Potato virus Y (PVY) [1], transmitted primarily by aphids, induces deleterious effects on potato plants. Conventional aphid counting involves deploying yellow sticky plates, necessitating manual counting which is a labour-intensive and time-consuming process. In our research, we have looked into alternative ways of counting aphids on the field, in the forms of object detection to detect aphids on sticky plates and tacking pictures of insects flying near the field using a camera setup and classifying these. We also looked into using Stable Diffusion to generate data to train on. The results have shown it is possible to count aphids automatically and taking pictures of insects near the field has proven itself as the better method, being able to get an accuracy around 85 percent consistently. Generating data using stable diffusion has also proven itself possible, being able to substitute as much as 75 percent of real images in a dataset, while maintaining similar results. During the research we found that the dataset used to train aphid detection models on has some imperfections such as incorrect annotations, which might have affected the results for the better or worse.
Link to the paper, first poster and second poster of this project.
