MAPPING OF TEA GARDENS FROM SATELLITE IMAGES-A FUZZY KNOWLEDGE-BASED IMAGE INTERPRETATION SYSTEM

Abstract

Tea is one of the most valuable natural resources of India. It commands a pivotal position in the nation's economy as it is one of the major forex earner for the country. Naturally there is always a demand for cost-effective techniques for continuous monitoring, assessment and management of tea gardens. Remote sensing technology offers numerous advantages over traditional methods of conducting agricultural resource survey and management. However, land cover classification is the first step for most of the application of remotely sensed satellite data. So the detection and identification leading to interpretation and mapping of the tea gardens (land cover class) from satellite images is the prerequisite for application of remote sensing technology to survey and management of tea gardens. The present research work is undertaken to develope an image interpretation system by synthesizing the cognition and reasoning processes of an expert image analyst with quantitative analysis leading to a versatile system for mapping of tea gardens from satellite images. Emulating an expert image analyst is a difficult task as expert assimilates and process information in qualitative or ambiguous terms during photointerpretation. Also, the cognition and inferencing during photointerpretation depends on the heuristics of the analyst which differ from one expert to the other. Since the success of a knowledge-based system depends on the strength of the knowledge base (KB) and on the use of proper heuristics, the extraction of knowledge and the heuristics used by experts to carry out interpretation of tea gardens from satellite data is crucial. (i J On the other hand, the mixed pixels present in satellite data are portend of interpretation ambiguity leading to errors in interpretation. It has been observed that experts generally use linguistic variables and fuzzy labels to address facts and domain knowledge of information classes. The reasoning of interpretation involves fuzzy truths, fuzzy connectives and fuzzy rules of inferences. Thus, it is found that, to emulate an expert image analyst, a fuzzy logic system provides the natural framework to deal with image cognition and interpretation process of expert photointerpreter. However, the actual interpretation of a satellite image depends not only on the spectral characteristics of the information classes but also on spatial and temporal characteristics of the scene. Further, it depends on the data to be used. For this study, an area in the district of Cacher in Assam (INDIA) lying between 24°45" to 25°00' North latitude and 92°45' to 93°00' East longitude is considered. The area is highly endowed with tea gardens. The IRS (Indian Remote Sensing Satellite) LISS (Linear Imaging Self Scanner) II geocoded data of November 8, 1989 (of the study area) has been taken for this study. In this study, a fuzzy knowledge-based image interpretation system is proposed emulating the multi-stage, multi-feature and multi-iteration heuristics of an expert image analyst. The acquisition of knowledge for identification and mapping of tea gardens from satellite images is achieved through spectral knowledge of land covers, domain knowledge and expert's heuristics. The knowledge base and feature attributes o\' the information classes are expressed by linguistic variables and fuzzy attributes. The inference mechanism is modelled on the basis of fuzzy logic. (ii) In the proposed system, the land covers of the study area are represented numerically by the spectral measurements at different bands of satellite data. The data is then preprocessed to remove the various errors present in the raw data followed by extraction of linguistic variables. The Knowledge Base carries out the feature selection process depending on the stage of the classifier which represents N-dimensional sample space by their Q-dimensional (Q<N) linguistic variables in terms of their fuzzy labels. The linguistic variables gets transformed to their fuzzy attributes by appropriate characteristic functions. Finally, according to the knowledge base, inference mechanism interpretes the possible information class of the sample. The system provides different outputs in each stage of its classification. These can be utilised as independent sources of land cover informations. At Stage I, the system provides the sub-pixel information of water against non-water of the scene. The graded variation of the water bodies are excellently depicted in the output of the system. At Stage II, the system provides sub-pixel vegetation information. It is found that the system works excellently for sub-pixel analysis of water and vegetation. Thus it provides a huge enhancement of information from satellite data regarding the information classes and their distribution. The vegetation information is further analysed by the system to provide information regarding agricultural and non-agricultural vegetation at Stage III. Finally, at Stage IV, the system provides a map of tea gardens from agricultural vegetation through graphics display. The proposed system provides sufficiently accurate information of the different land cover classes of interest and can be used reliably for mapping of different types of land covers, over and above tea gardens. The liii) results obtained from the working of the system in each stage of operation as well as from the experimental study show that the performance of the system is better than the minimum classification accuracy required i.e., 85% to justify the operational capability of a system. However, the ultimate aim of developing a humanistic image interpretation system is to have a viable alternative to expert image analysts who are rare and not readily available. Despite the progress has been achieved in this work, the developed system is a primitive one to replace an expert photointerpreter. To achieve a workable alternative, some potent research topics are suggested. If these steps can be taken care of in the spirit of humanistic approach and can be integrated in the proposed system, a viable solution can be hoped for. (