Object-oriented Classification Method Application In Sichuan Hilly Forest Classification Research

Forest resources survey is one of the major national conditions and strength surveys. The purpose is to comprehensively identify the national utilization of forest, grasp the real basis data of woodland, and provide a reliable basis for the national macro-control consumption of forest resources, draw up the national economic plan, control and guide the forestry production plan. In recent years, with the development of remote sensing technology, the time resolution remote sensing, spatial resolution and spectral resolution has been greatly improved, to provide people with extremely rich surface information. Meanwhile, for the investigation of forest resources provides a new scientific method, making precision survey of forest resources has been further improved, and greatly improves the operating efficiency. In most cases, traditional information extraction methods are used the similarity of the spectral characteristics of surface features to extract information. This extraction techniques which based on the information element, ignore the texture of the entire image, so that the extracted image recognition accuracy is affected; therefore, in order to solve this problem, a method of object-oriented classification is proposed. Object-oriented remote sensing image classification method using the full range of information spectrum, texture and space, to make up for the shortcomings of traditional classification methods, obtained better classification results, avoid salt and pepper phenomenon, breaking the traditional classification methods to Pixel basic classification and processing limitations unit.This article is based on eCognition software, making use of object-oriented classification to classify the forest in Yuzhen Town, Nanbu County, Sichuan Province, by SPOT5 remote sensing images, and verify the classification results. Aims to explore the forest classification method for remote sensing image hilly area of Sichuan Province, and improve the accuracy of SPOT5 image forest classification methods can learn from experience and to provide for the rapid, accurate and reasonable investigation of forest resources. In this paper, after a study concluded the following:(1) Though the mean area ratio method and repeated segmentation test, finalize the optimal segmentation scale and optimal segmentation parameters of the construction land, farmland and woodland, and paddy fields, are 10,30,70 respectively; shape factor are 0.7,0.3,0.5 respectively.(2) Making full use of the relationship between texture and spectral characteristics of image objects as well as between the various levels of segmentation, build the knowledge base for the area classification, fuzzy membership function category classification method woodland, open woodland, farmland, and construction extracted. This classification rules are suitable for classification of the forest area.(3) The accuracy of the evaluation results show that the maximum likelihood classification overall accuracy and kappa coefficients were 54.50% and 0.4921 respectively, object-oriented classification overall accuracy and kappa coefficients were 80.93% and 0.7672 respectively, significantly higher than the maximum likelihood method.Overall, the object-oriented classification method significantly improved the accuracy when faced high resolution image classification. The results in line with actual production requirements, this method is suitable for remote sensing classification of Sichuan Province hilly area of woodland.