Research On Mangrove Detection With Remote Sensing In Guangxi Coastal Areas

With the vigorous development of space remote sensing technology, the high resolution remote sensing image has been widely used. The research mainly focus onhowto effectively extract the various features form the image and converse these features. The space remote sensing technology provides a quick and accurate method for the monitoring and protection of the natural ecology of mangrove forests, so the research has a high practical value and practical significance. In reality, only depending on the parameters of the spectral image coefficients or spectral values, the classification results show that the required target area are unable to be obtained. Most of the previous researches did not have a good combination of spectral information and non-spectral information, so, in the paper, a variety of features are combined and used to monitor the growth area of mangroves.The selection of feature parameters has a great influence on the experimental classification results when the mangrove needs to be recognized. The ability of recognizing the target area is various widely for different feature parameters. Because of the growth characteristic of the mangroves, some features cannot effectively distinguish the mangroves, sea water and paddy field. To improve the recognition ability for different regions, the paper introduces the temperature and humidity index (TMI) as a new recognition feature and analysis the divisibility of multiple features. By the normalization for the mean of distance, the result proves that the characteristic parameter can effectively identify the target area.The target recognition area, in the paper, is divided into five categories:artificial land, mangrove, land plant, paddy field and water body. The Following up studies were focused on the 5 types.The gray level co-occurrence matrix is mainly aimed at extracting texture features from different remote sensing multispectral images. Those effective statistics features, such as the contrast, correlation, energy, entropy constitute the texture feature vector, are used as a training set for SVM classification model. The 5 types of the target area is described above. But the actual classification accuracy is less than 30%, so the feature cannot effectively distinguish between the paddy field and land plant.The vegetation coverage and its growth vigor have been widely used in the qualitative and quantitative evaluation of the vegetation index. The green index, the brightness index, the difference index, the normalized difference vegetation index are used in the remote sensing multispectral image processing. Using these 4 features to recognize all target area by the SVM classification model, the experiment show that the features can distinguish between land vegetation and mangrove forest. However, the classification results between the paddy field and mangrove forest appears a certain degree of pixel confusion. Therefore, the result is not ideal.The KT transformation bases on the experience matrix and is a linear transformation for different sensor fixed transform matrix. A number of parameter matrixes, such as surface brightness index, green vegetation index, humidity index, are obtained by processing the data of remote sensing multi spectral images. Using the three features as the data source to tanning SVM classification model, the classifier can distinguish the paddy field and the mangrove forest in a certain degree. Unfortunately, for some land plants and mangroves, the classifier is easy confusing on some mixed pixels.Knowledge rule uses the plant characteristic and the growth characteristic of the mangrove forest itself to generate the knowledge rule, which enhances the classification accuracy and the extraction ability of characteristic parameters. Thus, the image characteristic value is effectively processed. Combined with the feature vector by Tasseled Cap Value, the mangrove area is effectively identified. The experimental result shows that the classification accuracy is about 90%.Comparing the four kinds of classifier described above, the classification accuracy of the last three is more than 70%. And the accuracy of the classifier based on knowledge rule can reach 90%. The method, only using spectral values and the texture feature extracted from the gray level co-occurrence matrix to classify, cannot effectively distinguish the growth range of the mangroves. All experimental results show that the classification accuracy of the gray level co-occurrence matrix is lowest. And the classification results of vegetation index and tasseled cap transformation are prone to local confusion. The classification results based on knowledge rules are better and the accuracy is higher. It is feasible to use a variety of feature vectors to classify the target areas, and the knowledge rule-based SVM neural network model can provide the effective identification of the mangroves.