Monitoring the tea plant pest population is critical for researchers and farmers. Pseudaulacaspis pentagona, a mulberry scale insect species, has been a severe pest for these plants. The short time window for spraying pesticides makes the species difficult to control. Therefore, monitoring is required to predict the optimal timing for control. The present study aimed at identifying the insect species and automatically counting the number of individuals. The dataset for the study included scanned images of sticky traps that captured the pest insects as well as their parasite wasps (Thomsonisca amathus). Two methods were used depending on the density of the captured insects. The object detection algorithm YOLOv4 was used for the RGB images. Accuracy was evaluated in terms of precision, recall, and F-measure, and for low-density insects their respective values were 0.89, 0.87, and 0.88 for P. pentagona, and 0.93, 0.93, and 0.93 for T. amathus. For high-density insects, the respective values were 0.68, 0.40, and 0.50 for P. pentagona, and 0.94, 0.90, and 0.92 for T. amathus. For high-density insects, a linear regression analysis was applied to the binarized images. RMSE, relative RMSE, bias, and relative bias were evaluated as 1.2 102, 1.1 10%, 5.1 10, and 4.7%, respectively. Combining the YOLOv4 and linear regression analysis successfully achieved insect species detection and population counting, which will increase monitoring capabilities for trap-captured pests and their natural enemies in tea gardens.
