Please login first
Drone-based Smart Weed Localization from Limited Training Data and Radiometric Calibration Parameters
* 1 , * 2
1  School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Tehran, Iran
2  School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran
Academic Editor: Luca Lelli


Increased world population growth will demand more high-quality food production, which can only be achieved by applying a sustainable method for increasing crop yields. According to the Food and Agriculture Organization (FAO) report, weed grasses increase the environmental and economic costs of pesticide use by spreading them across farm boundaries, and their competition with agricultural crops reduces quantity and quality output. Among the pests, weed grasses are considered a crucial biotic constraint to food production. In traditional pest control methods in agriculture, most farm fields are spatially variable in grass weed infestation to a certain degree, but general weed management methods for herbicide application are based on the assumption that grass weeds are distributed uniformly in agricultural fields. However, a smart weed localization system for optimized herbicide dose in the agricultural field is a crucial step for smart farming and is still an open problem in pest control methods. While many object detection models appear to understand localization with a huge training data, a few-shot learning strategy potentially improves scene understanding with limited training data. In this study, the purpose of weed grasses localization from drone-based multispectral images is to locate the weed on large-scale images by using pixel-wise classification. Weed grasses localization with the use of uncertainty modeling in few-shot learning for drone-based multispectral images potentially improves multispectral scene understanding with a small training dataset, while many weed detection methods appear to understand single-time localization with a big training dataset. Few-shot learning can perform on unseen tasks after training a few annotated data and considers several tasks to produce a predictive function, and is an inductive transfer system whose main goal is to improve generalization ability for multiple tasks. Weed grasses localization of the trained model can be just blindly assumed accurate but the truth is not for decision making.

Keywords: Few-shot Learning, Weed Grasses Monitoring, Drone Imagery, Deep Learning, Limited Training Data.
Comments on this paper
Currently there are no comments available.