In this paper the searching capability of build up an object oriented classification system which is capable of
classification a given forest scene into its various constituents. To simplify the problem, six categories of forest
structures were defined. These categories are trees, bushes, grasses, foliage, sky and background sky .They are
sufficient to represent typical forest scenes dealt with in this application. To implement such classification system. We
proposed a Genetic Algorithm (GA) to segmentation image and find the best seed for each category. According to this
scheme, an image is divided evenly into small block. Then it is processed block by block. For each block, Discrete
Cosine Transform (DCT) is applied to determine some of DCT coefficient in compressed domain as the feature vectors.
Then take the seed values for each segment and the DCT coefficients to represented the inputs of feed forward neural
network. These system successes in classification all objects in image although used different kinds of activation
functions (hyperbolic functions), compares among them and find the best of it in obtaining on fast results. As a result
the A soft computing method will be higher classification accuracy than that of traditional pixel-based supervised
classification and gives convenient environment to use.