Medicinal plants images recognition by computer vision approach

Abstract

Utilizing machine learning to solve a real problem is the focus of many researchers worldwide. Specifically, using computer vision to aid in plant classification could help find plant diseases or classify plant species. Unfortunately, although there are many studies on plant identification and machine-learning techniques, the lack of a proper plant dataset hindered the development in this aspect. With the recently published dataset VNPlant-200, this study intends to experiment with the performance of various techniques to deploy a model to classify herbal plants in Vietnam. As the dataset captured plant images in a real-world setting, the model aims to deliver functional accuracy to help classify the plant species in a complex environment. Furthermore, this study will experiment with the image resolution provided, 256 × 256 and 512 × 512, to find the most optimal solutions. The methods are to use Histogram of Oriented Gradients (HOG), Local Binary Pattern (LBP), Oriented FAST and Rotated BRIEF (ORB), and various Convolutional Neural Networks (CNN) to extract features that represent the images. Then, the features will feed forward to the classifiers, such as K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Passive Aggressive Classifier (PA), and Logistic Regression (LR). The experiment will test this method with both provided image resolutions.