논문 상세보기
Springtails (class Collembola) play a crucial role in soil ecosystems. They are commonly used as standard species in soil toxicity assessments. According to the ISO 11267 guidelines established by the International Organization for Standardization (ISO), Allonychiurus kimi uses adult survival and juvenile production as toxicity assessment endpoint. Conventional toxicity assessment methods require manually counting adults and larvae under a microscope after experiments, which is time-consuming and laborintensive. To overcome these limitations, this study developed a model using YOLOv8 to detect and count both adults and juveniles of A. kimi. An AI model was trained using a training dataset and evaluated using a validation dataset. Both training and validation datasets used for AI model were created by picturing plate images that included adults and larvae. Statistical comparison of validation dataset showed no significant difference between manual and automatic counts. Additionally, the model achieved high accuracies (Precision=1.0, Recall=0.95 for adults; Precision=0.95, Recall=0.83 for juveniles). This indicates that the model can successfully detect objects. Additionally, the system can automatically measure body areas of individuals, enabling more detailed assessments related to growth and development. Therefore, this study establishes that AI-based counting methods in toxicity assessments with offer high levels of accuracy and efficiency can effectively replace traditional manual counting methods. This method significantly enhances the efficiency of large-scale toxicity evaluations while reducing researcher workload.
1. 서 론
2. 재료 및 방법
2.1. 실험 생물
2.2. 인공지능 모델 학습 및 검증용 데이터셋 구축
2.3. 개체 탐지 인공지능 모델
2.4. 통계분석 및 모델 성능평가지표
3. 결 과
3.1. 성충/유충 계수
3.2. 개체 크기 측정
3.3. 성충/유충 탐지 성능
4. 고 찰
적 요
CRediT authorship contribution statement
Declaration of Competing Interest
사 사
REFERENCES
APPENDIX
