Machine vision system for the automatic segmentation of plants under different lighting conditions

Precision agriculture needs the use new technologies for identification. Using digital images and analysing their colour is one of the most useful methods for the segmentation of plants from their background and is a basic operation in all machine vision applications. A machine vision system is presented based on hybrid artificial neural network-harmony search (ANN-HS) classifiers for the segmentation of different plants in different growth stages, different conditions of the day and one controlled state and different imaging situations. This system works in two stages; the first stage is to specify photography state and the second stage is to apply an appropriate threshold. In total, 23,899 images were taken from eight different states during the day and one control state. Five features among 126 extracting features of five colour spaces RGB, CMY, HSI, HSV, YIQ and YCbCr for use in classification unit were selected using hybrid artificial neural network - differential evolution algorithm. Meta-heuristics and statistical classifiers were used for classification. The results showed that the accuracies of meta-heuristics method of the hybrid artificial neural network-harmony search and k-nearest neighbour statistical method were 99.69% and 94.06% respectively. In order to determine appropriate thresholds an improved YCbCr colour space was proposed. The results showed that among eight different states during day and one control state, the level of threshold for six states must be determined in third channel related to this colour space and the rest should be determined in HSV and YIQ colour spaces. The suggested machine vision system segments each image during 0.37 s. Finally, it can be claimed that this system is applicable in all machine vision systems related to fields and has high accuracy and speed.