-Visual inspection time of 1 hour or more can be reduced to 3-4 minutes-
The National Agricultural Food Research Organization (NARO) has developed an AI system that can automatically count rice planthoppers from pictorial images of survey boards. By automatically recognizing and counting rice planthopper individuals with an accuracy of more than 90%, the inspection time, which may take more than an hour per survey board by visual inspection, can be reduced to 3-4 minutes. This AI system will be useful for the accurate prediction of crop damage due to pests and their control.
The rice planthoppers (brown planthopper, white-backed planthopper, small brown planthopper) are designated as "harmful animals" under the Plant Protection Act for stable rice production in Japan. Therefore, the outbreak status of the rice planthoppers in paddy fields is monitored regularly, and if crop damage due to them is predicted, warnings and advisories are issued to producers. In the above-mentioned pest forecasting project carried out by the national government, the prefectural pest control stations conduct regular surveys of rice planthoppers in paddy fields at least twice a month, at about 3,000 sites nationwide.
To investigate the outbreak of crop pests, skilled experts who can identify the pests are needed. The rice planthoppers are up to about 5 mm in size, and pest experts visually count planthopper individuals on the survey board one by one for surveys of rice planthoppers. The NARO has successfully trained an AI to automatically recognize only rice planthoppers by selecting them from other insects and plant pieces on the survey board. This specialized AI can distinguish three types of rice planthopper individuals into 18 categories (species, developmental stages, and sexes and wing-forms in adults) with an accuracy of about 90%, and notably, it can distinguish individuals of brown planthopper, which causes the most serious damage of the rice crop, into categories with an accuracy of more than 95%.
By the existing visual inspection method, counting rice planthopper individuals may take more than an hour per survey board. However, by using the developed AI system, the same counting task can be completed within 3-4 minutes, including the process of imaging the survey board. Thus, this AI system can greatly reduce labor costs and speed up the surveys of rice planthoppers with uniform accuracy, which will be useful for the accurate prediction of crop damage due to pests and their control. As a next step, demonstration tests will be conducted for spreading this system throughout Japan.
Takayama et al. (2021) Evaluation of detection accuracy of image recognition for automatic counting of rice planthoppers captured on sticky boards. Agricultural Information Research, 30(4): 174-184.
Yashiro & Sanada-Morimura (2021) A rapid multiplex PCR assay for species identification of Asian rice planthoppers (Hemiptera: Delphacidae) and its application to early-instar nymphs in paddy fields. PLoS ONE, 16(4): e0250471.