Remote sensing monitoring of a bamboo forest based on BP neural network

Abstract

The collection of information on bamboo forests plays a crucial role in the calculation of carbon content reserves, and the acquisition of high-precision information will be good for reducing estimation errors. High precision is obtained with the adoption of a back propagation (BP) neural network to extract information on bamboo forests from Enhanced Thematic Mapper+ (ETM+) remote sensing images with the assistance of neural network modules provided by Matlab. We obtained a production precision of 84.04% and a user precision of 98.75%. We also conducted a comparison of classification differences of three training functions, i.e., the, Levenberg-Marquardt BP algorithm function (Trainlm), a gradient decreasing function of adaptive learning rate BP (Traingda), and a gradient lowering momentum BP algorithm function (Traingdm). Our analysis suggests that Traingda had the highest precision while Trainlm function required the shortest training time.

Key words： forest management; Back Propagation (BP) neural network; bamboo forest; classification; remote sensing; Enhanced Thematic Mapper+ (ETM+)