Accurate root feature extraction

Ground-Penetrating Radar (GPR) has been used as a non-destructive and non-invasive tool for tree root inspection. Data collection of parameters useful in the imaging and monitoring of tree roots has been greatly facilitated by GPR radargrams. However, the task of extracting useful root-related information is challenging, as the data recorded by these scans are often affected by various factors that can affect the accuracy and reliability of the results. These include but are not limited to soil conditions; where the saturation of the soil can significantly reduce signal penetration depth and yield scan data with poor resolutions, the heterogeneity of the soil, including soil texture and moisture content, which can introduce clutter in GPR datagrams, and the variability of root architecture; where complex branching patterns and irregular shapes of tree roots can lead to uncertainties in the interpretation of GPR data [1]. This paper presents a comprehensive study on feature extraction techniques that aid in obtaining useful root-related information from noisy and cluttered GPR radargrams. The proposed method employs the application of machine and deep learning techniques and a deep convolutional neural network to perform feature extraction and data processing. The effectiveness of the proposed method is demonstrated through experiments conducted on real GPR data.Bachelor of Engineering (Information Engineering and Media