A Stable Diffusion Based Image Generation Method for Few-Shot Emerging Pest Detection in the Wild

Detecting few-shot emerging pests in the wild presents a significant challenge in the field of agricultural computer vision. However, there has been minimal research focused on generating emerging pest images with different natural scenarios that align with the authentic pest living environments. With the advent of the controllable generative stable diffusion model recent years, we proposed an innovative image generation method based on stable diffusion techniques as well as introducing text prompt and coordinate information into the vanilla model. Both qualitative and quantitative experimental results demonstrate that our proposed approach can generate anthropomorphic images which are both visually recognized by human experts and various image quality assessment performance metrics.