Quantitative analysis of landscape morphology using improved self-organizing mapping network

Abstract

To achieve high-quality development of rural landscapes, this study quantitatively analyzes and identifies the landscape morphology of polder fields through an improved self-organizing mapping neural network (SOM NN). The study first uses landscape morphology index to quantitatively translate the polder landscape, and then preprocesses the data through factor analysis to reduce dimensionality. On this basis, the K-means algorithm is introduced for pre-classification, and the SOM NN is improved by combining the confidence matrix. The clustering performance of the algorithm is first tested on the Iris data set, which contains measurements of three different species of iris (Iris), including calyx length and width, and petal length and width. The experimental outcomes denote that the average contour score and average adjusted Rand index of the improved SOM NN are 0.549 and 0.978, respectively. Its clustering accuracy and recall are 0.934 and 0.923, respectively, which are higher than other algorithms. In computation time, the average computation time of the improved SOM NN is only 0.022 ms, far lower than other algorithms. Moreover, the effectiveness of the improved SOM NN in identifying different types of polder landscape forms is verified through practical application analysis of Tangpu polder and Taixing polder in Jiangsu Province. The above results indicate that the improved SOM NN has great potential for application in landscape morphology quantification research, and can provide scientific basis for the management and protection of polder landscapes.