An adaptive and quantitative rubber nutrient status analyzing system by digital foliar images

Methods of plant nutrient status analysis using the leaf image features like color, textures and shape have been proposed in recent years, which facilitate the rapid development of nondestructive detecting techniques that are better than traditional soil and plant detection in many ways. However, most of these methods suffer from the following drawbacks: 1) device dependent due to the distortions caused by different imaging systems, thus making the detecting results of the foliar image captured with different devices apparently inconsistent; 2) sensitive feature extraction due to varying ambient interference of illuminations directly leads to unstable detection result; 3) qualitative classifier that is hard to achieve precision fertilizer guidance. In this paper, we present a system that addresses the issues using a normalized agro-chemical model and an implicit correction on color and shape distortions of the foliar images. The operation records have shown that the detection result can be obtained at least one week ahead of the traditional observation, and its computational error of the nutrient deficiency is within the range of the plant fertilizing guidance.