2019 · 23 citations · 10 references
ComputingAutonomous SystemsComputer ScienceComputer VisionMachine LearningUnmanned Aircraft SystemsDeep LearningSurveillance ApplicationsWater Resource MonitoringConvolutional Neural NetworksDeep Learning For InspectionDeep Learning-based DetectionDeep Learning-based MonitoringDocking Station DemandManual Visual InspectionReal-time MonitoringUnmanned Aerial Vehicle-based Application
Floating trash in the river presents a major environmental problem that endangers the lives of river inhabitants. However, conventional trash monitoring relies on labor-intensive manual inspection by dispatching inspectors to the filed, which is highly cost-ineffective. Despite that Unmanned Aerial Vehicles (UAVs) have been widely proposed for many real-time monitoring applications, limited power supply remains the most challenging bottleneck for such UAV-assisted monitoring applications. In this work, we propose an automated river trash monitoring system called SuperDock. SuperDock consists of a remote processing unit, a docking station and a UAV. SuperDock lands the UAV precisely onto the docking station followed by performing automated battery replacement. As a result, the UAV can continue the monitoring task after the battery replacement. Furthermore, SuperDock includes a deep learning-assisted river trash detection module based on YOLOv3 that runs much faster than the conventional Convolution Neural Networks (CNN). In addition, a data set has been generated for training and testing the deep learning network, specifically for the floating trash detection application. Finally, SuperDock enables the UAV to communicate wirelessly with a remote computer in a real-time manner. Experimental and simulation results show that SuperDock is highly effective in monitoring the floating trash in the river.
10
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Ross Girshick, Jeff Donahue, Trevor Darrell et al. · 2014 · 31.8K citations
YOLO9000: Better, Faster, Stronger
Joseph Redmon, Ali Farhadi · 2017 · 19K citations
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Shaoqing Ren, Kaiming He, Ross Girshick et al. · arXiv (Cornell University) · 2015 · 18.2K citations · Full text