As urbanization continues to expand year by year, the impact of urban pests on human life is increasing. To achieve precise control of urban pests and reduce the use of synthetic chemical pesticides, rapid and accurate species identification is the crucial first step. Cockroaches (Blattodea: Blattaria) are common public health pests in Taiwan, posing risks of various diseases. This study focuses on the common urban cockroach species in Taiwan, including the American cockroach (Periplaneta americana), German cockroach (Blattella germanica), and brown-banded cockroach (Supella longipalpa). It aims to obtain images of cockroaches from different angles, including adult males, adult females, and nymphs, and to identify their species, gender, and developmental stages using an artificial intelligence (AI) system. The study involved capturing wild and laboratory-reared cockroaches and placing them in transparent acrylic boxes measuring 21.5 × 13 × 30 cm. An Olympus Tough TG-6 camera was mounted above the boxes, and yellow, white, and blue sticky trap papers were used as backgrounds. Indoor lighting conditions were maintained at 500-600 lux, and additional flash was used for photography. The camera was connected to the OM image share app on a smartphone to control image capture and transfer the images to the cloud storage. Software was then used for image annotation and extraction of target images. The study has collected 9,000 images across 9 categories, each comprising 1,000 photographs. These include three types of cockroaches (Periplaneta americana, Blattella germanica, and Supella longipalpa), each in three distinct morphotypes (male, female, and nymph). After training with the YOLO (You Only Look Once) v8 model, the system can now accurately detect and identify cockroach species. Further tests included cross-validation, the addition of a new species (Blatta lateralis), identification of outdoor cockroach remnants, and assessments in various environments (sticky insect trap and under different light sources). These tests subdivided the species and their stages into ten distinct morphotype categories. The results show the recognition accuracy and offer an evaluation of its feasibility and potential future applications.

Artificial Intelligence, Urban Pests, Species Identification