Analysis of traditional computer vision techniques used for hemp leaf water stress detection and classification

Cannabis sativa L. has risen in popularity due to its large variety of uses and environmentally friendly impact. C. sativa L. is extremely sensitive and displays phenotypic responses to water stress in its leaf and stem structure. Optimizing the use of water in the agricultural process of cultivating hemp requires the determining of the water potential in the hemp plant. Computer Vision techniques to determine water potential can be used as opposed to traditional destructive and complex to implement techniques. The goal of this study is to prove that water stress detection in hemp leaves can be achieved using computer vision as well to create a model and compare computer vision techniques. This study used a dataset pooling technique to create the dataset of hemp leaves. The dataset is split randomly at an 80–20% ratio of training data and testing data, respectively. Two derivatives of the traditional pattern recognition pipelining model were used. The first pipeline employed traditional computer vision techniques such as Canny Edge Detection, Contour Analysis, SIFT, and SVM Classification. The second pipeline embraced an object detection approach by implementing Haar Cascades. The results of the study vary greatly leading to researchers to believe that more work needs to be done to improve performance.