Artificial Neural Networks Application in Population Chlorophyll Content Forecast from Cotton Plant Digital Images

摘要

Leaf chlorophyll content in a population of crops, if obtained in a timely manner, serves as a key indicator essential for crop planting and growth management, as well as crop problem diagnosis. In this paper, an artificial neural network based prediction system is reported for the extraction of leaf population chlorophyll content from the cotton plant images. This system needs to be trained with leaf green pixels extracted from a set of cotton plant images following a two-step procedure. In the first step, a global thresholding and pixel-labeling algorithm is applied to the image to help identify green leaf pixels and all the identified green pixels will be labeled as red. In this step, however, the background noise can not be distinguished from the real green leaves and thus mistakenly labeled as red. This problem is addressed in the second step, where an omnidirectional scan noise filtering coupled with the hue histogram statistic method is designed to suppress that background noise. It has been confirmed by the experimental results that the cotton plant leaf population chlorophyll content determined by using the proposed system is in sound agreement with those obtained from the traditional method using manually operated equipment; of the fourteen test images, the average prediction error for chlorophyll density (mg/cm~2) is within 6.2 %.