Plants are normally identified by humans using the morphological properties of the leaf. The leaf of the plant is the major characteristic feature that is mostly available in all season. The computer vision system which tries to replicate the human vision system requires some intelligence to detect and identify the object. The leaf can be easily identified by the vision based system if it is present in normal or specified texture of background but in the case of images captured from natural condition it need some extra effort. Most of the research works are done on identification of plant using leaf feature with constraint background. The proposed system is designed to detect plant leaf even in natural condition which may contain interference and overlapping.
The image is captured from the camera and the particular plant leaves are segmented from the image and a single leaf is segmented using marker-controlled watershed segmentation. The marker is generated automatically using morphological operations. The shape of the leaf is obtained and it is converted into feature with Hu moments. The Support Vector Machine (SVM) act as a classifier is responsible for identification of the plant species and it is trained with more than 300 leaves of three different plants. This work is done in BeagleBone Black with the help of an open source library of programming functions mainly aimed at real-time computer vision called OpenCV. The proposed system is tested on leaf image taken in real time and the system shows 86.66% average identification accuracy.