Proposed algorithm for hybrid satellite image classification

Abstract EN

The satellite image classification system intends to distinguish between the classes being present in the image.

It is highly challenging because the coverage area of the satellite is large such that the classes appear so small, this makes the process of object distinguish complex.

Additionally, the classification accuracy is an important factor, which the classification system must pass through .

This work presents a satellite image classification system which can classify between the vegetation, soil and water bodies ,etc.

Satellite image classification needs selection of appropriate classification method based on the requirements .In this paper the Support Vector Machine (SVM) and (K-means) are applied on classification of high resolution and low resolution satellite images.

Several different performance measures ,including time ,accuracy, sensitivity and simple contextual information were evaluated.

Additionally , the image was segmented using k-means method in order to improve the classification accuracy ,sensitivity and reduce of time.

The Support Vector Machine was flexible and powerful but still not perfectly suited for high resolution images.

Classifying such images requires contextual information to be taken into consideration and the SVM could not efficiently learn correct context from training examples.

Without including the contextual information obtained from the use of k-means.

The proposed model can be used to obtain the appropriate classification of satellite to show the layers of the earth covering buildings ,roads, agricultural , desert lands, water, and vehicles, etc.

We note that the time does not exceed 32 seconds for the number of images used within our database.