DURIAN_Outline_Defense_SplitBackground_v2.pptx

It is a defense presentation about durian maturity

1. D.U.R.I.A.N.: UAV-Assisted Detection and Maturity Assessment for Puyat Durian using YOLOv8 Outline Defense Researcher: amanita • Calinan, Davao

2. Outline of the Presentation Title slide Outline of the Presentation Background of the Study — Context Background of the Study — Problem Background of the Study — Objective Problem and Objectives (Summary) Significance of the Study (aligned with the University Research Agenda) Scope and Limitation of the Study Methodology (Materials and Methods) Barriers to Success Timeline References

3. Background of the Study — Context Puyat durian orchards in Davao require consistent maturity and quality for local and export markets. Manual maturity/defect checks are subjective, slow, and risky to scale. UAVs provide safe canopy coverage; edge AI (Jetson Nano + YOLOv8) enables on-site inference. Need geotagged, per‑tree information to support harvest planning and traceability. Reference: D.U.R.I.A.N. Chapters 1–2 manuscript (2025), internal draft.

4. Background of the Study — Problem No fast, objective maturity/defect assessment specific to Puyat durian at orchard scale. Inconsistent manual inspection and lack of tree‑level traceability. Hard to map results to GPS/tree IDs for quality control. Double counting across overlapping frames/images inflates yield estimates. Requires de‑duplication to ensure accurate per‑tree counts. Field constraints (lighting, occlusion, battery/wind) reduce reliability of manual methods.

5. Background of the Study — Objective Develop an end‑to‑end UAV + edge‑AI system that: Detects durian fruits and classifies maturity/defect from images (YOLOv8). Geotags detections and associates them with unique tree IDs. Aggregates counts per tree to produce a Maturity Map and Per‑Tree Yield Table.

6. Problem and Objectives (Summary) Problem: Absence of rapid, objective, geotagged maturity/defect assessment for Puyat durian. Objective 1: Train YOLOv8 to classify ripe, unripe, and defect fruits from UAV images. Objective 2: Implement geotagging and per‑tree aggregation with de‑duplication. Objective 3: Evaluate accuracy (mAP, Precision/Recall/F1) and operational performance.

7. Significance of the Study (University Research Agenda) Advances smart agriculture and digital transformation in the region. Improves productivity, quality control, and export readiness for durian growers. Reusable edge‑AI blueprint for other crops and orchard traits.

8. Scope and Limitation of the Study Scope: Puyat cultivar; outdoor orchards in Calinan, Davao; UAV RGB still imagery. Counting rule: fruit visibility ≥ 75% to qualify detections. Limitations: occlusion, lighting variability, generalization beyond Puyat; battery/wind constraints.

9. Methodology (Materials and Methods) Design: Modified Waterfall—Planning → Design/Implementation → Testing → Evaluation. Acquisition: RC2 → USB‑C file transfer to Jetson Nano (near‑real‑time or post‑flight). Pre‑processing: Quality/visibility gate; optional contrast/denoise. Model: YOLOv8 (n/s) optimized on Jetson (FP16/INT8). Geotagging: Timestamp sync with telemetry; per‑tree assignment and aggregation. Evaluation: mAP, confusion matrices, Accuracy/Precision/Recall/F1; vs manual inspection.

10. Barriers to Success Image quality (motion blur, glare, heavy occlusion). Mitigation: fast shutter, exposure lock, planned overlap. Telemetry misalignment affecting tree assignment. Mitigation: clock sync; buffer radius; post‑flight verification. Class imbalance (rare defects) during training. Mitigation: targeted sampling, augmentation, focal loss. Edge compute/storage limits during batch processing. Mitigation: FP16/INT8, batch tuning, periodic offload.

11. Timeline M1–M2: Permits, equipment setup, protocol finalization. M3–M4: Data collection flights and labeling. M5: Model training/ablations; geotagging & aggregation module. M6: Field validation; accuracy & performance evaluation. M7: GUI polish; documentation; final defense.

12. References D.U.R.I.A.N. Chapters 1–2 manuscript (2025). Ultralytics YOLOv8 documentation and related literature on small‑object detection. UAV‑based orchard mapping and phenotyping studies (selected). NVIDIA Jetson Nano edge‑AI deployment guides.