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.

Detecting and Counting Soybean Aphids Using Convolutional Neural Network

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  2018 ASABE Annual International Meeting  1800317.(doi:10.13031/aim.201800317)
Authors:   Kai-Yang Hsieh, Yan-Fu Kuo, Chuan-Kai Kuo
Keywords:   Convolutional neural network, Aphid, Count, Pest management, Economic threshold

Abstract. The soybean aphid (Aphis glycines) is a major pest of soybean, one of the most important crops worldwide. It has been reported to substantially reduce soybean growth and production in Asia and North America. Soybean pest management typically applies pesticides to control aphid populations when the aphid density reaches the economic threshold of 250 aphids per plant. Hence, precise estimation of aphid population density becomes an essential component in effective pest management. Conventional manual counting for estimating aphid population size is laborious and prone to human error (e.g., variation in experience). To improve the counting efficiency that is critical to effective pest management, this study proposed to detect and count aphid numbers automatically using machine vision and deep learning. We first cultivated aphids on soybean plants in environmental chambers. Images of soybean leaves were acquired and segmented into small patches. A convolutional neural network classifier was then developed to differentiate the patches into 3 classes: winged aphids (high colonization ability), wingless aphid (high reproductive ability), and background. The proposed method was compared to the conventional manual method for evaluating its accuracy.

(Download PDF)    (Export to EndNotes)