Proceedings
Authors
| Filter results6 paper(s) found. |
|---|
1. Evaluating Flight Path Strategy for Uav-based Phenotyping of Individual Muskmelon Plant in Greenhouse EnvironmentsUnmanned Aerial Vehicle (UAV)-based phenotyping is an emerging non-invasive method for high-throughput trait measurement in controlled environments. This study examines how UAV flight trajectory affects reconstruction fidelity and trait accuracy for muskmelon and grape plants in a GPS-denied greenhouse. Two strategies - circular loop and vertical hop - were flown using a UAV with RGB-D SLAM navigation, capturing data with a RunCam Thumb Pro. Data were processed through a GLOMAP structure-from-motion... T. Lin |
2. Integration of a Real-time Dairy Cow Eye Temperature Monitoring System Based on Deep Learning and Thermal ImagingEarly detection of heat stress and illness in dairy cows is critical for maintaining herd health and optimizing milk production. Among various physiological signals, body temperature is a key indicator of health status. In this study, we present a real-time, non-contact monitoring system that integrates dual-channel thermal imaging and deep learning for precise and automated surveillance. The system processes RGB and thermal video streams in parallel: in the RGB channel, YOLO detects faces, ByteTrack... T. Lin |
3. Plantsaga: Integrating Segment Anything Model with Gaussian Splatting for Plant Organ-level 3d SegmentationOrgan-level 3D phenotyping is essential for crop breeding but remains limited by the high cost of manual annotations. To address this challenge, PlantSAGA (Plant Segment Anything Gaussian Splatting) is introduced as a reference-based framework that enables accurate organ segmentation with minimal annotation. Multi-view muskmelon plants were reconstructed using COLMAP for camera pose estimation and Gaussian Splatting for 3D modeling, while 1~10 reference masks guided organ-level discrimination.... T. Lin |
4. Deep Learning-based Insect Detection on Sticky Traps Captured Via Mobile Phones Under Field Lighting ConditionsInsect pests pose a major threat to agricultural production, requiring effective integrated pest management (IPM) strategies that depend on accurate identification and counting of pests captured on sticky traps. However, mobile phone images taken under natural field lighting often suffer from inconsistent illumination, shadow interference, and low visibility of small insect targets, which significantly reduce the reliability of automated monitoring systems. To address these challenges, this study... T. Lin |
5. Modeling and Characterization of Unimodal and Bimodal Diurnal Pollen Foraging Patterns in Honeybee ColoniesPollen foraging patterns in honeybee colonies provide essential information on their ecological adaptation strategies. This study proposes a statistical modeling framework to characterize diurnal pollen foraging patterns in honeybee colonies. To support this, data were collected from healthy honeybee colonies during controlled experimental period. The raw pollen harvest data were then segmented into daily time series and converted into hourly histograms to capture foraging rhythms more effectively.... T. Lin |
6. Edge-AI-based Dairy Calf Behavior Monitoring System Using Computer Vision and Iot TechnologiesWe present an edge-AI, IoT system for real-time monitoring of dairy calf behavior that runs on embedded system and streams only compact results to the cloud. A lightweight, quantized MoViNet-A2 model deployed on a Raspberry Pi 4 classifies seven behaviors (non-active/active lying, non-active/active standing, feeding, drinking, ruminating) from 4-s clips captured once per minute, and publishes JSON outputs to AWS for dashboards. Field trials on three Holstein calves at the National Taiwan University... T. Lin |