Machine vision-based precision irrigation system for Sunagoke moss production

There are many methods for sensing water condition in Sunagoke moss. The direct measurement of canopy parameters is considered to be relatively inefficient and destructive. One alternative is the use of indirect measurement such as machine vision which can provide accurate result and non-destructive method. This study investigated the use of machine vision for monitoring water content in Sunagoke moss. Feature selection and features extraction are the most important steps in predicting Sunagoke moss water content using machine vision. The main goal of this study was utilizing machine vision as non-destructive sensing and Genetic-Neural Algorithm as feature selection techniques to determine water content of Sunagoke moss using non-invasive method. In our system, we extracted 106 features consisting of color features, textural features (gray level cooccurrence matrix, RGB color co-occurrence matrix, HSV and HSL color co-occurrence matrix textural features) and morphological features. Ten textural features were calculated, including Entropy, Energy (Angular Second Moment), Contrast, Homogeneity, Sum Mean, Variance, Correlation, Maximum Probability, Inverse Difference Moment and Cluster Tendency. There are (2¹⁰⁶-1) possible feature subsets. The specificities of this study was that we were not looking for single feature but several associations of features that may be involved in determining water content of Sunagoke moss. The genetic algorithms managed to select interesting features and artificial neural network was able to predict water content according to the selected features. We propose a model of neural network based precision irrigation system for Sunagoke moss plant factory to realize precision agriculture.