Short-term Vegetation Stress Monitoring Mapping With Machine Learning

The study focused on short-time vegetation stress monitoring mapping using machine learning technique. Two Russiches Basilicum plants were experimented to observed changes in plants health trend by measuring the vegetative indices. One of the plants was regularly watered and exposed to sunlight to absorb chlorophyll while the other plant was deprived of water and exposure to sunlight. Cannon Rebel Tli 500D NDVI camera with two bands of near infrared and blue was used to capture the images of both plants daily. Three vegetation indices of Normalized Difference Vegetation Index (NDVI), Transformed Difference Vegetation Index (TDVI), and Infrared Percentage Vegetation Index (IPVI) were computed as basic indices for vegetation stressed map evaluation and observation. MATLAB multi-paradigm programming language and numeric computing was used for the image processing and calculation of values for various vegetation indices computed for this project. K-means unsupervised classification method was used to separate the background from the foreground for robust analysis by assigning labels of 0 to the background and 1 to the foreground. The plant pixels were categorized into0.2 – 0.4 as normal, 0.4 – 0.6 were moderately healthy, and above 0.6 as healthy. A time series analysis of the various indices was carried out to visualize the result in graphs in order to evaluate and compare the trend of depreciation of the unhealthy plant as well as growth trend of the healthy plant. The result shows that the stressed plant died after 38 days with 0 pixel count of its health condition while the healthy plant still continue to blossom arithmetically at the average pixel count of 509 pixel growth increased daily.