No abstract

Problem statement:Change detection is the process of identifying difference of the state of an object or phenomena by observing it at different time. Essentially, it involves the ability to quantify temporal effects using multi-temporal data sets. Information about change is necessary for evaluating land cover and the management of natural resources. Approach: A neural network model based on both spectral and textural analysis is developed. Change detection system in this study is presented using modified version of back-propagation-training algorithm with dynamic learning rate and momentum. Through proposed model, the two images at different dates are fed into the input layer of neural network, in addition with Variance, Skewness and Eculedian for each image that represent different texture measure. This leads to better discrimination process. Results: The results showed that the trained network with texture measures achieve 23% higher accuracy than that without textural parameters. Conclusion: Adding textural parameters of satellite images through training phase increases the efficiently of change detection process also, it provides adequate information about the type of changes. It also found, when using dynamic momentum and learning rate, time and effort needed to select their appropriate value is reduced.

No abstract

No abstract

No abstract