An initial model to automatically identify stored product Lepidoptera in next generation traps using artificial intelligence

Automated traps that regularly take images of captures are a next generation technology that can support integrated pest management programs at food facilities. The development algorithms based on artificial intelligence may assist pest control operators in long-term monitoring programs by distinguishing among species of stored product insects automatically. In food facilities, stored product Lepidoptera, including Plodia interpunctella (Indian meal moth), Ephestia kuehniella (Mediterranean flour moth), Cadra cautella (almond moth), Tineola bisselliella (clothes moth) closely resemble each other. This study used a collection of high-resolution photographs taken with a 3D imaging setup to train an algorithm to detect and identify these four different species of moth using Roboflow. Clean, frozen moths that were bred in the laboratory and an equal number of mated or unmated, females and males were used from each species. Images were taken of the dorsal, ventral, anterior, and posterior aspects of each individual and then input to train the model. We found the initial model was both precise and accurate when validated using a field data set of images collected from a pilot-scale flour mill. We were able to track the number of insects detected over time, providing pest control operators with information on phenological dynamics of stored product Lepidoptera. Moving forward, the model can be further refined to include other flying species, and will be of general use to facilities engaged in long-term monitoring of insects.