Detection of Bark-Beetle Infestation Using an Autonomous UAV
Detekce stromů nakažených kůrovcem pomocí autonomního bezpilotního letounu
Authors
Supervisors
Reviewers
Editors
Other contributors
Journal Title
Journal ISSN
Volume Title
Publisher
České vysoké učení technické v Praze
Czech Technical University in Prague
Czech Technical University in Prague
Date
Abstract
Tato práce prezentuje UAV systém, který detekuje stromy nakažené lýkožroutem smrkovým. Tento systém je navržen, aby identifikoval malé dírky, které vytvořil lýkožrout, když se zavrtal do kůry. Díky tomu jsou nakažené stromy detekovány velmi brzo po nákaze. Úloha je rozdělena na tři části: segmentace kmene stromu, detekce děr a finální klasifikace. Použití RGB kamery s vysokým rozlišením a LiDAR senzoru umožňuje zachycení detailních obrázků a přesné měření vzdálenosti. Segmentace kmene stromu používá sítě natrénované ResNet50, aby segmentovala pixely odpovídající kmeni v RGB obrázcích. Co se týče detekce děr, je použit algoristmus MSER (Maximally Stable Extremal Regions), který detekuje podezřelé skvrny. Spolu s ekvalizací histogramu a filtrováním založeném na kulatosti a intenzitě jsme schopni detekovat dírky způsobené lýkožroutem. Dodatečně, je natrénována síť YOLOv7 a porovnána s navrženým detektorem. Finální klasifikace používá "díry na plochu" metriku, poměr počtu detekovaných děr k viditelné ploše kůry. Strom je klasifikován pomocí této metriky a histogramů nakažených a zdravých stromů z trénovacích dat. Vyvinutý systém demonstruje svou efektivitu v brzké detekci a kontrole lýkožrouta smrkového a poskytuje lesním hospodářům cenný nástroj v boji proti lýkožroutu.
This thesis presents a UAV-based system for the detection of European spruce bark beetle-infested trees. The system is designed to identify small holes made by the bark beetle as it drills into the tree's bark and phloem, allowing for early detection of infestations. The pipeline consists of three stages: tree trunk segmentation, hole detection, and final classification. Integration of a high-resolution RGB camera and a LiDAR sensor enables detailed image capture and accurate distance measurements. The tree trunk segmentation stage employs a ResNet50 network trained to segment pixels corresponding to tree trunks in RGB images. For the hole detection, a Maximally Stable Extremal Regions (MSER) blob detection algorithm is applied, enhanced by histogram equalization and filtering based on circularity and intensity properties of the detected blobs. Additionally, a YOLOv7 model is trained to compare it with the proposed detector. The final classification utilizes the "holes per area" metric, the ratio of the number of detected holes to the visible bark area. A tree is classified using this metric based on histograms of healthy and infected trees in the training dataset. The developed system demonstrates its effectiveness in early detection and monitoring of European spruce bark beetle infestations, providing forest managers with a valuable tool for proactive forest health management and minimizing economic losses associated with bark beetle outbreaks.
This thesis presents a UAV-based system for the detection of European spruce bark beetle-infested trees. The system is designed to identify small holes made by the bark beetle as it drills into the tree's bark and phloem, allowing for early detection of infestations. The pipeline consists of three stages: tree trunk segmentation, hole detection, and final classification. Integration of a high-resolution RGB camera and a LiDAR sensor enables detailed image capture and accurate distance measurements. The tree trunk segmentation stage employs a ResNet50 network trained to segment pixels corresponding to tree trunks in RGB images. For the hole detection, a Maximally Stable Extremal Regions (MSER) blob detection algorithm is applied, enhanced by histogram equalization and filtering based on circularity and intensity properties of the detected blobs. Additionally, a YOLOv7 model is trained to compare it with the proposed detector. The final classification utilizes the "holes per area" metric, the ratio of the number of detected holes to the visible bark area. A tree is classified using this metric based on histograms of healthy and infected trees in the training dataset. The developed system demonstrates its effectiveness in early detection and monitoring of European spruce bark beetle infestations, providing forest managers with a valuable tool for proactive forest health management and minimizing economic losses associated with bark beetle outbreaks.