YOLO-Poppy: Opium Poppy Detection Algorithm for Complex Aerial Scenes

The detection of poppies based on drone aerial images is more scientific and efficient compared to traditional methods such as field inspections, site visits, and informant reports. However, due to its unique imaging perspective and specific target background, the captured image data often contains considerable variations in target scale, low-quality target features caused by various concealed planting methods, and various deformations of targets caused by diverse perspectives and occlusions. To address these challenges, we propose a downsampling enhancement mechanism to mitigate the loss of low-quality targets by enhancing and reconstructing features, and integrate vegetation indices to better refine critical target features. Then for scenarios with heavy occlusion and interference in illegal cultivation, a joint measure loss function was created to prioritize critical feature extraction. Experiments on the Air-Poppy dataset with other models of the same magnitude and experiments on other plants datasets confirm the effectiveness and generalization of our algorithm.