Research on an Improved YOLO Model-Based Algorithm for UAV Poppy Recognition

Hu Li; Yushuang Feng; Jiajun Su; Bo Yuan

doi:10.54097/x9yjy202

Authors

Hu Li
Yushuang Feng
Jiajun Su
Bo Yuan

DOI:

https://doi.org/10.54097/x9yjy202

Keywords:

Poppy detection; YOLOv9; Loss function; Small target detection; Recognition accuracy

Abstract

To address the challenges faced in drone-based poppy detection tasks, for instance small size, confusing backgrounds, and high recognition accuracy requirements, this paper come up with a poppy small target recognition method based on enhanced YOLOv9. Initially, the CAM unit was added to the backbone network, which enhances the ability of model to express fine-grained features of poppies by integrating Channel Attention and spatial Mixing mechanisms. Secondly, a content-guided attention model is incorporated into the neck of the feature pyramid, leveraging deep semantic information to adaptively weight features of different scales. This enhancement leads to an improvement in multifaceted feature integration and a substantial enhancement in the responsiveness of the detection system to diminutive objects. Finally, the issue of existing object detection loss functions not being sufficiently sensitive to bounding box regression is addressed by the introduction of the Inner IoU loss function. Finally, the issue of existing object detection loss functions not being sufficiently sensitive to bounding box regression is addressed by the creation of the Inner IoU loss as a solution. The latter focuses more on maximizing the overlap between the actual box and the predicted box, which can further improve the accuracy of position regression. Experiments conducted on a dataset of aerial photographs of poppies demonstrate that the designed algorithm enhances mAP by 13.5% and detection accuracy by 16.8% in comparison with the original YOLOv9 model. While ensuring real-time processing, it delivers higher recognition accuracy and stronger robustness, providing an effective solution for rapid and accurate poppy identification in drug control operations. Ensuring real-time processing, it delivers higher recognition accuracy and stronger robustness, offering an adequate means by which to swiftly and accurately identify poppies in drug control operations.

References

[1]Zhou J, Tian Y, Yuan C, et al. Improved UAV opium poppy detection using an updated YOLOv3 model[J]. Sensors, 2019, 19(22): 4851.

[2]Radovic M, Adarkwa O, Wang Q. Object recognition in aerial images using convolutional neural networks[J]. Journal of Imaging, 2017, 3(2): 21.

[3]Zhang Z, Xia W, Xie G, et al. Fast opium poppy detection in Unmanned Aerial Vehicle (UAV) imagery based on deep neural network[J]. Drones, 2023, 7(9): 559.

[4]Liu X, Tian Y, Yuan C, et al. Opium poppy detection using deep learning[J]. Remote Sensing, 2018, 10(12): 1886.

[5]Rominger K R, Meyer S E. Drones, deep learning, and endangered plants: A method for population-level census using image analysis[J]. Drones, 2021, 5(4): 126.

[6]Wang Q, Wang C, Wu H, et al. A two-stage low-altitude remote sensing papaver somniferum image detection system based on YOLOv5s+ DenseNet121[J]. Remote Sensing, 2022, 14(8): 1834.

[7]Rominger K, Meyer S E. Application of UAV-based methodology for census of an endangered plant species in a fragile habitat[J]. Remote Sensing, 2019, 11(6): 719.

[8]Zou C, Jeon W S, Rhee S Y. Research on the multiple small target detection methodology in remote sensing[J]. Sensors, 2024, 24(10): 3211.

[9]Qin W, Yang X, Wang Y, et al. YOLOPoul: Performance evaluation of a novel YOLO object detectors benchmark for multi-class manure identification to warn about poultry digestive diseases[J]. Smart Agricultural Technology, 2025: 101145.

[10]Ye R, Gao Q, Qian Y, et al. Improved yolov8 and sahi model for the collaborative detection of small targets at the micro scale: A case study of pest detection in tea[J]. Agronomy, 2024, 14(5): 1034.

[11]Song X, Yan L, Liu S, et al. Agricultural Image Processing: Challenges, Advances, and Future Trends[J]. Applied Sciences, 2025.

[12]Sun Z, Yang H, Zhang Z, et al. An improved YOLOv5-based tapping trajectory detection method for natural rubber trees[J]. Agriculture, 2022, 12(9): 1309.

[13]Nedungatt S, KS B, Lal S. MDEANet: modified detail-enhanced convolution and attention-based network for dehazing of remote sensing images[J]. Multimedia Tools and Applications, 2025, 84(18): 18943-18966.

[14]Yuan M, Meng H, Yan T, et al. Fa YOLO: fog-aware instance segmentation of ships with feature refinement in foggy scenes: M. Yuan et al[J]. Multimedia Systems, 2025, 31(4): 291.

[15]Chen Z, He Z, Lu Z M. DEA-Net: Single image dehazing based on detail-enhanced convolution and content-guided attention[J]. IEEE transactions on image processing, 2024, 33: 1002-1015.

[16]Yekeben Y, Cheng S, Du A. CGFTNet: Content-Guided Frequency Domain Transform Network for Face Super-Resolution[J]. Information, 2024, 15(12): 765.

[17]Sun Z, Yang H, Zhang Z, et al. An improved YOLOv5-based tapping trajectory detection method for natural rubber trees[J]. Agriculture, 2022, 12(9): 1309.

[18]Yu Z, Liu Y, Yu S, et al. Automatic detection method of dairy cow feeding behavior based on YOLO improved model and edge computing[J]. Sensors, 2022, 22(9): 3271.

[19]Dong L, Zhu H, Ren H, et al. Developing YOLOv5s model with enhancement mechanisms for precision parts with irregular shapes[J]. Advanced Engineering Informatics, 2025, 65: 103257.

[20]Xue X W, Dai Q, Zhao M. SiamMask RAM for Underwater Target Tracking in Sonar Images[C]Journal of Physics: Conference Series. IOP Publishing, 2024, 2891(3): 032031.

[21]Yin Y, Zhang Z, Wei L, et al. Pedestrian detection algorithm integrating large kernel attention and YOLOV5 lightweight model[J]. PLoS one, 2023, 18(11): e0294865.

[22]Chen B, Ding F, Ma B, et al. A method for real-time recognition of safflower filaments in unstructured environments using the YOLO-SaFi model[J]. Sensors, 2024, 24(13): 4410.