Efficiency of Artificial Neural Networks in the Use and Soil Classification of the Japaratuba River Hydrographic Basin - SE

Abstract

The objective of this work was to evaluate the efficiency of the use of Artificial Neural Networks (ANN) for the classification of land use of the Japaratuba River Basin - SE, from a remote sensing image. A classification by the Maximum Likelihood method was performed to be compared with the classifications generated by ANN, since the first method is already consolidated in the literature. To assess the classifications efficiency, the Kappa Index, Global Accuracy and Root Mean Square Error (RMSE) were analyzed. The Maximum Likelihood classification obtained Kappa Index of 0.95 and Global Accuracy of 96.43%. The ANN architecture more efficient obtained a Kappa Index of 0.93, Global Accuracy of 94.14% and RMSE oscillating between 0.35 and 0.45 during its 10,000 iterations, while the value stipulated as excellent was 0.10. The ANN proved to be efficient in the classification of land use from remote sensing images, since the results of the accuracy parameters showed values ​​that indicate an almost perfect agreement in the classification performed by six of the eight ANN's architectures tested. Thus the generated products can be used as a technical-managerial tool for the environmental management of the study area.

Keywords:
remote sensing; digital image processing; non-parametric methods