Optimized Supervised ML for Medicinal Plant Detection - An FPGA Implementation

Abstract

Medicinal plants have a huge significance today as it is the root resource to treat several ailments and medical disorders that do not find a satisfactory cure using allopathy. The manual and physical identification of such plants requires experience and expertise and it can be a gradual and cumbersome task, in addition to resulting in inaccurate decisions. In an attempt to automate this decision making, a data set of leaves of 10 medicinal plant species were prepared and the Gray-level Co-occurence Matrix (GLCM) features were extracted. From our earlier implementations of the several machine learning algorithms, the k-nearest neighbor (KNN) algorithm was identified as best suited for classification using MATLAB 2019a and has been adopted here. Based on the confusion matrices for various k values, the optimum k was selected and the hardware implementation was implemented for the classifier on FPGA in this work. An accuracy of 88.3% was obtained for the classifier from the confusion chart. A custom intellectual property (IP) for the design is created and its verification is done on the ZedBoard for three classes of plants.