Multispectral Remote Sensing for Weed Detection in West Australian Agricultural Lands

TL;DR

The paper tackles pervasive weed infestations in the Kondinin region of Western Australia by building a WA-specific multispectral UAV dataset and an end-to-end deep learning framework for weed detection. It introduces a four-year data collection campaign using the DJI Matrice 300 RTK and P4 Multispectral, covering 0.6046 km^2 and yielding 12,627 raw images with ground-truth labels. Five DL architectures are evaluated using vegetation indices NDVI, GNDVI, EVI, SAVI, and MSAVI, with ResNet-50 achieving the best performance (accuracy 0.9213, F1 0.8735, mIOU 0.7888, mDC 0.8865). The dataset and pipeline provide a practical baseline for scalable weed management in WA, advancing precision agriculture in regional farming systems and informing future localization and deployment efforts.

Abstract

The Kondinin region in Western Australia faces significant agricultural challenges due to pervasive weed infestations, causing economic losses and ecological impacts. This study constructs a tailored multispectral remote sensing dataset and an end-to-end framework for weed detection to advance precision agriculture practices. Unmanned aerial vehicles were used to collect raw multispectral data from two experimental areas (E2 and E8) over four years, covering 0.6046 km^{2} and ground truth annotations were created with GPS-enabled vehicles to manually label weeds and crops. The dataset is specifically designed for agricultural applications in Western Australia. We propose an end-to-end framework for weed detection that includes extensive preprocessing steps, such as denoising, radiometric calibration, image alignment, orthorectification, and stitching. The proposed method combines vegetation indices (NDVI, GNDVI, EVI, SAVI, MSAVI) with multispectral channels to form classification features, and employs several deep learning models to identify weeds based on the input features. Among these models, ResNet achieves the highest performance, with a weed detection accuracy of 0.9213, an F1-Score of 0.8735, an mIOU of 0.7888, and an mDC of 0.8865, validating the efficacy of the dataset and the proposed weed detection method.

Figures (9)

• Figure 1: Workflow of our proposed end-to-end weed detection framework with a ResNet-50 model: The workflow starts with data retrieval, preprocessing, feature selection, and image slicing. The sliced sub-images are divided into three groups: training set, validation set, and test set. The training set is augmented and used to train a ResNet-50 model. The validation set is used to determine the optimal hyperparameters. These optimal hyperparameters are applied to train the optimized model, which is then tested and analyzed with the test set.
• Figure 2: SIFT feature detection for image alignment. Keypoints are shown on green and the correlations are shown in blue.
• Figure 3: Image stitching from raw data (RGB) to a map.
• Figure 4: 5 selected features are calculated respectively using all 5 stitched multispectral images
• Figure 5: Example (Left)Flight Path Planning for E2 experimental area(Right) Flight Path Planning for E8 experimental area with GSPro