Remote Sensing for Weed Detection and Control

Ishita Bansal,Peder Olsen,Roberto Estevão

2024-10-30

Abstract:Italian ryegrass is a grass weed commonly found in winter wheat fields that are competitive with winter wheat for moisture and nutrients. Ryegrass can cause substantial reductions in yield and grain quality if not properly controlled with the use of herbicides. To control the cost and environmental impact we detect weeds in drone and satellite imagery. Satellite imagery is too coarse to be used for precision spraying, but can aid in planning drone flights and treatments. Drone images on the other hand have sufficiently good resolution for precision spraying. However, ryegrass is hard to distinguish from the crop and annotation requires expert knowledge. We used the Python segmentation models library to test more than 600 different neural network architectures for weed segmentation in drone images and we map accuracy versus the cost of the model prediction for these. Our best system applies herbicides to over 99% of the weeds while only spraying an area 30% larger than the annotated weed area. These models yield large savings if the weed covers a small part of the field.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper attempts to address the problem of efficiently and accurately detecting and controlling weeds in farmland, particularly Italian ryegrass. This weed competes with crops for water and nutrients in winter wheat fields, and if not controlled, can lead to significant reductions in yield and grain quality. To reduce the cost of herbicides and minimize environmental impact, researchers utilize drone and satellite imagery for weed detection.

Specifically, the paper addresses the following issues:

1. **High-resolution weed detection**: Achieving high-resolution weed detection through drone imagery to enable precise spraying. The resolution of drone images is high enough for precise spraying but requires expert knowledge for annotation.

2. **Application of low-resolution satellite imagery**: Although the resolution of satellite images is lower and cannot be directly used for precise spraying, they can assist in planning drone flights and treatment schemes. Satellite images can be used to estimate the weed coverage rate on each plot, helping farmers prioritize plots with higher weed density.

3. **Model optimization**: Various neural network architectures and loss functions were studied to find the optimal weed segmentation model. These models can accurately detect weeds while minimizing the spraying area to cover most weeds, thereby reducing costs and environmental impact.

4. **Ensemble models**: Improving prediction accuracy and robustness by integrating multiple models. For example, using a Voting Regressor to combine multiple regression models improved the prediction accuracy of weed coverage rates in satellite images.

5. **Performance evaluation in practical applications**: Experiments were conducted to validate the performance of different models in practical applications, including metrics such as over-spraying rate, model size, and inference speed, to ensure the feasibility and efficiency of the models in real-world applications.

In summary, the paper aims to enhance the efficiency and sustainability of farmland management through advanced image processing techniques and machine learning methods, reducing the use of chemical pesticides and protecting the environment.