Evaluation of the level of accuracy of matching multispectral aerial images through the "Siamese" neural network model

Abstract

This study evaluates the accuracy of the "Siamese" neural network for matching multispectral aerial images. A set of image data was preprocessed, and the neural network was trained. The precision, recall and F1-score metrics were analyzed, finding that the average precision was 56.35%, with values varying between 25.00% and 91.18%. Recall was more stable with a mean of 52.44%, while precision had a mean of 56.57%. It is concluded that the "Siamese" neural network is effective for matching multispectral aerial images, although the accuracy depends on the training configuration and the characteristics of the data set.