IoT-based Plant Identification using Multi-level Classification

This paper presents a two-level classification scheme to identify different plant types according to their color and shape attributes. The proposed scheme can achieve accuracy without increasing the complexity by combining two simple Naive-Bayes classifiers in a hierarchical structure with a low computational cost. The algorithm utilizes hand-crafted features in machine learning, which relies essentially on the information present in the image. These features are extracted and used on two levels depending on the complexity of the samples. The first-level classifier serves as a filter that identifies the majority of input image samples with high accuracy using simple color features with low computational cost. If an image cannot be classified in the first stage, it is deferred to the second level where additional color and shape attributes are extracted. Thus, the second-level classifier is trained with new data generated through augmenting features produced in the first stage. An Internet of Things (IoT)-enabled data acquisition framework is developed to experimentally evaluate the proposed scheme, which collects image data in real time and processes the information based on the proposed classification scheme. Four types of artificial plants were placed in a growth chamber equipped with image sensors, and LED lighting with the image data acquired through a cloud service. A color feature, a shape feature, and two variations of the proposed multilevel color-shape classifiers were implemented to acquire, classify, and compare the collected data. The conducted experiments demonstrate performance of the combined two-level classifier which indicates that improved performance over single level classifiers can be achieved.