Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Application of Convolutional Neural Networks on the Development of Plant-Parasitic Nematode Image Identification System
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2021 ASABE Annual International Virtual Meeting 2100870.(doi:10.13031/aim.202100870)
Authors: Hsien-Hua Lai, Yu-Tang Chang, Jiue-In Yang, Shih-Fang Chen
Keywords: Plant-parasitic nematode, deep learning, region-based convolutional neural network
Abstract. Plant disease and pests are the third largest source of the loss rate of crops worldwide after the loss causing by natural disaster and transportation process. Plant-parasitic nematodes (PPN) are one of the main plant pests and cause over 100 million US dollars of agricultural loss worldwide every year. In general, identifying species of nematodes from their morphological characters is the first step for a nematologist. However, limited professionals cannot fulfill the huge demand. A fast identification system for nematodes applying imaging classification methods would be a potential solution to amend the practical needs. In recent years, the rise of deep learning brought a significant breakthrough in object detection. In this study, deep learning algorithms were applied to develop the plant-parasitic nematode (PPN) image identification system. The experimental dataset included four genera (e.g., Aphelenchoides, Bursaphelenchus, Meloidogyne, and Pratylenchus) and 10 species of common PPNs worldwide. In addition, one free-living nematodes, Caenorhabditis elegans, and two entomopathogenic nematodes, Heterocephalobellus sp. and Metarhabditis amsactae, were also collected as the control group for the recognition of the non-plant-parasitic nematodes. In total, 9483 images of nematode images were acquired. Faster Region-based Convolutional Neural Network (Faster RCNN) architecture was selected to develop the PPN identification model. As a result, the model with the ResNet-101 structure achieved the highest mean average precision (mAP) of 0.9018.
(Download PDF) (Export to EndNotes)