Plant Species Image Recognition using Artificial Intelligence on Jetson Nano Computational Platform

Abstract

The ongoing research for plant/animal species identification by computer vision engineers is exciting and vast. This paper describes a deep learning approach to identify plant species using image analysis. An efficient Artificial Intelligence System is designed and implemented with minimal components, including a camera and Jetson Nano (single-board embedded computing device). Convolutional Neural Networks are trained to capture the features from images and recognize the plant species. Thus, the experiment used, in particular, CNN architectures- AlexNet, ResNet50, and MobileNetv2, within Python’s Tensorflow framework, to accomplish species identification. Of these, AlexNet provided the best results, with 72% validation accuracy after 15 epochs. A portion of the LeafSnap dataset, containing 15 plant species and 30 images per species, was used to compare the performance of architectures.