Hyperspectral analysis of leaf copper accumulation in agronomic crop based on artificial neural network

Abstract

Copper is one kind of trace element in soil which is necessary for the growth and development of plants. Much more copper over the needed amount of agronomic crop is harmful to crop growth and becomes pollutants in soil. At present, there are few studies concerning the quantitative impact of heavy metal contamination on crops. This research investigates an alternative approach. Red edge parameters of rice canopy will be obtained based on the first order and second order derivative spectra, and its relationship with agricultural parameters will be analyzed. It is found that there is strong correlation between red edge position and leaf chlorophyll a / leaf chlorophyll b, red edge amplitude and carotenoid, red edge peak area and the leaf area index, margin and fresh leaves quality. There is no obvious correlation between moisture and red edge parameters. BP artificial neural network method is used to study quantitatively the inherent relation between the chlorophyll content of rice and copper contents in soil. Taking red edge parameters mentioned above which have strong correlation with agricultural parameters, as well as ph value as input, copper content as output, four layers BP neural network with five inputs, one output and two hidden layers will be established. It is tested that the network fitting accuracy reaches 98% and the model has a high fitting degree, which prediction accuracy also receives 85.4%. This study is helpful to improve the ability of monitoring the heavy metal contamination of soil and environment in agricultural region.