Power line corridor vegetation encroachment detection from satellite images using retinanet and support vector machine

• Main Author: Fathi Mahdi Elsiddig Haroun, Mr.
• Format: text::Thesis
• Language: English
• Published: 2023
• Subjects: Power line corridor vegetation encroachment detection from satellite images using retinanet and support vector machine

The overgrowth of vegetations beside a power line corridor right-of-way can cause flashovers when there is contact between tree branches and the transmission line, leading to a severe economic loss to power utility companies. There are many different methods used for detecting vegetation encroachment, such as patrol inspection, Light Detection And Ranging (LiDAR), Synthetic Aperture Radar (SAR), and airborne photography. However, these methods are costly concerning the coverage area. Highresolution satellite images can provide a wide geographical coverage at a relatively low cost. Several studies have provided solutions for detecting vegetation encroachment based on satellite images. Most of these detection solutions depend on the availability of multispectral satellite images and stereo satellite images. Satellite images that have only visible light bands can be a low-cost alternative compared with other types of satellite images. Besides this, many satellite imagery platforms provide free accessibility of such data. In this dissertation, a new vegetation encroachment detection method was proposed by studying the feasibility of using the visible-light band of highresolution satellite images using the RetinaNet deep learning model and Support Vector Machine algorithm (SVM). The proposed vegetation encroachment detection framework used the state-of-art RetinaNet model to identify and locate the power transmission towers from satellite images. A routing algorithm has been developed to create a routing path between all transmission towers in the image, which helps identify the power line path. Also, a corridor extraction algorithm has been developed to extract the region of interest (ROI) around the transmission towers. The SVM algorithm has been used to detect high- and low-density vegetation regions from the extracted ROI. This proposed method can help utility companies make vegetation management decisions at a lower cost. The mean average precision (mAP) of the transmission tower detection accuracy was 85.21% for an intersection over union (IoU) threshold ≥ 0.3 and 0.7245 for IoU≥0.5. Also, the achieved classification recall was 98.272% using the SVM Radial Basis Function (RBF) kernel. The results emphasize the possibility of using satellite images with only a visible-light band for detecting the vegetation encroachment along the power line corridor.