtain their soil characteristics during the process of farming. In this application, the key assumption is that - similar chromatograms have similar soil properties. This paper focuses on the definition of similarity measure and determining the weight model for the feature set needed for the application. © Springer-Verlag Berlin Heidelberg 2008.

Image understanding concerns the issues of how to facilitate interpretation of images. The work reported here is in support of a novel approach for machine interpretation of soil chromatograms. A chromatogram image is obtained from an alkaline solution of silver nitrate and soil. The patterns in the chromatogram contain information of mineral contents in the soil. We describe a combination of different methods for feature extraction. The objective is to segment the image into three regions, and measure the colour and dimension of different bands. The extracted features are used for image matching in a case based reasoning system.

A circular paper chromatogram is obtained from an alkaline solution of silver nitrate and soil. The shape, size, color and textural patterns of the chromatogram image are hypothesized to contain important information of the mineral content in the soil. We present a method to automatically analyze the chromatogram image for feature extraction. Image pre-processing is an important step before extracting the features of the image. Chromatogram image preprocessing involves detecting the center of the chromatogram, normalization and then segmentation into different concentric regions. Since chromatogram patterns are similar to iris (human eye) patterns, we have adopted iris-preprocessing methods. In this paper, we present a combination of different approaches: to detect the center, normalize and segment the chromatogram. Centre detection algorithm finds the center of the chromatogram which is assumed as the origin for normalization. Chromatogram normalization involves transforming from Cartesian to polar coordinates, so that chromatogram looks like an unwrapped polar image. Finally, color texture segmentation is used to detect different regions. Results of feature extraction are compared to that given by soil experts to test the accuracy of the system. © 2007 IEEE.