Use of unmanned aerial vehicle for environmental monitoring purposes and precision agriculture

Ahmed, Aishath Shara (2023) Use of unmanned aerial vehicle for environmental monitoring purposes and precision agriculture. Doctoral thesis, Northumbria University.

Preview

Text (Doctoral thesis) ahmed.aishath shara_phd(16038732).pdf - Submitted Version Download (13MB) | Preview

Abstract

Despite numerous research studies on remote sensing applications in forestry and precision agriculture, there is a limited availability of image analysis techniques that are less complex, reproducible, and applicable to diverse locations and under a wide range of environmental conditions. Image analysis techniques currently in use employ complex machine learning approaches (regression-based models), for example, to identify tree species in forestry and estimate crop yield in precision agriculture. However, many challenges must be overcome before these modern machine learning approaches can potentially see widespread commercial and non-commercial implementation in agriculture and forestry. As a result, there is a need to investigate and develop simple, dependable, and reproducible image analysis methods by utilising remote sensing data applicable in forestry and precision agriculture.

Hence, the current study focuses on using a remote sensing platform of multispectral unmanned aerial vehicle (UAV) to monitor native and invasive tree species in an ancient semi-natural woodland and investigate the performance of a variety of crops for precision agriculture, including oilseed rape, winter beans, and winter oats. The multispectral UAV data were analysed using simple yet effective image analysis techniques such as principal component analysis (PCA), spectral vegetation indices combined with image classification methods of thresholding and clustering (k-means and iso-cluster). Also, the image analysis methods were performed with effective data manipulation software such as MATLAB and ArcGIS.

Identification and quantification of native and invasive tree species was achieved by PCA derived spectral vegetation indices, thresholding and k-means clustering. Additionally, the use of spectral vegetation indices and iso-cluster classification method in precision agriculture of crops assisted in estimation of crop yield three months before harvest. Also, strong correlation was observed between the estimate and actual crop yield. Furthermore, a pilot study using a multinomial logistic regression model with high sensitivity and accuracy enabled the identification of soil nutrient concentration and crop quality features for very high oats yield.

The simple and effective image analysis methods on multispectral UAV data for forestry and precision agriculture must be employed more frequently than complex machine learning approaches. Also, the estimated crop yield prior to harvesting aids farmers for precision agriculture of crops to maintain its performance. Whereas, in forestry these methods can be employed frequently to monitor the native tree species and emergence of new invasive tree species and remove them effectively to maintain a sustainable ecosystem.

Actions (login required)

View Item

Downloads