Impact of Image Pre-processing Operations on Wheat Canopy Segmentation

This research work is an exploratory study of imaging methodologies that could aid in the development of wheat canopy segmentation and water stress detection systems. Through the planned experimentation process, it was found that the chlorophyll fluorescence images needed to be pre-processed before being given as input to the segmentation algorithm for efficient extraction of regions of interest. For efficient canopy extraction, the enhancement associated with contrast and removal of random noise must be done. Multiple pipelines were constructed and it was empirically verified that the TV-L1 denoising with Primal-Dual algorithm is best suited for denoising. The contrast stretching method, also referred to as Min-Max, is the most appropriate operation for pre-processing the images. These pre-processing methods were extremely useful for extracting the area of an image that has maximum photosynthetic activity. The criteria for the selection of pre-processing methods are based on the quality of the segmentation algorithm that is computed using the metric Intersection Over Union. The work demonstrates a most constructive way to separate the wheat canopy from the chlorophyll fluorescence images, for the creation of an automatic drought detection system.