北海道におけるニホンジカの自動個体識別と検出：農業保護のためのモジュール軽量化
Year: 2025
Type: dissertation
Abstract: This study explores the use of deep learning and camera trap technologies to enhance wildlife monitoring and conservation efforts, with a focus on Sika deer in Hokkaido, Japan. It addresses the challenges of manual wildlife identification and proposes a fully automated system to classify and identify species using deep learning models. The study evaluates the performance of two Convolutional Neural Networks (CNNs), EfficientNetB0 and VGG16, for multi-species classification, finding that EfficientNetB0 outperformed VGG16 in terms of accuracy, precision, and recall. The model demonstrated over 80% accuracy and achieved a top-5 accuracy of over 90%, making it highly effective for wildlife monitoring applications. Additionally, the research assesses the performance of three camera traps, highlighting the Solar-Powered 4k-Trail camera as the most effective in detecting Sika deer. The study emphasizes the importance of camera calibration and distortion correction for optimizing camera trap performance in ecological research. A key part of the study also focused on the identification of individual Sika deer using a Siamese Network Architecture with triplet loss functions. The ResNet152 model was the most successful, achieving high accuracy in classifying individual Sika deer based on distinctive features like spot patterns. The study also investigates human-wildlife conflict, particularly the agricultural damage caused by Sika deer and bears in Hokkaido. To mitigate this, a Sika deer detection system was developed using the YOLOv8-nano model deployed on a Raspberry Pi, integrated with a PIR sensor. The system provided real-time alerts to local residents via email and the LINE application, alongside audio notifications. While the buzzer's effectiveness in deterring deer was mixed, the alert system proved successful in monitoring deer activities, offering valuable insights for conservation strategies. Further research is recommended to explore alternative deterrents, such as drones, for improving wildlife management in agricultural areas.
Authors: SANDHYA, SHARMA, サンディヤ, シャルマ
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
SDG: Life on land
Open Access status: green
OpenAlex ID: W7112187681
Cites:
Cited by:
Related to: