Automatic monitoring of flying vegetable insect pests using an RGB camera and YOLO-SIP detector

• Published in: Precision agriculture Vol. 24; no. 2; pp. 436 - 457
• Main Authors: Guo, Qingwen, Wang, Chuntao, Xiao, Deqin, Huang, Qiong
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2023; Springer Nature B.V
• Subjects: Agriculture; Atmospheric Sciences; Biomedical and Life Sciences; Cameras; Chemistry and Earth Sciences; Computer Science; Computer vision; Datasets; Hypotheses; Insects; Life Sciences; Monitoring; Pest control; Pests; Physics; Population density; Precision farming; Remote Sensing/Photogrammetry; Sensors; Soil Science & Conservation; Statistics for Engineering; Traps; Vegetables; Insect pest monitoring; Yellow sticky traps; Computer-vision-based detector; Early pest management and control; Automatic
• ISSN: 1385-2256; 1573-1618
• DOI: 10.1007/s11119-022-09952-w

Pests cause heavy crop losses, so it is vital to conduct early pest management and control in precision agriculture. In general, pest monitoring is a foundation for early pest management and control. Conventional pest monitoring using manual sampling and detection is time consuming and labour intensive. Therefore, many studies have explored how to achieve automatic pest monitoring. However, few works have focused on automatic monitoring of flying vegetable insect pests. To close this gap, this study developed an automatic monitoring scheme for flying vegetable insect pests based on two hypotheses: (1) yellow sticky traps could provide reliable information to assess population density of flying vegetable insect pests, and (2) a computer-vision-based detector could accurately detect pests in images. Specifically, yellow sticky traps were exploited to sample flying vegetable insect pests, and an RGB camera was adopted to capture yellow-sticky-trap images; and a computer-vision-based detector called “YOLO for Small Insect Pests” (YOLO-SIP) was used to detect pests in captured images. The hypotheses were tested by using the Heuristics engineering method, installing yellow sticky traps and RGB cameras in vegetable fields, constructing a manually labelled image dataset, and applying YOLO-SIP to the constructed dataset with the mean average precision (mAP), average mean absolute error (aMAE), and average mean square error (aMSE) metrics. Experiments showed that the proposed scheme captured yellow-sticky-trap images automatically and obtained an mAP of 84.22%, an aMAE of 0.422, and an aMSE of 1.126. Thus, the proposed scheme is promising for the automatic monitoring of flying vegetable insect pests.