1. Aerial drones are providing a new source of high‐resolution imagery for mapping of plant species of interest, amongst other applications. On‐board detection algorithms could open the door to allow for applications in which drones can intelligently interact with their environment. However, the majority of plant detection studies have focused on detection in post‐flight processed orthomosaics. Greater research into developing detection algorithms robust to real‐world variations in environmental conditions is necessary, such that they are suitable for deployment in the field under variable conditions.
2. We outline the steps necessary to develop such a system, show by example how real‐world considerations can be addressed during model training and briefly illustrate the performance of our best performing model in the field when integrated with an aerial drone.
3. Our results show that introducing variations in brightness as an additional augmentation strategy during training is beneficial when dealing with real‐life data. We achieved a 27% improvement in the F1‐score obtained on the unseen test set when using this approach. Further improvements to the model performance were obtained through the use of weight map‐based loss, accounting for uncertainty in the annotation masks due to the indistinct nature of the edges of the target plants using weighting. This resulted in a 15% improvement in precision for the best configuration of hyper‐parameters, yielding a final model with an F1‐score of 83% and accuracy of 96%. Finally, results computed on the fly show that such a system is deployable in the field.
4. This study shows that it is possible for a commercially available drone, integrated with a deep learning model, to detect invasive plants in the field and demonstrates methodology which could be applied to developing similar systems for other plant species of interest. The ability to perform detection on the fly is necessary for future applications in which intelligent interaction between a drone and its environment is required.

中文翻译：

---

使用深度学习和无人机技术在野外检测植物物种

1. 除其他应用外，空中无人驾驶机提供了一种新的高分辨率图像来源，用于标绘感兴趣的植物物种。机载检测算法可以为无人机可以与其环境进行智能交互的应用打开大门。但是，大多数植物检测研究都集中在飞行后加工的正马赛克中进行检测。有必要对开发对环境条件的真实变化具有鲁棒性的检测算法进行更深入的研究，以使其适合在可变条件下现场部署。
2. 我们概述了开发这种系统的必要步骤，以示例方式展示了如何在模型训练期间解决现实世界中的注意事项，并简要说明了与空中无人驾驶飞机集成时我们在现场性能最佳的模型的性能。
3. 我们的结果表明，在训练过程中引入亮度变化作为额外的增强策略在处理真实数据时是有益的。使用这种方法时，在看不见的测试集上获得的F1分数提高了27％。通过使用基于权重图的损失，模型性​​能得到了进一步的改善，这说明了由于使用权重的目标植物边缘的模糊特性而导致的注释蒙版的不确定性。对于最佳的超参数配置，这使精度提高了15％，最终模型的F1得分为83％，准确度为96％。最后，即时计算的结果表明这种系统可以在现场部署。
4. 这项研究表明，与深度学习模型集成的商用无人机有可能在野外检测到入侵植物，并演示了可用于开发其他目标植物类似系统的方法。对于需要无人机与周围环境进行智能交互的未来应用，必须具备进行动态检测的能力。