Recognition of Formations of Volcanic Origin in Satellite Images using Neural Networks

Abstract

Here, we are trying to relate the formations of volcanic origin with data from satelite image (Landsat MSS). With this in mind, we approached theproblem with the neural network technique, The area under study is La Payuniq in the province of Mendoza, Argentina. The satellite image used is centered at35o55'72" iouth and 68o55'48" west. The neural network approach implies a minimization of the square error between the desired output and the actual output obtained. We were faced with a problem of optimization that was handled by the Simplex algorithm. The network used consists of four neurons.at the input layer, four neurons atil't" hidd"n layer and one neuron at output. This structure. matches the four spectral bands of the Landsat MSS data as input and the geological map as output. The neural network is trained over an area of known Iithology. Therefore. it rela- tes radiometric with geologic caracteristics. In the present case the results are quite coincident with-the aerìal-photographs interpretation of the area. The good results indicate the potential for the identification of volcanic formations in regions ofarduous access.