IMPROVEMENT OF CLASSIFICATION MODEL OF COASTAL SEDIMENTS USING CONVOLUTION NEURAL NETWORK BY GAUSSIAN BLUR

Abstract

 Distribution of coastal sediments needs to be known for effective coastal management periodicals. In this study, the classification model of coastal sediments using the convolution neural network, which was supposed in the last year by the author, was improved by adopting the Gaussian blur to pictures of training the model. The results of training of the model showed that the accuracy of training and test improved compared with the result of training with original pictures. Implementation of model for aerial pictures indicated the sand area were classified with not enough accuracy. This result shows Gaussian blur decrease features of sand in training pictures. In the results implemented to aerial pictures in the gravel area, increasing of strength of Gaussian blur provided improvement of classification accuracy of gravels. However, some area of gravels still classified as sand in all models. In order to improve of model additionally, it was estimated that the numbers of pictures for training should be increased.