Detection of the colorado potato beetle (Leptinotarsa decemlineata) (say) on the potato plant using object detection algorithm (the YOLOv4 algorithm ) and population monitoring opportunities

In this thesis, the detection of adult and late (3rd and 4th instars) larvae of potato beetle (Leptinotarsa decemliniata) on potato plant was investigated with the YOLOv4 object detection algorithm. At the end of the training, the evaluation metric of each 1000th iteration were examined and the most suitable model was tried to be determined by changing the input image resolution of the network to "416x416", "608x608", "832x832", "1024x1024" and "1440x1440" in the "cfg configuration" file. As a result of the training, the most suitable object detector model was obtained in the 3000th iteration by changing the input image resolution to "1024x1024" in the "cfg configuration" file. Recall 0.78, precision 0.85, F1-score 0.81, mAP 87.53 and IoU 57.99 were obtained as the evaluation metric of the suggested model. In addition, determining the emergence time of overwintered edult potato beetle by using yellow sticky traps, pitfall traps and visual inspection method is discussed. The trapping of the overwintered potato beetle with yellow sticky traps and pitfall traps was examined, and theirs usability in early warning systems was investigated by using the YOLOv4 object detection algorithm. Adult potato beetles were detected in pitfall traps for the first time on 25.05.2021. They were seen by using the visual control method on 01.06.2021. By using the pitfall trap method, the emergence of the potato beetle was earlier compared to the visual control method. In this study, no potato beetle was caught in the yellow sticky traps.